about summary refs log tree commit diff
path: root/tvix/nix-compat
diff options
context:
space:
mode:
Diffstat (limited to 'tvix/nix-compat')
-rw-r--r--tvix/nix-compat/Cargo.toml47
-rw-r--r--tvix/nix-compat/benches/derivation_parse_aterm.rs31
-rw-r--r--tvix/nix-compat/benches/narinfo_parse.rs69
-rw-r--r--tvix/nix-compat/default.nix7
-rw-r--r--tvix/nix-compat/src/aterm/escape.rs27
-rw-r--r--tvix/nix-compat/src/aterm/mod.rs7
-rw-r--r--tvix/nix-compat/src/aterm/parser.rs118
-rw-r--r--tvix/nix-compat/src/bin/drvfmt.rs42
-rw-r--r--tvix/nix-compat/src/derivation/errors.rs58
-rw-r--r--tvix/nix-compat/src/derivation/mod.rs289
-rw-r--r--tvix/nix-compat/src/derivation/output.rs146
-rw-r--r--tvix/nix-compat/src/derivation/parse_error.rs81
-rw-r--r--tvix/nix-compat/src/derivation/parser.rs510
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/duplicate.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv.json23
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv.json19
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv.json23
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv.json19
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv.json16
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv.json23
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv.json23
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv.json21
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv.json23
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv1
-rw-r--r--tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv.json21
-rw-r--r--tvix/nix-compat/src/derivation/tests/mod.rs416
-rw-r--r--tvix/nix-compat/src/derivation/validate.rs158
-rw-r--r--tvix/nix-compat/src/derivation/write.rs196
-rw-r--r--tvix/nix-compat/src/lib.rs7
-rw-r--r--tvix/nix-compat/src/nar/mod.rs4
-rw-r--r--tvix/nix-compat/src/nar/reader/mod.rs293
-rw-r--r--tvix/nix-compat/src/nar/reader/read.rs109
-rw-r--r--tvix/nix-compat/src/nar/reader/test.rs120
-rw-r--r--tvix/nix-compat/src/nar/tests/complicated.narbin0 -> 840 bytes
-rw-r--r--tvix/nix-compat/src/nar/tests/helloworld.narbin0 -> 128 bytes
-rw-r--r--tvix/nix-compat/src/nar/tests/symlink.narbin0 -> 136 bytes
-rw-r--r--tvix/nix-compat/src/nar/wire/mod.rs133
-rw-r--r--tvix/nix-compat/src/nar/wire/tag.rs165
-rw-r--r--tvix/nix-compat/src/nar/writer/async.rs235
-rw-r--r--tvix/nix-compat/src/nar/writer/mod.rs9
-rw-r--r--tvix/nix-compat/src/nar/writer/sync.rs224
-rw-r--r--tvix/nix-compat/src/nar/writer/test.rs133
-rw-r--r--tvix/nix-compat/src/narinfo/fingerprint.rs50
-rw-r--r--tvix/nix-compat/src/narinfo/mod.rs527
-rw-r--r--tvix/nix-compat/src/narinfo/public_keys.rs149
-rw-r--r--tvix/nix-compat/src/narinfo/signature.rs132
-rw-r--r--tvix/nix-compat/src/nixbase32.rs203
-rw-r--r--tvix/nix-compat/src/nixhash/algos.rs56
-rw-r--r--tvix/nix-compat/src/nixhash/ca_hash.rs213
-rw-r--r--tvix/nix-compat/src/nixhash/mod.rs530
-rw-r--r--tvix/nix-compat/src/store_path/mod.rs430
-rw-r--r--tvix/nix-compat/src/store_path/utils.rs296
-rw-r--r--tvix/nix-compat/testdata/narinfo.zstbin0 -> 975945 bytes
62 files changed, 6442 insertions, 0 deletions
diff --git a/tvix/nix-compat/Cargo.toml b/tvix/nix-compat/Cargo.toml
new file mode 100644
index 000000000000..2e4bff71aa42
--- /dev/null
+++ b/tvix/nix-compat/Cargo.toml
@@ -0,0 +1,47 @@
+[package]
+name = "nix-compat"
+version = "0.1.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[features]
+async = ["futures-util"]
+
+[dependencies]
+bitflags = "2.4.1"
+bstr = { version = "1.6.0", features = ["alloc", "unicode", "serde"] }
+data-encoding = "2.3.3"
+ed25519 = "2.2.3"
+ed25519-dalek = "2.1.0"
+futures-util = { version = "0.3.30", features = ["io"], optional = true }
+glob = "0.3.0"
+nom = "7.1.3"
+serde = { version = "1.0", features = ["derive"] }
+serde_json = "1.0"
+sha2 = "0.10.6"
+thiserror = "1.0.38"
+
+[dev-dependencies]
+futures = { version = "0.3.30", default-features = false, features = ["executor"] }
+lazy_static = "1.4.0"
+serde_json = "1.0"
+test-case = "3.3.1"
+criterion = { version = "0.5", features = ["html_reports"] }
+hex-literal = "0.4.1"
+pretty_assertions = "1.4.0"
+zstd = "^0.9.0"
+
+[dev-dependencies.test-generator]
+# This fork of test-generator adds support for cargo workspaces, see
+# also https://github.com/frehberg/test-generator/pull/14
+git = "https://github.com/JamesGuthrie/test-generator.git"
+rev = "82e799979980962aec1aa324ec6e0e4cad781f41"
+
+[[bench]]
+name = "derivation_parse_aterm"
+harness = false
+
+[[bench]]
+name = "narinfo_parse"
+harness = false
diff --git a/tvix/nix-compat/benches/derivation_parse_aterm.rs b/tvix/nix-compat/benches/derivation_parse_aterm.rs
new file mode 100644
index 000000000000..4ace7d4480f4
--- /dev/null
+++ b/tvix/nix-compat/benches/derivation_parse_aterm.rs
@@ -0,0 +1,31 @@
+use std::path::Path;
+
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use nix_compat::derivation::Derivation;
+
+const RESOURCES_PATHS: &str = "src/derivation/tests/derivation_tests/ok";
+
+fn bench_aterm_parser(c: &mut Criterion) {
+    for drv in [
+        "0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv",
+        "292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv",
+        "4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv",
+        "52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv",
+        "9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv",
+        "ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv",
+        "h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv",
+        "m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv",
+        "ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv",
+        "x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv",
+    ] {
+        let drv_path = Path::new(RESOURCES_PATHS).join(drv);
+        let drv_bytes = &std::fs::read(drv_path).unwrap();
+
+        c.bench_function(drv, |b| {
+            b.iter(|| Derivation::from_aterm_bytes(black_box(drv_bytes)))
+        });
+    }
+}
+
+criterion_group!(benches, bench_aterm_parser);
+criterion_main!(benches);
diff --git a/tvix/nix-compat/benches/narinfo_parse.rs b/tvix/nix-compat/benches/narinfo_parse.rs
new file mode 100644
index 000000000000..7ffd24d12bc3
--- /dev/null
+++ b/tvix/nix-compat/benches/narinfo_parse.rs
@@ -0,0 +1,69 @@
+use criterion::{black_box, criterion_group, criterion_main, Criterion, Throughput};
+use lazy_static::lazy_static;
+use nix_compat::narinfo::NarInfo;
+use std::{io, str};
+
+const SAMPLE: &str = r#"StorePath: /nix/store/1pajsq519irjy86vli20bgq1wr1q3pny-banking-0.3.0
+URL: nar/0rdn027rxqbl42bv9jxhsipgq2hwqdapvwmdzligmzdmz2p9vybs.nar.xz
+Compression: xz
+FileHash: sha256:0rdn027rxqbl42bv9jxhsipgq2hwqdapvwmdzligmzdmz2p9vybs
+FileSize: 92828
+NarHash: sha256:0cfnydzp132y69bh20dj76yfd6hc3qdyblbwr9hwn59vfmnb09m0
+NarSize: 173352
+References: 03d4ncyfh76mgs6sfayl8l6zzdhm219w-python3.9-mt-940-4.23.0 0rhbw783qcjxv3cqln1760i1lmz2yb67-gsettings-desktop-schemas-41.0 1dm9ndgg56ylawpcbdzkhl03fg6777rr-python3.9-six-1.16.0 1pajsq519irjy86vli20bgq1wr1q3pny-banking-0.3.0 2ccy5zc89zpc2aznqxgvzp4wm1bwj05n-bzip2-1.0.6.0.2-bin 32gy3pqk4n725lscdm622yzsg9np3xvs-python3.9-cryptography-36.0.0-dev 35chvqbr7vp9icdki0132fc6np09vrx5-python3.9-bleach-4.1.0 53abh5cz9zi4yh75lfzg99xqy0fdgj4i-python3.9-xmlschema-1.9.2 5p96sifyavb407mnharhyzlw6pn6km1b-glib-2.70.2-bin 6hil8z0zkqcgvaw1qwjyqa8qyaa1lm3k-python3.9-pycairo-1.20.1 803ffb21rv4af521pplb72zjm1ygm9kk-python3.9-pyparsing-2.4.7 al95l8psvmq5di3vdwa75n8w2m0sj2sy-gdk-pixbuf-2.42.6 b09371lq1jjrv43h8jpp82v23igndsn2-python3.9-fints-3.0.1 b53hk557pdk5mq4lv1zrh71a54qazbsm-python3.9-certifi-2021.10.08 bl0cwvwgch92cfsnli4dsah2gxgdickp-gtk+3-3.24.30 cfkq9wi7ypqk26c75dzic5v3nxlzyi58-python3.9-cryptography-36.0.0 cyhg57whqvrx7xf7fvn70dr5836y7zak-python3.9-sepaxml-2.4.1 d810g729g1c4lvp3nv1n3ah6cvpwg7by-cairo-1.16.0-dev dn4fwp0yx6nsa85cr20cwvdmg64xwmcy-python3-3.9.9 dzsj2n0nmq8nv6w0hvy5vb61kim3rzmd-pango-1.50.0 fs6rcnhbjvpxsyw5qiq0q7jx378fjrq7-python3.9-webencodings-0.5.1 g08sxarx191yh2dh0yk2j8icja54aksf-harfbuzz-3.1.2 glanz2lv7m6ak8pql0jcpr3izyp5cxm5-python3.9-pycparser-2.21 gpzx6h0dp5yhcvkfj68zs444ghll7dzm-python3.9-html5lib-1.1 gxyhqkpahahn4h8wbanzfhr1zkxbysid-expat-2.4.2-dev gy3pnc7bpff1h4ylhrivs4cjlvmxl0dk-python3.9-packaging-20.9 hhpqldw0552mf4mjdm2q7zqwy9hpfchd-libpng-apng-1.6.37-dev ig2bdwmplvs6dyg07fdyh006ha768jh1-python3.9-cffi-1.15.0 ij5rm5y6lmqzrwqd1zxckhbii3dg2nq5-glib-2.70.2-dev j5raylzz6fsafbgayyfaydadjl0x22s0-freetype-2.11.1-dev j6w2fbsl49jska4scyr860gz4df9biha-gobject-introspection-1.70.0 jfc99f1hrca6ih6h0n4ax431hjlx96j0-python3.9-brotli-1.0.9 kbazcxnki2qz514rl1plhsj3587hl8bb-python3.9-pysocks-1.7.1 kkljrrrj80fnz59qyfgnv6wvv0cbmpql-libhandy-1.5.0 l82il2lbp757c0smi81qmj4crlcmdz9s-python3.9-pygobject-3.42.0-dev m4zflhr10wz4frhgxqfi43rwvapki1pi-fontconfig-2.13.94-bin mbsc1c7mq15vgfzcdma9fglczih9ncfy-python3.9-chardet-4.0.0 mfvaaf4illpwrflg30cij5x4rncp9jin-python3.9-text-unidecode-1.3 msiv2nkdcaf4gvaf2cfnxcjm66j8mjxz-python3.9-elementpath-2.4.0 nmwapds8fcx22vd30d81va7a7a51ywwx-gettext-0.21 pbfraw351mksnkp2ni9c4rkc9cpp89iv-bash-5.1-p12 r8cbf18vrd54rb4psf3m4zlk5sd2jsv3-python3.9-pygobject-3.42.0 rig6npd9sd45ashf6fxcwgxzm7m4p0l3-python3.9-requests-2.26.0 ryj72ashr27gf4kh0ssgi3zpiv8fxw53-librsvg-2.52.4 s2jjq7rk5yrzlv9lyralzvpixg4p6jh3-atk-2.36.0 w1lsr2i37fr0mp1jya04nwa5nf5dxm2n-python3.9-setuptools-57.2.0 whfykra99ahs814l5hp3q5ps8rwzsf3s-python3.9-brotlicffi-1.0.9.2 wqdmghdvc4s95jgpp13fj5v3xar8mlks-python3.9-charset-normalizer-2.0.8 x1ha2nyji1px0iqknbyhdnvw4icw5h3i-python3.9-idna-3.3 z9560qb4ygbi0352m9pglwhi332cxb1f-python3.9-urllib3-1.26.7
+Deriver: 2ch8jx910qk6721mp4yqsmvdfgj5c8ir-banking-0.3.0.drv
+Sig: cache.nixos.org-1:xcL67rBZPcdVZudDLpLeddkBa0KaFTw5A0udnaa0axysjrQ6Nvd9p3BLZ4rhKgl52/cKiU3c6aq60L8+IcE5Dw==
+"#;
+
+lazy_static! {
+    static ref CASES: &'static [&'static str] = {
+        let data =
+            zstd::decode_all(io::Cursor::new(include_bytes!("../testdata/narinfo.zst"))).unwrap();
+        let data = str::from_utf8(Vec::leak(data)).unwrap();
+        Vec::leak(
+            data.split_inclusive("\n\n")
+                .map(|s| s.strip_suffix('\n').unwrap())
+                .collect::<Vec<_>>(),
+        )
+    };
+}
+
+pub fn parse(c: &mut Criterion) {
+    let mut g = c.benchmark_group("parse");
+
+    {
+        g.throughput(Throughput::Bytes(SAMPLE.len() as u64));
+        g.bench_with_input("single", SAMPLE, |b, data| {
+            b.iter(|| {
+                black_box(NarInfo::parse(black_box(data)).ok().unwrap());
+            });
+        });
+    }
+
+    {
+        for &case in *CASES {
+            NarInfo::parse(case).expect("should parse");
+        }
+
+        g.throughput(Throughput::Bytes(
+            CASES.iter().map(|s| s.len() as u64).sum(),
+        ));
+        g.bench_with_input("many", &*CASES, |b, data| {
+            let mut vec = vec![];
+            b.iter(|| {
+                vec.clear();
+                vec.extend(
+                    black_box(data)
+                        .iter()
+                        .map(|s| NarInfo::parse(s).ok().unwrap()),
+                );
+                black_box(&vec);
+            });
+        });
+    }
+
+    g.finish();
+}
+
+criterion_group!(benches, parse);
+criterion_main!(benches);
diff --git a/tvix/nix-compat/default.nix b/tvix/nix-compat/default.nix
new file mode 100644
index 000000000000..d6169f1339cf
--- /dev/null
+++ b/tvix/nix-compat/default.nix
@@ -0,0 +1,7 @@
+{ depot, ... }:
+
+depot.tvix.crates.workspaceMembers.nix-compat.build.override {
+  runTests = true;
+  # make sure we also enable async here, so run the tests behind that feature flag.
+  features = [ "default" "async" ];
+}
diff --git a/tvix/nix-compat/src/aterm/escape.rs b/tvix/nix-compat/src/aterm/escape.rs
new file mode 100644
index 000000000000..06b550bbf02d
--- /dev/null
+++ b/tvix/nix-compat/src/aterm/escape.rs
@@ -0,0 +1,27 @@
+use bstr::ByteSlice;
+
+/// Escapes a byte sequence. Does not add surrounding quotes.
+pub fn escape_bytes<P: AsRef<[u8]>>(s: P) -> Vec<u8> {
+    let mut s: Vec<u8> = s.as_ref().to_vec();
+
+    s = s.replace(b"\\", b"\\\\");
+    s = s.replace(b"\n", b"\\n");
+    s = s.replace(b"\r", b"\\r");
+    s = s.replace(b"\t", b"\\t");
+    s = s.replace(b"\"", b"\\\"");
+
+    s
+}
+
+#[cfg(test)]
+mod tests {
+    use super::escape_bytes;
+    use test_case::test_case;
+
+    #[test_case(b"", b""; "empty")]
+    #[test_case(b"\"", b"\\\""; "doublequote")]
+    #[test_case(b":", b":"; "colon")]
+    fn escape(input: &[u8], expected: &[u8]) {
+        assert_eq!(expected, escape_bytes(input))
+    }
+}
diff --git a/tvix/nix-compat/src/aterm/mod.rs b/tvix/nix-compat/src/aterm/mod.rs
new file mode 100644
index 000000000000..8806b6caf2e5
--- /dev/null
+++ b/tvix/nix-compat/src/aterm/mod.rs
@@ -0,0 +1,7 @@
+mod escape;
+mod parser;
+
+pub(crate) use escape::escape_bytes;
+pub(crate) use parser::parse_bstr_field;
+pub(crate) use parser::parse_str_list;
+pub(crate) use parser::parse_string_field;
diff --git a/tvix/nix-compat/src/aterm/parser.rs b/tvix/nix-compat/src/aterm/parser.rs
new file mode 100644
index 000000000000..72648d5ef590
--- /dev/null
+++ b/tvix/nix-compat/src/aterm/parser.rs
@@ -0,0 +1,118 @@
+//! This module implements parsing code for some basic building blocks
+//! of the [ATerm][] format, which is used by C++ Nix to serialize Derivations.
+//!
+//! [ATerm]: http://program-transformation.org/Tools/ATermFormat.html
+use bstr::BString;
+use nom::branch::alt;
+use nom::bytes::complete::{escaped_transform, is_not, tag};
+use nom::character::complete::char as nomchar;
+use nom::combinator::{map, value};
+use nom::multi::separated_list0;
+use nom::sequence::delimited;
+use nom::IResult;
+
+/// Parse a bstr and undo any escaping.
+fn parse_escaped_bstr(i: &[u8]) -> IResult<&[u8], BString> {
+    escaped_transform(
+        is_not("\"\\"),
+        '\\',
+        alt((
+            value("\\".as_bytes(), nomchar('\\')),
+            value("\n".as_bytes(), nomchar('n')),
+            value("\t".as_bytes(), nomchar('t')),
+            value("\r".as_bytes(), nomchar('r')),
+            value("\"".as_bytes(), nomchar('\"')),
+        )),
+    )(i)
+    .map(|(i, v)| (i, BString::new(v)))
+}
+
+/// Parse a field in double quotes, undo any escaping, and return the unquoted
+/// and decoded `Vec<u8>`.
+pub(crate) fn parse_bstr_field(i: &[u8]) -> IResult<&[u8], BString> {
+    // inside double quotes…
+    delimited(
+        nomchar('\"'),
+        // There is
+        alt((
+            // …either is a bstr after unescaping
+            parse_escaped_bstr,
+            // …or an empty string.
+            map(tag(b""), |_| BString::default()),
+        )),
+        nomchar('\"'),
+    )(i)
+}
+
+/// Parse a field in double quotes, undo any escaping, and return the unquoted
+/// and decoded string, if it's a valid string. Or fail parsing if the bytes are
+/// no valid UTF-8.
+pub(crate) fn parse_string_field(i: &[u8]) -> IResult<&[u8], String> {
+    // inside double quotes…
+    delimited(
+        nomchar('\"'),
+        // There is
+        alt((
+            // either is a String after unescaping
+            nom::combinator::map_opt(parse_escaped_bstr, |escaped_bstr| {
+                String::from_utf8(escaped_bstr.into()).ok()
+            }),
+            // or an empty string.
+            map(tag(b""), |_| String::new()),
+        )),
+        nomchar('\"'),
+    )(i)
+}
+
+/// Parse a list of of string fields (enclosed in brackets)
+pub(crate) fn parse_str_list(i: &[u8]) -> IResult<&[u8], Vec<String>> {
+    // inside brackets
+    delimited(
+        nomchar('['),
+        separated_list0(nomchar(','), parse_string_field),
+        nomchar(']'),
+    )(i)
+}
+
+#[cfg(test)]
+mod tests {
+    use test_case::test_case;
+
+    #[test_case(br#""""#, b"", b""; "empty")]
+    #[test_case(br#""Hello World""#, b"Hello World", b""; "hello world")]
+    #[test_case(br#""\"""#, br#"""#, b""; "doublequote")]
+    #[test_case(br#"":""#, b":", b""; "colon")]
+    #[test_case(br#""\""Rest"#, br#"""#, b"Rest"; "doublequote rest")]
+    fn parse_bstr_field(input: &[u8], expected: &[u8], exp_rest: &[u8]) {
+        let (rest, parsed) = super::parse_bstr_field(input).expect("must parse");
+        assert_eq!(exp_rest, rest, "expected remainder");
+        assert_eq!(expected, parsed);
+    }
+
+    #[test_case(br#""""#, "", b""; "empty")]
+    #[test_case(br#""Hello World""#, "Hello World", b""; "hello world")]
+    #[test_case(br#""\"""#, r#"""#, b""; "doublequote")]
+    #[test_case(br#"":""#, ":", b""; "colon")]
+    #[test_case(br#""\""Rest"#, r#"""#, b"Rest"; "doublequote rest")]
+    fn parse_string_field(input: &[u8], expected: &str, exp_rest: &[u8]) {
+        let (rest, parsed) = super::parse_string_field(input).expect("must parse");
+        assert_eq!(exp_rest, rest, "expected remainder");
+        assert_eq!(expected, &parsed);
+    }
+
+    #[test]
+    fn parse_string_field_invalid_encoding_fail() {
+        let input: Vec<_> = vec![b'"', 0xc5, 0xc4, 0xd6, b'"'];
+
+        super::parse_string_field(&input).expect_err("must fail");
+    }
+
+    #[test_case(br#"["foo"]"#, vec!["foo".to_string()], b""; "single foo")]
+    #[test_case(b"[]", vec![], b""; "empty list")]
+    #[test_case(b"[]blub", vec![], b"blub"; "empty list with rest")]
+    fn parse_list(input: &[u8], expected: Vec<String>, exp_rest: &[u8]) {
+        let (rest, parsed) = super::parse_str_list(input).expect("must parse");
+        assert_eq!(exp_rest, rest, "expected remainder");
+        assert_eq!(expected, parsed);
+    }
+}
diff --git a/tvix/nix-compat/src/bin/drvfmt.rs b/tvix/nix-compat/src/bin/drvfmt.rs
new file mode 100644
index 000000000000..ddc1f0389f26
--- /dev/null
+++ b/tvix/nix-compat/src/bin/drvfmt.rs
@@ -0,0 +1,42 @@
+use std::{collections::BTreeMap, io::Read};
+
+use nix_compat::derivation::Derivation;
+use serde_json::json;
+
+/// construct a serde_json::Value from a Derivation.
+/// Some environment values can be non-valid UTF-8 strings.
+/// `serde_json` prints them out really unreadable.
+/// This is a tool to print A-Terms in a more readable fashion, so we brutally
+/// use the [std::string::ToString] implementation of [bstr::BString] to get
+/// a UTF-8 string (replacing invalid characters with the Unicode replacement
+/// codepoint).
+fn build_serde_json_value(drv: Derivation) -> serde_json::Value {
+    json!({
+        "args": drv.arguments,
+        "builder": drv.builder,
+        "env":   drv.environment.into_iter().map(|(k,v)| (k, v.to_string())).collect::<BTreeMap<String, String>>(),
+        "inputDrvs": drv.input_derivations,
+        "inputSrcs": drv.input_sources,
+        "outputs": drv.outputs,
+        "system": drv.system,
+    })
+}
+
+fn main() {
+    // read A-Term from stdin
+    let mut buf = Vec::new();
+    std::io::stdin()
+        .read_to_end(&mut buf)
+        .expect("failed to read from stdin");
+
+    match Derivation::from_aterm_bytes(&buf) {
+        Ok(drv) => {
+            println!(
+                "{}",
+                serde_json::to_string_pretty(&build_serde_json_value(drv))
+                    .expect("unable to serialize")
+            );
+        }
+        Err(e) => eprintln!("unable to parse derivation: {:#?}", e),
+    }
+}
diff --git a/tvix/nix-compat/src/derivation/errors.rs b/tvix/nix-compat/src/derivation/errors.rs
new file mode 100644
index 000000000000..7dcc3a5d0042
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/errors.rs
@@ -0,0 +1,58 @@
+//! Contains [DerivationError], exported as [crate::derivation::DerivationError]
+use crate::store_path;
+use thiserror::Error;
+
+use super::CAHash;
+
+/// Errors that can occur during the validation of Derivation structs.
+#[derive(Debug, Error, PartialEq)]
+pub enum DerivationError {
+    // outputs
+    #[error("no outputs defined")]
+    NoOutputs(),
+    #[error("invalid output name: {0}")]
+    InvalidOutputName(String),
+    #[error("encountered fixed-output derivation, but more than 1 output in total")]
+    MoreThanOneOutputButFixed(),
+    #[error("invalid output name for fixed-output derivation: {0}")]
+    InvalidOutputNameForFixed(String),
+    #[error("unable to validate output {0}: {1}")]
+    InvalidOutput(String, OutputError),
+    #[error("unable to validate output {0}: {1}")]
+    InvalidOutputDerivationPath(String, store_path::BuildStorePathError),
+    // input derivation
+    #[error("unable to parse input derivation path {0}: {1}")]
+    InvalidInputDerivationPath(String, store_path::Error),
+    #[error("input derivation {0} doesn't end with .drv")]
+    InvalidInputDerivationPrefix(String),
+    #[error("input derivation {0} output names are empty")]
+    EmptyInputDerivationOutputNames(String),
+    #[error("input derivation {0} output name {1} is invalid")]
+    InvalidInputDerivationOutputName(String, String),
+
+    // input sources
+    #[error("unable to parse input sources path {0}: {1}")]
+    InvalidInputSourcesPath(String, store_path::Error),
+
+    // platform
+    #[error("invalid platform field: {0}")]
+    InvalidPlatform(String),
+
+    // builder
+    #[error("invalid builder field: {0}")]
+    InvalidBuilder(String),
+
+    // environment
+    #[error("invalid environment key {0}")]
+    InvalidEnvironmentKey(String),
+}
+
+/// Errors that can occur during the validation of a specific
+// [crate::derivation::Output] of a [crate::derivation::Derviation].
+#[derive(Debug, Error, PartialEq)]
+pub enum OutputError {
+    #[error("Invalid output path {0}: {1}")]
+    InvalidOutputPath(String, store_path::Error),
+    #[error("Invalid CAHash: {:?}", .0)]
+    InvalidCAHash(CAHash),
+}
diff --git a/tvix/nix-compat/src/derivation/mod.rs b/tvix/nix-compat/src/derivation/mod.rs
new file mode 100644
index 000000000000..2f90c54d8f76
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/mod.rs
@@ -0,0 +1,289 @@
+use crate::store_path::{
+    self, build_ca_path, build_output_path, build_text_path, StorePath, StorePathRef,
+};
+use bstr::BString;
+use serde::{Deserialize, Serialize};
+use sha2::{Digest, Sha256};
+use std::collections::{BTreeMap, BTreeSet};
+use std::io;
+
+mod errors;
+mod output;
+mod parse_error;
+mod parser;
+mod validate;
+mod write;
+
+#[cfg(test)]
+mod tests;
+
+// Public API of the crate.
+pub use crate::nixhash::{CAHash, NixHash};
+pub use errors::{DerivationError, OutputError};
+pub use output::Output;
+
+#[derive(Clone, Debug, Default, Eq, PartialEq, Serialize, Deserialize)]
+pub struct Derivation {
+    #[serde(rename = "args")]
+    pub arguments: Vec<String>,
+
+    pub builder: String,
+
+    #[serde(rename = "env")]
+    pub environment: BTreeMap<String, BString>,
+
+    #[serde(rename = "inputDrvs")]
+    pub input_derivations: BTreeMap<String, BTreeSet<String>>,
+
+    #[serde(rename = "inputSrcs")]
+    pub input_sources: BTreeSet<String>,
+
+    pub outputs: BTreeMap<String, Output>,
+
+    pub system: String,
+}
+
+impl Derivation {
+    /// write the Derivation to the given [std::io::Write], in ATerm format.
+    ///
+    /// The only errors returns are these when writing to the passed writer.
+    pub fn serialize(&self, writer: &mut impl std::io::Write) -> Result<(), io::Error> {
+        use write::*;
+
+        writer.write_all(write::DERIVATION_PREFIX.as_bytes())?;
+        write_char(writer, write::PAREN_OPEN)?;
+
+        write_outputs(writer, &self.outputs)?;
+        write_char(writer, COMMA)?;
+
+        write_input_derivations(writer, &self.input_derivations)?;
+        write_char(writer, COMMA)?;
+
+        write_input_sources(writer, &self.input_sources)?;
+        write_char(writer, COMMA)?;
+
+        write_system(writer, &self.system)?;
+        write_char(writer, COMMA)?;
+
+        write_builder(writer, &self.builder)?;
+        write_char(writer, COMMA)?;
+
+        write_arguments(writer, &self.arguments)?;
+        write_char(writer, COMMA)?;
+
+        write_environment(writer, &self.environment)?;
+
+        write_char(writer, PAREN_CLOSE)?;
+
+        Ok(())
+    }
+
+    /// return the ATerm serialization.
+    pub fn to_aterm_bytes(&self) -> Vec<u8> {
+        let mut buffer: Vec<u8> = Vec::new();
+
+        // invoke serialize and write to the buffer.
+        // Note we only propagate errors writing to the writer in serialize,
+        // which won't panic for the string we write to.
+        self.serialize(&mut buffer).unwrap();
+
+        buffer
+    }
+
+    /// Parse an Derivation in ATerm serialization, and validate it passes our
+    /// set of validations.
+    pub fn from_aterm_bytes(b: &[u8]) -> Result<Derivation, parser::Error<&[u8]>> {
+        parser::parse(b)
+    }
+
+    /// Returns the drv path of a [Derivation] struct.
+    ///
+    /// The drv path is calculated by invoking [build_text_path], using
+    /// the `name` with a `.drv` suffix as name, all [Derivation::input_sources] and
+    /// keys of [Derivation::input_derivations] as references, and the ATerm string of
+    /// the [Derivation] as content.
+    pub fn calculate_derivation_path(&self, name: &str) -> Result<StorePath, DerivationError> {
+        // append .drv to the name
+        let name = &format!("{}.drv", name);
+
+        // collect the list of paths from input_sources and input_derivations
+        // into a (sorted, guaranteed by BTreeSet) list of references
+        let references: BTreeSet<String> = {
+            let mut inputs = self.input_sources.clone();
+            let input_derivation_keys: Vec<String> =
+                self.input_derivations.keys().cloned().collect();
+            inputs.extend(input_derivation_keys);
+            inputs
+        };
+
+        build_text_path(name, self.to_aterm_bytes(), references)
+            .map(|s| s.to_owned())
+            .map_err(|_e| DerivationError::InvalidOutputName(name.to_string()))
+    }
+
+    /// Returns the FOD digest, if the derivation is fixed-output, or None if
+    /// it's not.
+    /// TODO: this is kinda the string from [build_ca_path] with a
+    /// [CAHash::Flat], what's fed to `build_store_path_from_fingerprint_parts`
+    /// (except the out_output.path being an empty string)
+    fn fod_digest(&self) -> Option<[u8; 32]> {
+        if self.outputs.len() != 1 {
+            return None;
+        }
+
+        let out_output = self.outputs.get("out")?;
+        let ca_hash = &out_output.ca_hash.as_ref()?;
+
+        Some(
+            Sha256::new_with_prefix(format!(
+                "fixed:out:{}{}:{}",
+                ca_kind_prefix(ca_hash),
+                ca_hash.digest().to_nix_hex_string(),
+                out_output.path
+            ))
+            .finalize()
+            .into(),
+        )
+    }
+
+    /// Calculates the hash of a derivation modulo fixed-output subderivations.
+    ///
+    /// This is called `hashDerivationModulo` in nixcpp.
+    ///
+    /// It returns a [NixHash], created by calculating the sha256 digest of
+    /// the derivation ATerm representation, except that:
+    ///  -  any input derivation paths have beed replaced "by the result of a
+    ///     recursive call to this function" and that
+    ///  - for fixed-output derivations the special
+    ///    `fixed:out:${algo}:${digest}:${fodPath}` string is hashed instead of
+    ///    the A-Term.
+    ///
+    /// If the derivation is not a fixed derivation, it's up to the caller of
+    /// this function to provide a lookup function to lookup these calculation
+    /// results of parent derivations at `fn_get_hash_derivation_modulo` (by
+    /// drv path).
+    pub fn derivation_or_fod_hash<F>(&self, fn_get_derivation_or_fod_hash: F) -> NixHash
+    where
+        F: Fn(&StorePathRef) -> NixHash,
+    {
+        // Fixed-output derivations return a fixed hash.
+        // Non-Fixed-output derivations return a hash of the ATerm notation, but with all
+        // input_derivation paths replaced by a recursive call to this function.
+        // We use fn_get_derivation_or_fod_hash here, so callers can precompute this.
+        let digest = self.fod_digest().unwrap_or({
+            // This is a new map from derivation_or_fod_hash.digest (as lowerhex)
+            // to list of output names
+            let mut replaced_input_derivations: BTreeMap<String, BTreeSet<String>> =
+                BTreeMap::new();
+
+            // For each input_derivation, look up the
+            // derivation_or_fod_hash, and replace the derivation path with it's HEXLOWER
+            // digest.
+            // This is not the [NixHash::to_nix_hash_string], but without the sha256: prefix).
+            for (drv_path_str, output_names) in &self.input_derivations {
+                // parse drv_path to StorePathRef
+                let drv_path = StorePathRef::from_absolute_path(drv_path_str.as_bytes())
+                    .expect("invalid input derivation path");
+                replaced_input_derivations.insert(
+                    data_encoding::HEXLOWER
+                        .encode(fn_get_derivation_or_fod_hash(&drv_path).digest_as_bytes()),
+                    output_names.clone(),
+                );
+            }
+
+            // construct a new derivation struct with these replaced input derivation strings
+            let replaced_derivation = Derivation {
+                input_derivations: replaced_input_derivations,
+                ..self.clone()
+            };
+
+            // write the ATerm of that to the hash function
+            let mut hasher = Sha256::new();
+            hasher.update(replaced_derivation.to_aterm_bytes());
+
+            hasher.finalize().into()
+        });
+
+        NixHash::Sha256(digest)
+    }
+
+    /// This calculates all output paths of a Derivation and updates the struct.
+    /// It requires the struct to be initially without output paths.
+    /// This means, self.outputs[$outputName].path needs to be an empty string,
+    /// and self.environment[$outputName] needs to be an empty string.
+    ///
+    /// Output path calculation requires knowledge of the
+    /// derivation_or_fod_hash [NixHash], which (in case of non-fixed-output
+    /// derivations) also requires knowledge of other hash_derivation_modulo
+    /// [NixHash]es.
+    ///
+    /// We solve this by asking the caller of this function to provide the
+    /// hash_derivation_modulo of the current Derivation.
+    ///
+    /// On completion, self.environment[$outputName] and
+    /// self.outputs[$outputName].path are set to the calculated output path for all
+    /// outputs.
+    pub fn calculate_output_paths(
+        &mut self,
+        name: &str,
+        derivation_or_fod_hash: &NixHash,
+    ) -> Result<(), DerivationError> {
+        // The fingerprint and hash differs per output
+        for (output_name, output) in self.outputs.iter_mut() {
+            // Assert that outputs are not yet populated, to avoid using this function wrongly.
+            // We don't also go over self.environment, but it's a sufficient
+            // footgun prevention mechanism.
+            assert!(output.path.is_empty());
+
+            let path_name = output_path_name(name, output_name);
+
+            // For fixed output derivation we use the per-output info, otherwise we use the
+            // derivation hash.
+            let abs_store_path = if let Some(ref hwm) = output.ca_hash {
+                build_ca_path(&path_name, hwm, Vec::<String>::new(), false).map_err(|e| {
+                    DerivationError::InvalidOutputDerivationPath(output_name.to_string(), e)
+                })?
+            } else {
+                build_output_path(derivation_or_fod_hash, output_name, &path_name).map_err(|e| {
+                    DerivationError::InvalidOutputDerivationPath(
+                        output_name.to_string(),
+                        store_path::BuildStorePathError::InvalidStorePath(e),
+                    )
+                })?
+            };
+
+            output.path = abs_store_path.to_absolute_path();
+            self.environment.insert(
+                output_name.to_string(),
+                abs_store_path.to_absolute_path().into(),
+            );
+        }
+
+        Ok(())
+    }
+}
+
+/// Calculate the name part of the store path of a derivation [Output].
+///
+/// It's the name, and (if it's the non-out output), the output name
+/// after a `-`.
+fn output_path_name(derivation_name: &str, output_name: &str) -> String {
+    let mut output_path_name = derivation_name.to_string();
+    if output_name != "out" {
+        output_path_name.push('-');
+        output_path_name.push_str(output_name);
+    }
+    output_path_name
+}
+
+/// For a [CAHash], return the "prefix" used for NAR purposes.
+/// For [CAHash::Flat], this is an empty string, for [CAHash::Nar], it's "r:".
+/// Panics for other [CAHash] kinds, as they're not valid in a derivation
+/// context.
+fn ca_kind_prefix(ca_hash: &CAHash) -> &'static str {
+    match ca_hash {
+        CAHash::Flat(_) => "",
+        CAHash::Nar(_) => "r:",
+        _ => panic!("invalid ca hash in derivation context: {:?}", ca_hash),
+    }
+}
diff --git a/tvix/nix-compat/src/derivation/output.rs b/tvix/nix-compat/src/derivation/output.rs
new file mode 100644
index 000000000000..b7fa1ac379b0
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/output.rs
@@ -0,0 +1,146 @@
+use crate::derivation::OutputError;
+use crate::nixhash::CAHash;
+use crate::store_path::StorePathRef;
+use serde::{Deserialize, Serialize};
+use serde_json::Map;
+
+#[derive(Clone, Debug, Default, Eq, PartialEq, Serialize)]
+pub struct Output {
+    pub path: String,
+
+    #[serde(flatten)]
+    pub ca_hash: Option<CAHash>, // we can only represent a subset here.
+}
+
+impl<'de> Deserialize<'de> for Output {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        let fields = Map::deserialize(deserializer)?;
+        Ok(Self {
+            path: fields
+                .get("path")
+                .ok_or(serde::de::Error::missing_field(
+                    "`path` is missing but required for outputs",
+                ))?
+                .as_str()
+                .ok_or(serde::de::Error::invalid_type(
+                    serde::de::Unexpected::Other("certainly not a string"),
+                    &"a string",
+                ))?
+                .to_owned(),
+            ca_hash: CAHash::from_map::<D>(&fields)?,
+        })
+    }
+}
+
+impl Output {
+    pub fn is_fixed(&self) -> bool {
+        self.ca_hash.is_some()
+    }
+
+    pub fn validate(&self, validate_output_paths: bool) -> Result<(), OutputError> {
+        if let Some(fixed_output_hash) = &self.ca_hash {
+            match fixed_output_hash {
+                CAHash::Flat(_) | CAHash::Nar(_) => {
+                    // all hashes allowed for Flat, and Nar.
+                }
+                _ => return Err(OutputError::InvalidCAHash(fixed_output_hash.clone())),
+            }
+        }
+
+        if validate_output_paths {
+            if let Err(e) = StorePathRef::from_absolute_path(self.path.as_bytes()) {
+                return Err(OutputError::InvalidOutputPath(self.path.to_string(), e));
+            }
+        }
+        Ok(())
+    }
+}
+
+/// This ensures that a potentially valid input addressed
+/// output is deserialized as a non-fixed output.
+#[test]
+fn deserialize_valid_input_addressed_output() {
+    let json_bytes = r#"
+    {
+      "path": "/nix/store/blablabla"
+    }"#;
+    let output: Output = serde_json::from_str(json_bytes).expect("must parse");
+
+    assert!(!output.is_fixed());
+}
+
+/// This ensures that a potentially valid fixed output
+/// output deserializes fine as a fixed output.
+#[test]
+fn deserialize_valid_fixed_output() {
+    let json_bytes = r#"
+    {
+        "path": "/nix/store/blablablabla",
+        "hash": "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba",
+        "hashAlgo": "r:sha256"
+    }"#;
+    let output: Output = serde_json::from_str(json_bytes).expect("must parse");
+
+    assert!(output.is_fixed());
+}
+
+/// This ensures that parsing an input with the invalid hash encoding
+/// will result in a parsing failure.
+#[test]
+fn deserialize_with_error_invalid_hash_encoding_fixed_output() {
+    let json_bytes = r#"
+    {
+        "path": "/nix/store/blablablabla",
+        "hash": "IAMNOTVALIDNIXBASE32",
+        "hashAlgo": "r:sha256"
+    }"#;
+    let output: Result<Output, _> = serde_json::from_str(json_bytes);
+
+    assert!(output.is_err());
+}
+
+/// This ensures that parsing an input with the wrong hash algo
+/// will result in a parsing failure.
+#[test]
+fn deserialize_with_error_invalid_hash_algo_fixed_output() {
+    let json_bytes = r#"
+    {
+        "path": "/nix/store/blablablabla",
+        "hash": "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba",
+        "hashAlgo": "r:sha1024"
+    }"#;
+    let output: Result<Output, _> = serde_json::from_str(json_bytes);
+
+    assert!(output.is_err());
+}
+
+/// This ensures that parsing an input with the missing hash algo but present hash will result in a
+/// parsing failure.
+#[test]
+fn deserialize_with_error_missing_hash_algo_fixed_output() {
+    let json_bytes = r#"
+    {
+        "path": "/nix/store/blablablabla",
+        "hash": "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba",
+    }"#;
+    let output: Result<Output, _> = serde_json::from_str(json_bytes);
+
+    assert!(output.is_err());
+}
+
+/// This ensures that parsing an input with the missing hash but present hash algo will result in a
+/// parsing failure.
+#[test]
+fn deserialize_with_error_missing_hash_fixed_output() {
+    let json_bytes = r#"
+    {
+        "path": "/nix/store/blablablabla",
+        "hashAlgo": "r:sha1024"
+    }"#;
+    let output: Result<Output, _> = serde_json::from_str(json_bytes);
+
+    assert!(output.is_err());
+}
diff --git a/tvix/nix-compat/src/derivation/parse_error.rs b/tvix/nix-compat/src/derivation/parse_error.rs
new file mode 100644
index 000000000000..26df13f5772b
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/parse_error.rs
@@ -0,0 +1,81 @@
+//! This contains error and result types that can happen while parsing
+//! Derivations from ATerm.
+use nom::IResult;
+
+use crate::nixhash;
+
+pub type NomResult<I, O> = IResult<I, O, NomError<I>>;
+
+#[derive(Debug, thiserror::Error, PartialEq)]
+pub enum ErrorKind {
+    /// duplicate key in map
+    #[error("duplicate map key: {0}")]
+    DuplicateMapKey(String),
+
+    /// Input derivation has two outputs with the same name
+    #[error("duplicate output name {1} for input derivation {0}")]
+    DuplicateInputDerivationOutputName(String, String),
+
+    #[error("duplicate input source: {0}")]
+    DuplicateInputSource(String),
+
+    #[error("nix hash error: {0}")]
+    NixHashError(nixhash::Error),
+
+    #[error("nom error: {0:?}")]
+    Nom(nom::error::ErrorKind),
+}
+
+/// Our own error type to pass along parser-related errors.
+#[derive(Debug, PartialEq)]
+pub struct NomError<I> {
+    /// position of the error in the input data
+    pub input: I,
+    /// error code
+    pub code: ErrorKind,
+}
+
+impl<I, E> nom::error::FromExternalError<I, E> for NomError<I> {
+    fn from_external_error(input: I, kind: nom::error::ErrorKind, _e: E) -> Self {
+        Self {
+            input,
+            code: ErrorKind::Nom(kind),
+        }
+    }
+}
+
+impl<I> nom::error::ParseError<I> for NomError<I> {
+    fn from_error_kind(input: I, kind: nom::error::ErrorKind) -> Self {
+        Self {
+            input,
+            code: ErrorKind::Nom(kind),
+        }
+    }
+
+    // FUTUREWORK: implement, so we have support for backtracking through the
+    // parse tree?
+    fn append(_input: I, _kind: nom::error::ErrorKind, other: Self) -> Self {
+        other
+    }
+}
+
+/// This wraps a [nom::error::Error] into our error.
+impl<I> From<nom::error::Error<I>> for NomError<I> {
+    fn from(value: nom::error::Error<I>) -> Self {
+        Self {
+            input: value.input,
+            code: ErrorKind::Nom(value.code),
+        }
+    }
+}
+
+/// This essentially implements
+/// `From<nom::Err<nom::error::Error<I>>>` for `nom::Err<NomError<I>>`,
+/// which we can't because `nom::Err<_>` is a foreign type.
+pub(crate) fn into_nomerror<I>(e: nom::Err<nom::error::Error<I>>) -> nom::Err<NomError<I>> {
+    match e {
+        nom::Err::Incomplete(n) => nom::Err::Incomplete(n),
+        nom::Err::Error(e) => nom::Err::Error(e.into()),
+        nom::Err::Failure(e) => nom::Err::Failure(e.into()),
+    }
+}
diff --git a/tvix/nix-compat/src/derivation/parser.rs b/tvix/nix-compat/src/derivation/parser.rs
new file mode 100644
index 000000000000..7ffa6fd46eb6
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/parser.rs
@@ -0,0 +1,510 @@
+//! This module constructs a [Derivation] by parsing its [ATerm][]
+//! serialization.
+//!
+//! [ATerm]: http://program-transformation.org/Tools/ATermFormat.html
+
+use bstr::BString;
+use nom::bytes::complete::tag;
+use nom::character::complete::char as nomchar;
+use nom::combinator::{all_consuming, map_res};
+use nom::multi::{separated_list0, separated_list1};
+use nom::sequence::{delimited, preceded, separated_pair, terminated, tuple};
+use std::collections::{btree_map, BTreeMap, BTreeSet};
+use thiserror;
+
+use crate::derivation::parse_error::{into_nomerror, ErrorKind, NomError, NomResult};
+use crate::derivation::{write, CAHash, Derivation, Output};
+use crate::{aterm, nixhash};
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error<I> {
+    #[error("parsing error: {0}")]
+    Parser(NomError<I>),
+    #[error("premature EOF")]
+    Incomplete,
+    #[error("validation error: {0}")]
+    Validation(super::DerivationError),
+}
+
+pub(crate) fn parse(i: &[u8]) -> Result<Derivation, Error<&[u8]>> {
+    match all_consuming(parse_derivation)(i) {
+        Ok((rest, derivation)) => {
+            // this shouldn't happen, as all_consuming shouldn't return.
+            debug_assert!(rest.is_empty());
+
+            // invoke validate
+            derivation.validate(true).map_err(Error::Validation)?;
+
+            Ok(derivation)
+        }
+        Err(nom::Err::Incomplete(_)) => Err(Error::Incomplete),
+        Err(nom::Err::Error(e) | nom::Err::Failure(e)) => Err(Error::Parser(e)),
+    }
+}
+
+/// Consume a string containing the algo, and optionally a `r:`
+/// prefix, and a digest (bytes), return a [CAHash::Nar] or [CAHash::Flat].
+fn from_algo_and_mode_and_digest<B: AsRef<[u8]>>(
+    algo_and_mode: &str,
+    digest: B,
+) -> crate::nixhash::Result<CAHash> {
+    Ok(match algo_and_mode.strip_prefix("r:") {
+        Some(algo) => nixhash::CAHash::Nar(nixhash::from_algo_and_digest(
+            algo.try_into()?,
+            digest.as_ref(),
+        )?),
+        None => nixhash::CAHash::Flat(nixhash::from_algo_and_digest(
+            algo_and_mode.try_into()?,
+            digest.as_ref(),
+        )?),
+    })
+}
+
+/// Parse one output in ATerm. This is 4 string fields inside parans:
+/// output name, output path, algo (and mode), digest.
+/// Returns the output name and [Output] struct.
+fn parse_output(i: &[u8]) -> NomResult<&[u8], (String, Output)> {
+    delimited(
+        nomchar('('),
+        map_res(
+            |i| {
+                tuple((
+                    terminated(aterm::parse_string_field, nomchar(',')),
+                    terminated(aterm::parse_string_field, nomchar(',')),
+                    terminated(aterm::parse_string_field, nomchar(',')),
+                    aterm::parse_bstr_field,
+                ))(i)
+                .map_err(into_nomerror)
+            },
+            |(output_name, output_path, algo_and_mode, encoded_digest)| {
+                // convert these 4 fields into an [Output].
+                let ca_hash_res = {
+                    if algo_and_mode.is_empty() && encoded_digest.is_empty() {
+                        None
+                    } else {
+                        match data_encoding::HEXLOWER.decode(&encoded_digest) {
+                            Ok(digest) => {
+                                Some(from_algo_and_mode_and_digest(&algo_and_mode, digest))
+                            }
+                            Err(e) => Some(Err(nixhash::Error::InvalidBase64Encoding(e))),
+                        }
+                    }
+                }
+                .transpose();
+
+                match ca_hash_res {
+                    Ok(hash_with_mode) => Ok((
+                        output_name,
+                        Output {
+                            path: output_path,
+                            ca_hash: hash_with_mode,
+                        },
+                    )),
+                    Err(e) => Err(nom::Err::Failure(NomError {
+                        input: i,
+                        code: ErrorKind::NixHashError(e),
+                    })),
+                }
+            },
+        ),
+        nomchar(')'),
+    )(i)
+}
+
+/// Parse multiple outputs in ATerm. This is a list of things acccepted by
+/// parse_output, and takes care of turning the (String, Output) returned from
+/// it to a BTreeMap.
+/// We don't use parse_kv here, as it's dealing with 2-tuples, and these are
+/// 4-tuples.
+fn parse_outputs(i: &[u8]) -> NomResult<&[u8], BTreeMap<String, Output>> {
+    let res = delimited(
+        nomchar('['),
+        separated_list1(tag(","), parse_output),
+        nomchar(']'),
+    )(i);
+
+    match res {
+        Ok((rst, outputs_lst)) => {
+            let mut outputs: BTreeMap<String, Output> = BTreeMap::default();
+            for (output_name, output) in outputs_lst.into_iter() {
+                if outputs.contains_key(&output_name) {
+                    return Err(nom::Err::Failure(NomError {
+                        input: i,
+                        code: ErrorKind::DuplicateMapKey(output_name),
+                    }));
+                }
+                outputs.insert(output_name, output);
+            }
+            Ok((rst, outputs))
+        }
+        // pass regular parse errors along
+        Err(e) => Err(e),
+    }
+}
+
+fn parse_input_derivations(i: &[u8]) -> NomResult<&[u8], BTreeMap<String, BTreeSet<String>>> {
+    let (i, input_derivations_list) = parse_kv::<Vec<String>, _>(aterm::parse_str_list)(i)?;
+
+    // This is a HashMap of drv paths to a list of output names.
+    let mut input_derivations: BTreeMap<String, BTreeSet<String>> = BTreeMap::new();
+
+    for (input_derivation, output_names) in input_derivations_list {
+        let mut new_output_names = BTreeSet::new();
+        for output_name in output_names.into_iter() {
+            if new_output_names.contains(&output_name) {
+                return Err(nom::Err::Failure(NomError {
+                    input: i,
+                    code: ErrorKind::DuplicateInputDerivationOutputName(
+                        input_derivation.to_string(),
+                        output_name.to_string(),
+                    ),
+                }));
+            } else {
+                new_output_names.insert(output_name);
+            }
+        }
+        input_derivations.insert(input_derivation, new_output_names);
+    }
+
+    Ok((i, input_derivations))
+}
+
+fn parse_input_sources(i: &[u8]) -> NomResult<&[u8], BTreeSet<String>> {
+    let (i, input_sources_lst) = aterm::parse_str_list(i).map_err(into_nomerror)?;
+
+    let mut input_sources: BTreeSet<_> = BTreeSet::new();
+    for input_source in input_sources_lst.into_iter() {
+        if input_sources.contains(&input_source) {
+            return Err(nom::Err::Failure(NomError {
+                input: i,
+                code: ErrorKind::DuplicateInputSource(input_source),
+            }));
+        } else {
+            input_sources.insert(input_source);
+        }
+    }
+
+    Ok((i, input_sources))
+}
+
+pub fn parse_derivation(i: &[u8]) -> NomResult<&[u8], Derivation> {
+    use nom::Parser;
+    preceded(
+        tag(write::DERIVATION_PREFIX),
+        delimited(
+            // inside parens
+            nomchar('('),
+            // tuple requires all errors to be of the same type, so we need to be a
+            // bit verbose here wrapping generic IResult into [NomATermResult].
+            tuple((
+                // parse outputs
+                terminated(parse_outputs, nomchar(',')),
+                // // parse input derivations
+                terminated(parse_input_derivations, nomchar(',')),
+                // // parse input sources
+                terminated(parse_input_sources, nomchar(',')),
+                // // parse system
+                |i| terminated(aterm::parse_string_field, nomchar(','))(i).map_err(into_nomerror),
+                // // parse builder
+                |i| terminated(aterm::parse_string_field, nomchar(','))(i).map_err(into_nomerror),
+                // // parse arguments
+                |i| terminated(aterm::parse_str_list, nomchar(','))(i).map_err(into_nomerror),
+                // parse environment
+                parse_kv::<BString, _>(aterm::parse_bstr_field),
+            )),
+            nomchar(')'),
+        )
+        .map(
+            |(
+                outputs,
+                input_derivations,
+                input_sources,
+                system,
+                builder,
+                arguments,
+                environment,
+            )| {
+                Derivation {
+                    arguments,
+                    builder,
+                    environment,
+                    input_derivations,
+                    input_sources,
+                    outputs,
+                    system,
+                }
+            },
+        ),
+    )(i)
+}
+
+/// Parse a list of key/value pairs into a BTreeMap.
+/// The parser for the values can be passed in.
+/// In terms of ATerm, this is just a 2-tuple,
+/// but we have the additional restriction that the first element needs to be
+/// unique across all tuples.
+pub(crate) fn parse_kv<'a, V, VF>(
+    vf: VF,
+) -> impl FnMut(&'a [u8]) -> NomResult<&'a [u8], BTreeMap<String, V>> + 'static
+where
+    VF: FnMut(&'a [u8]) -> nom::IResult<&'a [u8], V, nom::error::Error<&'a [u8]>> + Clone + 'static,
+{
+    move |i|
+    // inside brackets
+    delimited(
+        nomchar('['),
+        |ii| {
+            let res = separated_list0(
+                nomchar(','),
+                // inside parens
+                delimited(
+                    nomchar('('),
+                    separated_pair(
+                        aterm::parse_string_field,
+                        nomchar(','),
+                        vf.clone(),
+                    ),
+                    nomchar(')'),
+                ),
+            )(ii).map_err(into_nomerror);
+
+            match res {
+                Ok((rest, pairs)) => {
+                    let mut kvs: BTreeMap<String, V> = BTreeMap::new();
+                    for (k, v) in pairs.into_iter() {
+                        // collect the 2-tuple to a BTreeMap,
+                        // and fail if the key was already seen before.
+                        match kvs.entry(k) {
+                            btree_map::Entry::Vacant(e) => { e.insert(v); },
+                            btree_map::Entry::Occupied(e) => {
+                                return Err(nom::Err::Failure(NomError {
+                                    input: i,
+                                    code: ErrorKind::DuplicateMapKey(e.key().clone()),
+                                }));
+                            }
+                        }
+                    }
+                    Ok((rest, kvs))
+                }
+                Err(e) => Err(e),
+            }
+        },
+        nomchar(']'),
+    )(i)
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::{BTreeMap, BTreeSet};
+
+    use crate::derivation::{
+        parse_error::ErrorKind, parser::from_algo_and_mode_and_digest, CAHash, NixHash, Output,
+    };
+    use bstr::{BString, ByteSlice};
+    use hex_literal::hex;
+    use lazy_static::lazy_static;
+    use test_case::test_case;
+    const DIGEST_SHA256: [u8; 32] =
+        hex!("a5ce9c155ed09397614646c9717fc7cd94b1023d7b76b618d409e4fefd6e9d39");
+
+    lazy_static! {
+        pub static ref NIXHASH_SHA256: NixHash = NixHash::Sha256(DIGEST_SHA256);
+        static ref EXP_MULTI_OUTPUTS: BTreeMap<String, Output> = {
+            let mut b = BTreeMap::new();
+            b.insert(
+                "lib".to_string(),
+                Output {
+                    path: "/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib"
+                        .to_string(),
+                    ca_hash: None,
+                },
+            );
+            b.insert(
+                "out".to_string(),
+                Output {
+                    path: "/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out".to_string(),
+                    ca_hash: None,
+                },
+            );
+            b
+        };
+        static ref EXP_AB_MAP: BTreeMap<String, BString> = {
+            let mut b = BTreeMap::new();
+            b.insert("a".to_string(), b"1".as_bstr().to_owned());
+            b.insert("b".to_string(), b"2".as_bstr().to_owned());
+            b
+        };
+        static ref EXP_INPUT_DERIVATIONS_SIMPLE: BTreeMap<String, BTreeSet<String>> = {
+            let mut b = BTreeMap::new();
+            b.insert(
+                "/nix/store/8bjm87p310sb7r2r0sg4xrynlvg86j8k-hello-2.12.1.tar.gz.drv".to_string(),
+                {
+                    let mut output_names = BTreeSet::new();
+                    output_names.insert("out".to_string());
+                    output_names
+                },
+            );
+            b.insert(
+                "/nix/store/p3jc8aw45dza6h52v81j7lk69khckmcj-bash-5.2-p15.drv".to_string(),
+                {
+                    let mut output_names = BTreeSet::new();
+                    output_names.insert("out".to_string());
+                    output_names.insert("lib".to_string());
+                    output_names
+                },
+            );
+            b
+        };
+        static ref EXP_INPUT_DERIVATIONS_SIMPLE_ATERM: String = {
+            format!(
+                "[(\"{0}\",[\"out\"]),(\"{1}\",[\"out\",\"lib\"])]",
+                "/nix/store/8bjm87p310sb7r2r0sg4xrynlvg86j8k-hello-2.12.1.tar.gz.drv",
+                "/nix/store/p3jc8aw45dza6h52v81j7lk69khckmcj-bash-5.2-p15.drv"
+            )
+        };
+        static ref EXP_INPUT_SOURCES_SIMPLE: BTreeSet<String> = {
+            let mut b = BTreeSet::new();
+            b.insert("/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out".to_string());
+            b.insert("/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib".to_string());
+            b
+        };
+    }
+
+    /// Ensure parsing KVs works
+    #[test_case(b"[]", &BTreeMap::new(), b""; "empty")]
+    #[test_case(b"[(\"a\",\"1\"),(\"b\",\"2\")]", &EXP_AB_MAP, b""; "simple")]
+    fn parse_kv(input: &'static [u8], expected: &BTreeMap<String, BString>, exp_rest: &[u8]) {
+        let (rest, parsed) = super::parse_kv::<BString, _>(crate::aterm::parse_bstr_field)(input)
+            .expect("must parse");
+        assert_eq!(exp_rest, rest, "expected remainder");
+        assert_eq!(*expected, parsed);
+    }
+
+    /// Ensures the kv parser complains about duplicate map keys
+    #[test]
+    fn parse_kv_fail_dup_keys() {
+        let input: &'static [u8] = b"[(\"a\",\"1\"),(\"a\",\"2\")]";
+        let e = super::parse_kv::<BString, _>(crate::aterm::parse_bstr_field)(input)
+            .expect_err("must fail");
+
+        match e {
+            nom::Err::Failure(e) => {
+                assert_eq!(ErrorKind::DuplicateMapKey("a".to_string()), e.code);
+            }
+            _ => panic!("unexpected error"),
+        }
+    }
+
+    /// Ensure parsing input derivations works.
+    #[test_case(b"[]", &BTreeMap::new(); "empty")]
+    #[test_case(EXP_INPUT_DERIVATIONS_SIMPLE_ATERM.as_bytes(), &EXP_INPUT_DERIVATIONS_SIMPLE; "simple")]
+    fn parse_input_derivations(
+        input: &'static [u8],
+        expected: &BTreeMap<String, BTreeSet<String>>,
+    ) {
+        let (rest, parsed) = super::parse_input_derivations(input).expect("must parse");
+
+        assert_eq!(expected, &parsed, "parsed mismatch");
+        assert!(rest.is_empty(), "rest must be empty");
+    }
+
+    /// Ensures the input derivation parser complains about duplicate output names
+    #[test]
+    fn parse_input_derivations_fail_dup_output_names() {
+        let input_str = format!(
+            "[(\"{0}\",[\"out\"]),(\"{1}\",[\"out\",\"out\"])]",
+            "/nix/store/8bjm87p310sb7r2r0sg4xrynlvg86j8k-hello-2.12.1.tar.gz.drv",
+            "/nix/store/p3jc8aw45dza6h52v81j7lk69khckmcj-bash-5.2-p15.drv"
+        );
+        let e = super::parse_input_derivations(input_str.as_bytes()).expect_err("must fail");
+
+        match e {
+            nom::Err::Failure(e) => {
+                assert_eq!(
+                    ErrorKind::DuplicateInputDerivationOutputName(
+                        "/nix/store/p3jc8aw45dza6h52v81j7lk69khckmcj-bash-5.2-p15.drv".to_string(),
+                        "out".to_string()
+                    ),
+                    e.code
+                );
+            }
+            _ => panic!("unexpected error"),
+        }
+    }
+
+    /// Ensure parsing input sources works
+    #[test_case(b"[]", &BTreeSet::new(); "empty")]
+    #[test_case(b"[\"/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out\",\"/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib\"]", &EXP_INPUT_SOURCES_SIMPLE; "simple")]
+    fn parse_input_sources(input: &'static [u8], expected: &BTreeSet<String>) {
+        let (rest, parsed) = super::parse_input_sources(input).expect("must parse");
+
+        assert_eq!(expected, &parsed, "parsed mismatch");
+        assert!(rest.is_empty(), "rest must be empty");
+    }
+
+    /// Ensures the input sources parser complains about duplicate input sources
+    #[test]
+    fn parse_input_sources_fail_dup_keys() {
+        let input: &'static [u8] = b"[\"/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-foo\",\"/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-foo\"]";
+        let e = super::parse_input_sources(input).expect_err("must fail");
+
+        match e {
+            nom::Err::Failure(e) => {
+                assert_eq!(
+                    ErrorKind::DuplicateInputSource(
+                        "/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-foo".to_string()
+                    ),
+                    e.code
+                );
+            }
+            _ => panic!("unexpected error"),
+        }
+    }
+
+    #[test_case(
+        br#"("out","/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo","","")"#,
+        ("out".to_string(), Output {
+            path: "/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo".to_string(),
+            ca_hash: None
+        }); "simple"
+    )]
+    #[test_case(
+        br#"("out","/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar","r:sha256","08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba")"#,
+        ("out".to_string(), Output {
+            path: "/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar".to_string(),
+            ca_hash: Some(from_algo_and_mode_and_digest("r:sha256",
+                   data_encoding::HEXLOWER.decode(b"08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba").unwrap()            ).unwrap()),
+        }); "fod"
+     )]
+    fn parse_output(input: &[u8], expected: (String, Output)) {
+        let (rest, parsed) = super::parse_output(input).expect("must parse");
+        assert!(rest.is_empty());
+        assert_eq!(expected, parsed);
+    }
+
+    #[test_case(
+        br#"[("lib","/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib","",""),("out","/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out","","")]"#,
+        &EXP_MULTI_OUTPUTS;
+        "multi-out"
+    )]
+    fn parse_outputs(input: &[u8], expected: &BTreeMap<String, Output>) {
+        let (rest, parsed) = super::parse_outputs(input).expect("must parse");
+        assert!(rest.is_empty());
+        assert_eq!(*expected, parsed);
+    }
+
+    #[test_case("sha256", &DIGEST_SHA256, CAHash::Flat(NIXHASH_SHA256.clone()); "sha256 flat")]
+    #[test_case("r:sha256", &DIGEST_SHA256, CAHash::Nar(NIXHASH_SHA256.clone()); "sha256 recursive")]
+    fn test_from_algo_and_mode_and_digest(algo_and_mode: &str, digest: &[u8], expected: CAHash) {
+        assert_eq!(
+            expected,
+            from_algo_and_mode_and_digest(algo_and_mode, digest).unwrap()
+        );
+    }
+
+    #[test]
+    fn from_algo_and_mode_and_digest_failure() {
+        assert!(from_algo_and_mode_and_digest("r:sha256", []).is_err());
+        assert!(from_algo_and_mode_and_digest("ha256", DIGEST_SHA256).is_err());
+    }
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/duplicate.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/duplicate.drv
new file mode 100644
index 000000000000..072561a29e3a
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/duplicate.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo","","")],[("/nix/store/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv",["out"])],[],":",":",[],[("bar","/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar"),("builder",":"),("name","foo"),("name","bar"),("out","/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv
new file mode 100644
index 000000000000..a4fea3c5f486
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar","r:sha256","08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba")],[],[],":",":",[],[("builder",":"),("name","bar"),("out","/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar"),("outputHash","08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba"),("outputHashAlgo","sha256"),("outputHashMode","recursive"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv.json
new file mode 100644
index 000000000000..c8bbc4cbb5be
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv.json
@@ -0,0 +1,23 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "builder": ":",
+    "name": "bar",
+    "out": "/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar",
+    "outputHash": "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba",
+    "outputHashAlgo": "sha256",
+    "outputHashMode": "recursive",
+    "system": ":"
+  },
+  "inputDrvs": {},
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "hash": "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba",
+      "hashAlgo": "r:sha256",
+      "path": "/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv
new file mode 100644
index 000000000000..f0d9230a5a52
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/pzr7lsd3q9pqsnb42r9b23jc5sh8irvn-nested-json","","")],[],[],":",":",[],[("builder",":"),("json","{\"hello\":\"moto\\n\"}"),("name","nested-json"),("out","/nix/store/pzr7lsd3q9pqsnb42r9b23jc5sh8irvn-nested-json"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv.json
new file mode 100644
index 000000000000..9cb0b43b4c09
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/292w8yzv5nn7nhdpxcs8b7vby2p27s09-nested-json.drv.json
@@ -0,0 +1,19 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "builder": ":",
+    "json": "{\"hello\":\"moto\\n\"}",
+    "name": "nested-json",
+    "out": "/nix/store/pzr7lsd3q9pqsnb42r9b23jc5sh8irvn-nested-json",
+    "system": ":"
+  },
+  "inputDrvs": {},
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "path": "/nix/store/pzr7lsd3q9pqsnb42r9b23jc5sh8irvn-nested-json"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv
new file mode 100644
index 000000000000..a2cf9d31f92e
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo","","")],[("/nix/store/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv",["out"])],[],":",":",[],[("bar","/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar"),("builder",":"),("name","foo"),("out","/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv.json
new file mode 100644
index 000000000000..957a85ccab82
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv.json
@@ -0,0 +1,23 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "bar": "/nix/store/4q0pg5zpfmznxscq3avycvf9xdvx50n3-bar",
+    "builder": ":",
+    "name": "foo",
+    "out": "/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo",
+    "system": ":"
+  },
+  "inputDrvs": {
+    "/nix/store/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv": [
+      "out"
+    ]
+  },
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "path": "/nix/store/5vyvcwah9l9kf07d52rcgdk70g2f4y13-foo"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv
new file mode 100644
index 000000000000..bbe88c02c739
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/vgvdj6nf7s8kvfbl2skbpwz9kc7xjazc-unicode","","")],[],[],":",":",[],[("builder",":"),("letters","räksmörgås\nrødgrød med fløde\nLübeck\n肥猪\nこんにちは / 今日は\n🌮\n"),("name","unicode"),("out","/nix/store/vgvdj6nf7s8kvfbl2skbpwz9kc7xjazc-unicode"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv.json
new file mode 100644
index 000000000000..f8f33c1bba17
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv.json
@@ -0,0 +1,19 @@
+{
+  "outputs": {
+    "out": {
+      "path": "/nix/store/vgvdj6nf7s8kvfbl2skbpwz9kc7xjazc-unicode"
+    }
+  },
+  "inputSrcs": [],
+  "inputDrvs": {},
+  "system": ":",
+  "builder": ":",
+  "args": [],
+  "env": {
+    "builder": ":",
+    "letters": "räksmörgås\nrødgrød med fløde\nLübeck\n肥猪\nこんにちは / 今日は\n🌮\n",
+    "name": "unicode",
+    "out": "/nix/store/vgvdj6nf7s8kvfbl2skbpwz9kc7xjazc-unicode",
+    "system": ":"
+  }
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv
new file mode 100644
index 000000000000..4b9338c0b953
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/6a39dl014j57bqka7qx25k0vb20vkqm6-structured-attrs","","")],[],[],":",":",[],[("__json","{\"builder\":\":\",\"name\":\"structured-attrs\",\"system\":\":\"}"),("out","/nix/store/6a39dl014j57bqka7qx25k0vb20vkqm6-structured-attrs")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv.json
new file mode 100644
index 000000000000..74e3d7df55c5
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv.json
@@ -0,0 +1,16 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "__json": "{\"builder\":\":\",\"name\":\"structured-attrs\",\"system\":\":\"}",
+    "out": "/nix/store/6a39dl014j57bqka7qx25k0vb20vkqm6-structured-attrs"
+  },
+  "inputDrvs": {},
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "path": "/nix/store/6a39dl014j57bqka7qx25k0vb20vkqm6-structured-attrs"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv
new file mode 100644
index 000000000000..1699c2a75e48
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/fhaj6gmwns62s6ypkcldbaj2ybvkhx3p-foo","","")],[("/nix/store/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv",["out"])],[],":",":",[],[("bar","/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar"),("builder",":"),("name","foo"),("out","/nix/store/fhaj6gmwns62s6ypkcldbaj2ybvkhx3p-foo"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv.json
new file mode 100644
index 000000000000..831d27956d86
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv.json
@@ -0,0 +1,23 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "bar": "/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar",
+    "builder": ":",
+    "name": "foo",
+    "out": "/nix/store/fhaj6gmwns62s6ypkcldbaj2ybvkhx3p-foo",
+    "system": ":"
+  },
+  "inputDrvs": {
+    "/nix/store/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv": [
+      "out"
+    ]
+  },
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "path": "/nix/store/fhaj6gmwns62s6ypkcldbaj2ybvkhx3p-foo"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv
new file mode 100644
index 000000000000..523612238c76
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv
@@ -0,0 +1 @@
+Derive([("lib","/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib","",""),("out","/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out","","")],[],[],":",":",[],[("builder",":"),("lib","/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib"),("name","has-multi-out"),("out","/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out"),("outputs","out lib"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv.json
new file mode 100644
index 000000000000..0bd7a2991cc7
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv.json
@@ -0,0 +1,23 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "builder": ":",
+    "lib": "/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib",
+    "name": "has-multi-out",
+    "out": "/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out",
+    "outputs": "out lib",
+    "system": ":"
+  },
+  "inputDrvs": {},
+  "inputSrcs": [],
+  "outputs": {
+    "lib": {
+      "path": "/nix/store/2vixb94v0hy2xc6p7mbnxxcyc095yyia-has-multi-out-lib"
+    },
+    "out": {
+      "path": "/nix/store/55lwldka5nyxa08wnvlizyqw02ihy8ic-has-multi-out"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv
new file mode 100644
index 000000000000..6a7a35c58c3f
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/drr2mjp9fp9vvzsf5f9p0a80j33dxy7m-cp1252","","")],[],[],":",":",[],[("builder",":"),("chars",""),("name","cp1252"),("out","/nix/store/drr2mjp9fp9vvzsf5f9p0a80j33dxy7m-cp1252"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv.json
new file mode 100644
index 000000000000..9d6ba8b7977f
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv.json
@@ -0,0 +1,21 @@
+{
+  "/nix/store/m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv": {
+    "outputs": {
+      "out": {
+        "path": "/nix/store/drr2mjp9fp9vvzsf5f9p0a80j33dxy7m-cp1252"
+      }
+    },
+    "inputSrcs": [],
+    "inputDrvs": {},
+    "system": ":",
+    "builder": ":",
+    "args": [],
+    "env": {
+      "builder": ":",
+      "chars": "",
+      "name": "cp1252",
+      "out": "/nix/store/drr2mjp9fp9vvzsf5f9p0a80j33dxy7m-cp1252",
+      "system": ":"
+    }
+  }
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv
new file mode 100644
index 000000000000..559e93ed0ed6
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar","r:sha1","0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33")],[],[],":",":",[],[("builder",":"),("name","bar"),("out","/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar"),("outputHash","0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33"),("outputHashAlgo","sha1"),("outputHashMode","recursive"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv.json
new file mode 100644
index 000000000000..e297d271592f
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv.json
@@ -0,0 +1,23 @@
+{
+  "args": [],
+  "builder": ":",
+  "env": {
+    "builder": ":",
+    "name": "bar",
+    "out": "/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar",
+    "outputHash": "0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33",
+    "outputHashAlgo": "sha1",
+    "outputHashMode": "recursive",
+    "system": ":"
+  },
+  "inputDrvs": {},
+  "inputSrcs": [],
+  "outputs": {
+    "out": {
+      "hash": "0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33",
+      "hashAlgo": "r:sha1",
+      "path": "/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar"
+    }
+  },
+  "system": ":"
+}
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv
new file mode 100644
index 000000000000..b19fd8eb2ce4
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv
@@ -0,0 +1 @@
+Derive([("out","/nix/store/x1f6jfq9qgb6i8jrmpifkn9c64fg4hcm-latin1","","")],[],[],":",":",[],[("builder",":"),("chars",""),("name","latin1"),("out","/nix/store/x1f6jfq9qgb6i8jrmpifkn9c64fg4hcm-latin1"),("system",":")])
\ No newline at end of file
diff --git a/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv.json b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv.json
new file mode 100644
index 000000000000..ffd5c08da830
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/derivation_tests/ok/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv.json
@@ -0,0 +1,21 @@
+{
+  "/nix/store/x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv": {
+    "outputs": {
+      "out": {
+        "path": "/nix/store/x1f6jfq9qgb6i8jrmpifkn9c64fg4hcm-latin1"
+      }
+    },
+    "inputSrcs": [],
+    "inputDrvs": {},
+    "system": ":",
+    "builder": ":",
+    "args": [],
+    "env": {
+      "builder": ":",
+      "chars": "",
+      "name": "latin1",
+      "out": "/nix/store/x1f6jfq9qgb6i8jrmpifkn9c64fg4hcm-latin1",
+      "system": ":"
+    }
+  }
+}
diff --git a/tvix/nix-compat/src/derivation/tests/mod.rs b/tvix/nix-compat/src/derivation/tests/mod.rs
new file mode 100644
index 000000000000..168e11d46f1e
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/tests/mod.rs
@@ -0,0 +1,416 @@
+use super::parse_error::ErrorKind;
+use crate::derivation::output::Output;
+use crate::derivation::parse_error::NomError;
+use crate::derivation::parser::Error;
+use crate::derivation::Derivation;
+use crate::store_path::StorePath;
+use bstr::{BStr, BString};
+use std::collections::BTreeSet;
+use std::fs::File;
+use std::io::Read;
+use std::path::Path;
+use std::str::FromStr;
+use test_case::test_case;
+use test_generator::test_resources;
+
+const RESOURCES_PATHS: &str = "src/derivation/tests/derivation_tests";
+
+fn read_file(path: &str) -> BString {
+    let path = Path::new(path);
+    let mut file = File::open(path).unwrap();
+    let mut file_contents = Vec::new();
+
+    file.read_to_end(&mut file_contents).unwrap();
+
+    file_contents.into()
+}
+
+#[test_resources("src/derivation/tests/derivation_tests/ok/*.drv")]
+fn check_serialization(path_to_drv_file: &str) {
+    // skip JSON files known to fail parsing
+    if path_to_drv_file.ends_with("cp1252.drv") || path_to_drv_file.ends_with("latin1.drv") {
+        return;
+    }
+    let json_bytes = read_file(&format!("{}.json", path_to_drv_file));
+    let derivation: Derivation =
+        serde_json::from_slice(&json_bytes).expect("JSON was not well-formatted");
+
+    let mut serialized_derivation = Vec::new();
+    derivation.serialize(&mut serialized_derivation).unwrap();
+
+    let expected = read_file(path_to_drv_file);
+
+    assert_eq!(expected, BStr::new(&serialized_derivation));
+}
+
+#[test_resources("src/derivation/tests/derivation_tests/ok/*.drv")]
+fn validate(path_to_drv_file: &str) {
+    // skip JSON files known to fail parsing
+    if path_to_drv_file.ends_with("cp1252.drv") || path_to_drv_file.ends_with("latin1.drv") {
+        return;
+    }
+    let json_bytes = read_file(&format!("{}.json", path_to_drv_file));
+    let derivation: Derivation =
+        serde_json::from_slice(&json_bytes).expect("JSON was not well-formatted");
+
+    derivation
+        .validate(true)
+        .expect("derivation failed to validate")
+}
+
+#[test_resources("src/derivation/tests/derivation_tests/ok/*.drv")]
+fn check_to_aterm_bytes(path_to_drv_file: &str) {
+    // skip JSON files known to fail parsing
+    if path_to_drv_file.ends_with("cp1252.drv") || path_to_drv_file.ends_with("latin1.drv") {
+        return;
+    }
+    let json_bytes = read_file(&format!("{}.json", path_to_drv_file));
+    let derivation: Derivation =
+        serde_json::from_slice(&json_bytes).expect("JSON was not well-formatted");
+
+    let expected = read_file(path_to_drv_file);
+
+    assert_eq!(expected, BStr::new(&derivation.to_aterm_bytes()));
+}
+
+/// Reads in derivations in ATerm representation, parses with that parser,
+/// then compares the structs with the ones obtained by parsing the JSON
+/// representations.
+#[test_resources("src/derivation/tests/derivation_tests/ok/*.drv")]
+fn from_aterm_bytes(path_to_drv_file: &str) {
+    // Read in ATerm representation.
+    let aterm_bytes = read_file(path_to_drv_file);
+    let parsed_drv = Derivation::from_aterm_bytes(&aterm_bytes).expect("must succeed");
+
+    // For where we're able to load JSON fixtures, parse them and compare the structs.
+    // For where we're not, compare the bytes manually.
+    if path_to_drv_file.ends_with("cp1252.drv") || path_to_drv_file.ends_with("latin1.drv") {
+        assert_eq!(
+            &[0xc5, 0xc4, 0xd6][..],
+            parsed_drv.environment.get("chars").unwrap(),
+            "expected bytes to match",
+        );
+    } else {
+        let json_bytes = read_file(&format!("{}.json", path_to_drv_file));
+        let fixture_derivation: Derivation =
+            serde_json::from_slice(&json_bytes).expect("JSON was not well-formatted");
+
+        assert_eq!(fixture_derivation, parsed_drv);
+    }
+
+    // Finally, write the ATerm serialization to another buffer, ensuring it's
+    // stable (and we compare all fields we couldn't compare in the non-utf8
+    // derivations)
+
+    assert_eq!(
+        &aterm_bytes,
+        &parsed_drv.to_aterm_bytes(),
+        "expected serialized ATerm to match initial input"
+    );
+}
+
+#[test]
+fn from_aterm_bytes_duplicate_map_key() {
+    let buf: Vec<u8> = read_file(&format!("{}/{}", RESOURCES_PATHS, "duplicate.drv")).into();
+
+    let err = Derivation::from_aterm_bytes(&buf).expect_err("must fail");
+
+    match err {
+        Error::Parser(NomError { input: _, code }) => {
+            assert_eq!(code, ErrorKind::DuplicateMapKey("name".to_string()));
+        }
+        _ => {
+            panic!("unexpected error");
+        }
+    }
+}
+
+/// Read in a derivation in ATerm, but add some garbage at the end.
+/// Ensure the parser detects and fails in this case.
+#[test]
+fn from_aterm_bytes_trailer() {
+    let mut buf: Vec<u8> = read_file(&format!(
+        "{}/ok/{}",
+        RESOURCES_PATHS, "0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv"
+    ))
+    .into();
+
+    buf.push(0x00);
+
+    Derivation::from_aterm_bytes(&buf).expect_err("must fail");
+}
+
+#[test_case("bar","0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv"; "fixed_sha256")]
+#[test_case("foo", "4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv"; "simple-sha256")]
+#[test_case("bar", "ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv"; "fixed-sha1")]
+#[test_case("foo", "ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv"; "simple-sha1")]
+#[test_case("has-multi-out", "h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv"; "multiple-outputs")]
+#[test_case("structured-attrs", "9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv"; "structured-attrs")]
+#[test_case("unicode", "52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv"; "unicode")]
+fn derivation_path(name: &str, expected_path: &str) {
+    let json_bytes = read_file(&format!("{}/ok/{}.json", RESOURCES_PATHS, expected_path));
+    let derivation: Derivation =
+        serde_json::from_slice(&json_bytes).expect("JSON was not well-formatted");
+
+    assert_eq!(
+        derivation.calculate_derivation_path(name).unwrap(),
+        StorePath::from_str(expected_path).unwrap()
+    );
+}
+
+/// This trims all output paths from a Derivation struct,
+/// by setting outputs[$outputName].path and environment[$outputName] to the empty string.
+fn derivation_with_trimmed_output_paths(derivation: &Derivation) -> Derivation {
+    let mut trimmed_env = derivation.environment.clone();
+    let mut trimmed_outputs = derivation.outputs.clone();
+
+    for (output_name, output) in &derivation.outputs {
+        trimmed_env.insert(output_name.clone(), "".into());
+        assert!(trimmed_outputs.contains_key(output_name));
+        trimmed_outputs.insert(
+            output_name.to_string(),
+            Output {
+                path: "".to_string(),
+                ..output.clone()
+            },
+        );
+    }
+
+    // replace environment and outputs with the trimmed variants
+    Derivation {
+        environment: trimmed_env,
+        outputs: trimmed_outputs,
+        ..derivation.clone()
+    }
+}
+
+#[test_case("0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv", "sha256:724f3e3634fce4cbbbd3483287b8798588e80280660b9a63fd13a1bc90485b33"; "fixed_sha256")]
+#[test_case("ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv", "sha256:c79aebd0ce3269393d4a1fde2cbd1d975d879b40f0bf40a48f550edc107fd5df";"fixed-sha1")]
+fn derivation_or_fod_hash(drv_path: &str, expected_nix_hash_string: &str) {
+    // read in the fixture
+    let json_bytes = read_file(&format!("{}/ok/{}.json", RESOURCES_PATHS, drv_path));
+    let drv: Derivation = serde_json::from_slice(&json_bytes).expect("must deserialize");
+
+    let actual = drv.derivation_or_fod_hash(|_| panic!("must not be called"));
+
+    assert_eq!(expected_nix_hash_string, actual.to_nix_hex_string());
+}
+
+/// This reads a Derivation (in A-Term), trims out all fields containing
+/// calculated output paths, then triggers the output path calculation and
+/// compares the struct to match what was originally read in.
+#[test_case("bar","0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv"; "fixed_sha256")]
+#[test_case("foo", "4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv"; "simple-sha256")]
+#[test_case("bar", "ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv"; "fixed-sha1")]
+#[test_case("foo", "ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv"; "simple-sha1")]
+#[test_case("has-multi-out", "h32dahq0bx5rp1krcdx3a53asj21jvhk-has-multi-out.drv"; "multiple-outputs")]
+#[test_case("structured-attrs", "9lj1lkjm2ag622mh4h9rpy6j607an8g2-structured-attrs.drv"; "structured-attrs")]
+#[test_case("unicode", "52a9id8hx688hvlnz4d1n25ml1jdykz0-unicode.drv"; "unicode")]
+#[test_case("cp1252", "m1vfixn8iprlf0v9abmlrz7mjw1xj8kp-cp1252.drv"; "cp1252")]
+#[test_case("latin1", "x6p0hg79i3wg0kkv7699935f7rrj9jf3-latin1.drv"; "latin1")]
+fn output_paths(name: &str, drv_path_str: &str) {
+    // read in the derivation
+    let expected_derivation = Derivation::from_aterm_bytes(
+        read_file(&format!("{}/ok/{}", RESOURCES_PATHS, drv_path_str)).as_ref(),
+    )
+    .expect("must succeed");
+
+    // create a version with trimmed output paths, simulating we constructed
+    // the struct.
+    let mut derivation = derivation_with_trimmed_output_paths(&expected_derivation);
+
+    // calculate the derivation_or_fod_hash of derivation
+    // We don't expect the lookup function to be called for most derivations.
+    let calculated_derivation_or_fod_hash = derivation.derivation_or_fod_hash(|parent_drv_path| {
+        // 4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv may lookup /nix/store/0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv
+        // ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv may lookup /nix/store/ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv
+        if name == "foo"
+            && ((drv_path_str == "4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv"
+                && parent_drv_path.to_string() == "0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv")
+                || (drv_path_str == "ch49594n9avinrf8ip0aslidkc4lxkqv-foo.drv"
+                    && parent_drv_path.to_string() == "ss2p4wmxijn652haqyd7dckxwl4c7hxx-bar.drv"))
+        {
+            // do the lookup, by reading in the fixture of the requested
+            // drv_name, and calculating its drv replacement (on the non-stripped version)
+            // In a real-world scenario you would have already done this during construction.
+
+            let json_bytes = read_file(&format!(
+                "{}/ok/{}.json",
+                RESOURCES_PATHS,
+                Path::new(&parent_drv_path.to_string())
+                    .file_name()
+                    .unwrap()
+                    .to_string_lossy()
+            ));
+
+            let drv: Derivation = serde_json::from_slice(&json_bytes).expect("must deserialize");
+
+            // calculate derivation_or_fod_hash for each parent.
+            // This may not trigger subsequent requests, as both parents are FOD.
+            drv.derivation_or_fod_hash(|_| panic!("must not lookup"))
+        } else {
+            // we only expect this to be called in the "foo" testcase, for the "bar derivations"
+            panic!("may only be called for foo testcase on bar derivations");
+        }
+    });
+
+    derivation
+        .calculate_output_paths(name, &calculated_derivation_or_fod_hash)
+        .unwrap();
+
+    // The derivation should now look like it was before
+    assert_eq!(expected_derivation, derivation);
+}
+
+/// Exercises the output path calculation functions like a constructing client
+/// (an implementation of builtins.derivation) would do:
+///
+/// ```nix
+/// rec {
+///   bar = builtins.derivation {
+///     name = "bar";
+///     builder = ":";
+///     system = ":";
+///     outputHash = "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba";
+///     outputHashAlgo = "sha256";
+///     outputHashMode = "recursive";
+///   };
+///
+///   foo = builtins.derivation {
+///     name = "foo";
+///     builder = ":";
+///     system = ":";
+///     inherit bar;
+///   };
+/// }
+/// ```
+/// It first assembles the bar derivation, does the output path calculation on
+/// it, then continues with the foo derivation.
+///
+/// The code ensures the resulting Derivations match our fixtures.
+#[test]
+fn output_path_construction() {
+    // create the bar derivation
+    let mut bar_drv = Derivation {
+        builder: ":".to_string(),
+        system: ":".to_string(),
+        ..Default::default()
+    };
+
+    // assemble bar env
+    let bar_env = &mut bar_drv.environment;
+    bar_env.insert("builder".to_string(), ":".into());
+    bar_env.insert("name".to_string(), "bar".into());
+    bar_env.insert("out".to_string(), "".into()); // will be calculated
+    bar_env.insert(
+        "outputHash".to_string(),
+        "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba".into(),
+    );
+    bar_env.insert("outputHashAlgo".to_string(), "sha256".into());
+    bar_env.insert("outputHashMode".to_string(), "recursive".into());
+    bar_env.insert("system".to_string(), ":".into());
+
+    // assemble bar outputs
+    bar_drv.outputs.insert(
+        "out".to_string(),
+        Output {
+            path: "".to_string(), // will be calculated
+            ca_hash: Some(crate::nixhash::CAHash::Nar(
+                crate::nixhash::from_algo_and_digest(
+                    crate::nixhash::HashAlgo::Sha256,
+                    &data_encoding::HEXLOWER
+                        .decode(
+                            "08813cbee9903c62be4c5027726a418a300da4500b2d369d3af9286f4815ceba"
+                                .as_bytes(),
+                        )
+                        .unwrap(),
+                )
+                .unwrap(),
+            )),
+        },
+    );
+
+    // calculate bar output paths
+    let bar_calc_result = bar_drv.calculate_output_paths(
+        "bar",
+        &bar_drv.derivation_or_fod_hash(|_| panic!("is FOD, should not lookup")),
+    );
+    assert!(bar_calc_result.is_ok());
+
+    // ensure it matches our bar fixture
+    let bar_json_bytes = read_file(&format!(
+        "{}/ok/{}.json",
+        RESOURCES_PATHS, "0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv"
+    ));
+    let bar_drv_expected: Derivation =
+        serde_json::from_slice(&bar_json_bytes).expect("must deserialize");
+    assert_eq!(bar_drv_expected, bar_drv);
+
+    // now construct foo, which requires bar_drv
+    // Note how we refer to the output path, drv name and replacement_str (with calculated output paths) of bar.
+    let bar_output_path = &bar_drv.outputs.get("out").expect("must exist").path;
+    let bar_drv_derivation_or_fod_hash =
+        bar_drv.derivation_or_fod_hash(|_| panic!("is FOD, should not lookup"));
+
+    let bar_drv_path = bar_drv
+        .calculate_derivation_path("bar")
+        .expect("must succeed");
+
+    // create foo derivation
+    let mut foo_drv = Derivation {
+        builder: ":".to_string(),
+        system: ":".to_string(),
+        ..Default::default()
+    };
+
+    // assemble foo env
+    let foo_env = &mut foo_drv.environment;
+    foo_env.insert("bar".to_string(), bar_output_path.to_owned().into());
+    foo_env.insert("builder".to_string(), ":".into());
+    foo_env.insert("name".to_string(), "foo".into());
+    foo_env.insert("out".to_string(), "".into()); // will be calculated
+    foo_env.insert("system".to_string(), ":".into());
+
+    // asssemble foo outputs
+    foo_drv.outputs.insert(
+        "out".to_string(),
+        Output {
+            path: "".to_string(), // will be calculated
+            ca_hash: None,
+        },
+    );
+
+    // assemble foo input_derivations
+    foo_drv.input_derivations.insert(
+        bar_drv_path.to_absolute_path(),
+        BTreeSet::from(["out".to_string()]),
+    );
+
+    // calculate foo output paths
+    let foo_calc_result = foo_drv.calculate_output_paths(
+        "foo",
+        &foo_drv.derivation_or_fod_hash(|drv_path| {
+            if drv_path.to_string() != "0hm2f1psjpcwg8fijsmr4wwxrx59s092-bar.drv" {
+                panic!("lookup called with unexpected drv_path: {}", drv_path);
+            }
+            bar_drv_derivation_or_fod_hash.clone()
+        }),
+    );
+    assert!(foo_calc_result.is_ok());
+
+    // ensure it matches our foo fixture
+    let foo_json_bytes = read_file(&format!(
+        "{}/ok/{}.json",
+        RESOURCES_PATHS, "4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv",
+    ));
+    let foo_drv_expected: Derivation =
+        serde_json::from_slice(&foo_json_bytes).expect("must deserialize");
+    assert_eq!(foo_drv_expected, foo_drv);
+
+    assert_eq!(
+        StorePath::from_str("4wvvbi4jwn0prsdxb7vs673qa5h9gr7x-foo.drv").expect("must succeed"),
+        foo_drv
+            .calculate_derivation_path("foo")
+            .expect("must succeed")
+    );
+}
diff --git a/tvix/nix-compat/src/derivation/validate.rs b/tvix/nix-compat/src/derivation/validate.rs
new file mode 100644
index 000000000000..d711f5ce1de2
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/validate.rs
@@ -0,0 +1,158 @@
+use crate::derivation::{Derivation, DerivationError};
+use crate::store_path::{self, StorePathRef};
+
+impl Derivation {
+    /// validate ensures a Derivation struct is properly populated,
+    /// and returns a [DerivationError] if not.
+    ///
+    /// if `validate_output_paths` is set to false, the output paths are
+    /// excluded from validation.
+    ///
+    /// This is helpful to validate struct population before invoking
+    /// [Derivation::calculate_output_paths].
+    pub fn validate(&self, validate_output_paths: bool) -> Result<(), DerivationError> {
+        // Ensure the number of outputs is > 1
+        if self.outputs.is_empty() {
+            return Err(DerivationError::NoOutputs());
+        }
+
+        // Validate all outputs
+        for (output_name, output) in &self.outputs {
+            // empty output names are invalid.
+            //
+            // `drv` is an invalid output name too, as this would cause
+            // a `builtins.derivation` call to return an attrset with a
+            // `drvPath` key (which already exists) and has a different
+            // meaning.
+            //
+            // Other output names that don't match the name restrictions from
+            // [StorePathRef] will fail the [StorePathRef::validate_name] check.
+            if output_name.is_empty()
+                || output_name == "drv"
+                || store_path::validate_name(output_name.as_bytes()).is_err()
+            {
+                return Err(DerivationError::InvalidOutputName(output_name.to_string()));
+            }
+
+            if output.is_fixed() {
+                if self.outputs.len() != 1 {
+                    return Err(DerivationError::MoreThanOneOutputButFixed());
+                }
+                if output_name != "out" {
+                    return Err(DerivationError::InvalidOutputNameForFixed(
+                        output_name.to_string(),
+                    ));
+                }
+            }
+
+            if let Err(e) = output.validate(validate_output_paths) {
+                return Err(DerivationError::InvalidOutput(output_name.to_string(), e));
+            }
+        }
+
+        // Validate all input_derivations
+        for (input_derivation_path, output_names) in &self.input_derivations {
+            // Validate input_derivation_path
+            if let Err(e) = StorePathRef::from_absolute_path(input_derivation_path.as_bytes()) {
+                return Err(DerivationError::InvalidInputDerivationPath(
+                    input_derivation_path.to_string(),
+                    e,
+                ));
+            }
+
+            if !input_derivation_path.ends_with(".drv") {
+                return Err(DerivationError::InvalidInputDerivationPrefix(
+                    input_derivation_path.to_string(),
+                ));
+            }
+
+            if output_names.is_empty() {
+                return Err(DerivationError::EmptyInputDerivationOutputNames(
+                    input_derivation_path.to_string(),
+                ));
+            }
+
+            for output_name in output_names.iter() {
+                // empty output names are invalid.
+                //
+                // `drv` is an invalid output name too, as this would cause
+                // a `builtins.derivation` call to return an attrset with a
+                // `drvPath` key (which already exists) and has a different
+                // meaning.
+                //
+                // Other output names that don't match the name restrictions from
+                // [StorePath] will fail the [StorePathRef::validate_name] check.
+                if output_name.is_empty()
+                    || output_name == "drv"
+                    || store_path::validate_name(output_name.as_bytes()).is_err()
+                {
+                    return Err(DerivationError::InvalidInputDerivationOutputName(
+                        input_derivation_path.to_string(),
+                        output_name.to_string(),
+                    ));
+                }
+            }
+        }
+
+        // Validate all input_sources
+        for input_source in self.input_sources.iter() {
+            if let Err(e) = StorePathRef::from_absolute_path(input_source.as_bytes()) {
+                return Err(DerivationError::InvalidInputSourcesPath(
+                    input_source.to_string(),
+                    e,
+                ));
+            }
+        }
+
+        // validate platform
+        if self.system.is_empty() {
+            return Err(DerivationError::InvalidPlatform(self.system.to_string()));
+        }
+
+        // validate builder
+        if self.builder.is_empty() {
+            return Err(DerivationError::InvalidBuilder(self.builder.to_string()));
+        }
+
+        // validate env, none of the keys may be empty.
+        // We skip the `name` validation seen in go-nix.
+        for k in self.environment.keys() {
+            if k.is_empty() {
+                return Err(DerivationError::InvalidEnvironmentKey(k.to_string()));
+            }
+        }
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use std::collections::BTreeMap;
+
+    use crate::derivation::{CAHash, Derivation, Output};
+
+    /// Regression test: produce a Derivation that's almost valid, except its
+    /// fixed-output output has the wrong hash specified.
+    #[test]
+    fn output_validate() {
+        let mut outputs = BTreeMap::new();
+        outputs.insert(
+            "out".to_string(),
+            Output {
+                path: "".to_string(),
+                ca_hash: Some(CAHash::Text([0; 32])), // This is disallowed
+            },
+        );
+
+        let drv = Derivation {
+            arguments: vec![],
+            builder: "/bin/sh".to_string(),
+            outputs,
+            system: "x86_64-linux".to_string(),
+            ..Default::default()
+        };
+
+        drv.validate(false).expect_err("must fail");
+    }
+}
diff --git a/tvix/nix-compat/src/derivation/write.rs b/tvix/nix-compat/src/derivation/write.rs
new file mode 100644
index 000000000000..0981dfccae22
--- /dev/null
+++ b/tvix/nix-compat/src/derivation/write.rs
@@ -0,0 +1,196 @@
+//! This module implements the serialisation of derivations into the
+//! [ATerm][] format used by C++ Nix.
+//!
+//! [ATerm]: http://program-transformation.org/Tools/ATermFormat.html
+
+use crate::aterm::escape_bytes;
+use crate::derivation::{ca_kind_prefix, output::Output};
+use bstr::BString;
+use std::{
+    collections::{BTreeMap, BTreeSet},
+    io,
+    io::Error,
+    io::Write,
+};
+
+pub const DERIVATION_PREFIX: &str = "Derive";
+pub const PAREN_OPEN: char = '(';
+pub const PAREN_CLOSE: char = ')';
+pub const BRACKET_OPEN: char = '[';
+pub const BRACKET_CLOSE: char = ']';
+pub const COMMA: char = ',';
+pub const QUOTE: char = '"';
+
+// Writes a character to the writer.
+pub(crate) fn write_char(writer: &mut impl Write, c: char) -> io::Result<()> {
+    let mut buf = [0; 4];
+    let b = c.encode_utf8(&mut buf).as_bytes();
+    writer.write_all(b)
+}
+
+// Write a string `s` as a quoted field to the writer.
+// The `escape` argument controls whether escaping will be skipped.
+// This is the case if `s` is known to only contain characters that need no
+// escaping.
+pub(crate) fn write_field<S: AsRef<[u8]>>(
+    writer: &mut impl Write,
+    s: S,
+    escape: bool,
+) -> io::Result<()> {
+    write_char(writer, QUOTE)?;
+
+    if !escape {
+        writer.write_all(s.as_ref())?;
+    } else {
+        writer.write_all(&escape_bytes(s.as_ref()))?;
+    }
+
+    write_char(writer, QUOTE)?;
+
+    Ok(())
+}
+
+fn write_array_elements<S: AsRef<[u8]>>(
+    writer: &mut impl Write,
+    elements: &[S],
+) -> Result<(), io::Error> {
+    for (index, element) in elements.iter().enumerate() {
+        if index > 0 {
+            write_char(writer, COMMA)?;
+        }
+
+        write_field(writer, element, true)?;
+    }
+
+    Ok(())
+}
+
+pub fn write_outputs(
+    writer: &mut impl Write,
+    outputs: &BTreeMap<String, Output>,
+) -> Result<(), io::Error> {
+    write_char(writer, BRACKET_OPEN)?;
+    for (ii, (output_name, output)) in outputs.iter().enumerate() {
+        if ii > 0 {
+            write_char(writer, COMMA)?;
+        }
+
+        write_char(writer, PAREN_OPEN)?;
+
+        let mut elements: Vec<&str> = vec![output_name, &output.path];
+
+        let (mode_and_algo, digest) = match &output.ca_hash {
+            Some(ca_hash) => (
+                format!("{}{}", ca_kind_prefix(ca_hash), ca_hash.digest().algo()),
+                data_encoding::HEXLOWER.encode(ca_hash.digest().digest_as_bytes()),
+            ),
+            None => ("".to_string(), "".to_string()),
+        };
+
+        elements.push(&mode_and_algo);
+        elements.push(&digest);
+
+        write_array_elements(writer, &elements)?;
+
+        write_char(writer, PAREN_CLOSE)?;
+    }
+    write_char(writer, BRACKET_CLOSE)?;
+
+    Ok(())
+}
+
+pub fn write_input_derivations(
+    writer: &mut impl Write,
+    input_derivations: &BTreeMap<String, BTreeSet<String>>,
+) -> Result<(), io::Error> {
+    write_char(writer, BRACKET_OPEN)?;
+
+    for (ii, (input_derivation, output_names)) in input_derivations.iter().enumerate() {
+        if ii > 0 {
+            write_char(writer, COMMA)?;
+        }
+
+        write_char(writer, PAREN_OPEN)?;
+        write_field(writer, input_derivation.as_str(), false)?;
+        write_char(writer, COMMA)?;
+
+        write_char(writer, BRACKET_OPEN)?;
+        write_array_elements(
+            writer,
+            &output_names
+                .iter()
+                .map(String::as_bytes)
+                .collect::<Vec<_>>(),
+        )?;
+        write_char(writer, BRACKET_CLOSE)?;
+
+        write_char(writer, PAREN_CLOSE)?;
+    }
+
+    write_char(writer, BRACKET_CLOSE)?;
+
+    Ok(())
+}
+
+pub fn write_input_sources(
+    writer: &mut impl Write,
+    input_sources: &BTreeSet<String>,
+) -> Result<(), io::Error> {
+    write_char(writer, BRACKET_OPEN)?;
+    write_array_elements(
+        writer,
+        &input_sources.iter().map(String::from).collect::<Vec<_>>(),
+    )?;
+    write_char(writer, BRACKET_CLOSE)?;
+
+    Ok(())
+}
+
+pub fn write_system(writer: &mut impl Write, platform: &str) -> Result<(), Error> {
+    write_field(writer, platform, true)?;
+    Ok(())
+}
+
+pub fn write_builder(writer: &mut impl Write, builder: &str) -> Result<(), Error> {
+    write_field(writer, builder, true)?;
+    Ok(())
+}
+
+pub fn write_arguments(writer: &mut impl Write, arguments: &[String]) -> Result<(), io::Error> {
+    write_char(writer, BRACKET_OPEN)?;
+    write_array_elements(
+        writer,
+        &arguments
+            .iter()
+            .map(|s| s.as_bytes().to_vec().into())
+            .collect::<Vec<BString>>(),
+    )?;
+    write_char(writer, BRACKET_CLOSE)?;
+
+    Ok(())
+}
+
+pub fn write_environment<E, K, V>(writer: &mut impl Write, environment: E) -> Result<(), io::Error>
+where
+    E: IntoIterator<Item = (K, V)>,
+    K: AsRef<[u8]>,
+    V: AsRef<[u8]>,
+{
+    write_char(writer, BRACKET_OPEN)?;
+
+    for (i, (k, v)) in environment.into_iter().enumerate() {
+        if i > 0 {
+            write_char(writer, COMMA)?;
+        }
+
+        write_char(writer, PAREN_OPEN)?;
+        write_field(writer, k, false)?;
+        write_char(writer, COMMA)?;
+        write_field(writer, v, true)?;
+        write_char(writer, PAREN_CLOSE)?;
+    }
+
+    write_char(writer, BRACKET_CLOSE)?;
+
+    Ok(())
+}
diff --git a/tvix/nix-compat/src/lib.rs b/tvix/nix-compat/src/lib.rs
new file mode 100644
index 000000000000..dd161cc1f944
--- /dev/null
+++ b/tvix/nix-compat/src/lib.rs
@@ -0,0 +1,7 @@
+pub(crate) mod aterm;
+pub mod derivation;
+pub mod nar;
+pub mod narinfo;
+pub mod nixbase32;
+pub mod nixhash;
+pub mod store_path;
diff --git a/tvix/nix-compat/src/nar/mod.rs b/tvix/nix-compat/src/nar/mod.rs
new file mode 100644
index 000000000000..058977f4fcd1
--- /dev/null
+++ b/tvix/nix-compat/src/nar/mod.rs
@@ -0,0 +1,4 @@
+mod wire;
+
+pub mod reader;
+pub mod writer;
diff --git a/tvix/nix-compat/src/nar/reader/mod.rs b/tvix/nix-compat/src/nar/reader/mod.rs
new file mode 100644
index 000000000000..fa7ddc77f96c
--- /dev/null
+++ b/tvix/nix-compat/src/nar/reader/mod.rs
@@ -0,0 +1,293 @@
+//! Parser for the Nix archive format, aka NAR.
+//!
+//! NAR files (and their hashed representations) are used in C++ Nix for
+//! a variety of things, including addressing fixed-output derivations
+//! and transferring store paths between Nix stores.
+
+use std::io::{
+    self, BufRead,
+    ErrorKind::{InvalidData, UnexpectedEof},
+    Read, Write,
+};
+
+// Required reading for understanding this module.
+use crate::nar::wire;
+
+mod read;
+#[cfg(test)]
+mod test;
+
+pub type Reader<'a> = dyn BufRead + Send + 'a;
+
+/// Start reading a NAR file from `reader`.
+pub fn open<'a, 'r>(reader: &'a mut Reader<'r>) -> io::Result<Node<'a, 'r>> {
+    read::token(reader, &wire::TOK_NAR)?;
+    Node::new(reader)
+}
+
+pub enum Node<'a, 'r> {
+    Symlink {
+        target: Vec<u8>,
+    },
+    File {
+        executable: bool,
+        reader: FileReader<'a, 'r>,
+    },
+    Directory(DirReader<'a, 'r>),
+}
+
+impl<'a, 'r> Node<'a, 'r> {
+    /// Start reading a [Node], matching the next [wire::Node].
+    ///
+    /// Reading the terminating [wire::TOK_PAR] is done immediately for [Node::Symlink],
+    /// but is otherwise left to [DirReader] or [FileReader].
+    fn new(reader: &'a mut Reader<'r>) -> io::Result<Self> {
+        Ok(match read::tag(reader)? {
+            wire::Node::Sym => {
+                let target = read::bytes(reader, wire::MAX_TARGET_LEN)?;
+
+                if target.is_empty() || target.contains(&0) {
+                    return Err(InvalidData.into());
+                }
+
+                read::token(reader, &wire::TOK_PAR)?;
+
+                Node::Symlink { target }
+            }
+            tag @ (wire::Node::Reg | wire::Node::Exe) => {
+                let len = read::u64(reader)?;
+
+                Node::File {
+                    executable: tag == wire::Node::Exe,
+                    reader: FileReader::new(reader, len)?,
+                }
+            }
+            wire::Node::Dir => Node::Directory(DirReader::new(reader)),
+        })
+    }
+}
+
+/// File contents, readable through the [Read] trait.
+///
+/// It comes with some caveats:
+///  * You must always read the entire file, unless you intend to abandon the entire archive reader.
+///  * You must abandon the entire archive reader upon the first error.
+///
+/// It's fine to read exactly `reader.len()` bytes without ever seeing an explicit EOF.
+///
+/// TODO(edef): enforce these in `#[cfg(debug_assertions)]`
+pub struct FileReader<'a, 'r> {
+    reader: &'a mut Reader<'r>,
+    len: u64,
+    /// Truncated original file length for padding computation.
+    /// We only care about the 3 least significant bits; semantically, this is a u3.
+    pad: u8,
+}
+
+impl<'a, 'r> FileReader<'a, 'r> {
+    /// Instantiate a new reader, starting after [wire::TOK_REG] or [wire::TOK_EXE].
+    /// We handle the terminating [wire::TOK_PAR] on semantic EOF.
+    fn new(reader: &'a mut Reader<'r>, len: u64) -> io::Result<Self> {
+        // For zero-length files, we have to read the terminating TOK_PAR
+        // immediately, since FileReader::read may never be called; we've
+        // already reached semantic EOF by definition.
+        if len == 0 {
+            read::token(reader, &wire::TOK_PAR)?;
+        }
+
+        Ok(Self {
+            reader,
+            len,
+            pad: len as u8,
+        })
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.len == 0
+    }
+
+    pub fn len(&self) -> u64 {
+        self.len
+    }
+}
+
+impl FileReader<'_, '_> {
+    /// Equivalent to [BufRead::fill_buf]
+    ///
+    /// We can't directly implement [BufRead], because [FileReader::consume] needs
+    /// to perform fallible I/O.
+    pub fn fill_buf(&mut self) -> io::Result<&[u8]> {
+        if self.is_empty() {
+            return Ok(&[]);
+        }
+
+        let mut buf = self.reader.fill_buf()?;
+
+        if buf.is_empty() {
+            return Err(UnexpectedEof.into());
+        }
+
+        if buf.len() as u64 > self.len {
+            buf = &buf[..self.len as usize];
+        }
+
+        Ok(buf)
+    }
+
+    /// Analogous to [BufRead::consume], differing only in that it needs
+    /// to perform I/O in order to read padding and terminators.
+    pub fn consume(&mut self, n: usize) -> io::Result<()> {
+        if n == 0 {
+            return Ok(());
+        }
+
+        self.len = self
+            .len
+            .checked_sub(n as u64)
+            .expect("consumed bytes past EOF");
+
+        self.reader.consume(n);
+
+        if self.is_empty() {
+            self.finish()?;
+        }
+
+        Ok(())
+    }
+
+    /// Copy the (remaining) contents of the file into `dst`.
+    pub fn copy(&mut self, mut dst: impl Write) -> io::Result<()> {
+        while !self.is_empty() {
+            let buf = self.fill_buf()?;
+            let n = dst.write(buf)?;
+            self.consume(n)?;
+        }
+
+        Ok(())
+    }
+}
+
+impl Read for FileReader<'_, '_> {
+    fn read(&mut self, mut buf: &mut [u8]) -> io::Result<usize> {
+        if buf.is_empty() || self.is_empty() {
+            return Ok(0);
+        }
+
+        if buf.len() as u64 > self.len {
+            buf = &mut buf[..self.len as usize];
+        }
+
+        let n = self.reader.read(buf)?;
+        self.len -= n as u64;
+
+        if n == 0 {
+            return Err(UnexpectedEof.into());
+        }
+
+        if self.is_empty() {
+            self.finish()?;
+        }
+
+        Ok(n)
+    }
+}
+
+impl FileReader<'_, '_> {
+    /// We've reached semantic EOF, consume and verify the padding and terminating TOK_PAR.
+    /// Files are padded to 64 bits (8 bytes), just like any other byte string in the wire format.
+    fn finish(&mut self) -> io::Result<()> {
+        let pad = (self.pad & 7) as usize;
+
+        if pad != 0 {
+            let mut buf = [0; 8];
+            self.reader.read_exact(&mut buf[pad..])?;
+
+            if buf != [0; 8] {
+                return Err(InvalidData.into());
+            }
+        }
+
+        read::token(self.reader, &wire::TOK_PAR)
+    }
+}
+
+/// A directory iterator, yielding a sequence of [Node]s.
+/// It must be fully consumed before reading further from the [DirReader] that produced it, if any.
+pub struct DirReader<'a, 'r> {
+    reader: &'a mut Reader<'r>,
+    /// Previous directory entry name.
+    /// We have to hang onto this to enforce name monotonicity.
+    prev_name: Option<Vec<u8>>,
+}
+
+pub struct Entry<'a, 'r> {
+    pub name: Vec<u8>,
+    pub node: Node<'a, 'r>,
+}
+
+impl<'a, 'r> DirReader<'a, 'r> {
+    fn new(reader: &'a mut Reader<'r>) -> Self {
+        Self {
+            reader,
+            prev_name: None,
+        }
+    }
+
+    /// Read the next [Entry] from the directory.
+    ///
+    /// We explicitly don't implement [Iterator], since treating this as
+    /// a regular Rust iterator will surely lead you astray.
+    ///
+    ///  * You must always consume the entire iterator, unless you abandon the entire archive reader.
+    ///  * You must abandon the entire archive reader on the first error.
+    ///  * You must abandon the directory reader upon the first [None].
+    ///  * Even if you know the amount of elements up front, you must keep reading until you encounter [None].
+    ///
+    /// TODO(edef): enforce these in `#[cfg(debug_assertions)]`
+    #[allow(clippy::should_implement_trait)]
+    pub fn next(&mut self) -> io::Result<Option<Entry>> {
+        // COME FROM the previous iteration: if we've already read an entry,
+        // read its terminating TOK_PAR here.
+        if self.prev_name.is_some() {
+            read::token(self.reader, &wire::TOK_PAR)?;
+        }
+
+        // Determine if there are more entries to follow
+        if let wire::Entry::None = read::tag(self.reader)? {
+            // We've reached the end of this directory.
+            return Ok(None);
+        }
+
+        let name = read::bytes(self.reader, wire::MAX_NAME_LEN)?;
+
+        if name.is_empty()
+            || name.contains(&0)
+            || name.contains(&b'/')
+            || name == b"."
+            || name == b".."
+        {
+            return Err(InvalidData.into());
+        }
+
+        // Enforce strict monotonicity of directory entry names.
+        match &mut self.prev_name {
+            None => {
+                self.prev_name = Some(name.clone());
+            }
+            Some(prev_name) => {
+                if *prev_name >= name {
+                    return Err(InvalidData.into());
+                }
+
+                name[..].clone_into(prev_name);
+            }
+        }
+
+        read::token(self.reader, &wire::TOK_NOD)?;
+
+        Ok(Some(Entry {
+            name,
+            node: Node::new(&mut self.reader)?,
+        }))
+    }
+}
diff --git a/tvix/nix-compat/src/nar/reader/read.rs b/tvix/nix-compat/src/nar/reader/read.rs
new file mode 100644
index 000000000000..1ce161376424
--- /dev/null
+++ b/tvix/nix-compat/src/nar/reader/read.rs
@@ -0,0 +1,109 @@
+//! Helpers for reading [crate::nar::wire] format.
+
+use std::io::{
+    self,
+    ErrorKind::{Interrupted, InvalidData, UnexpectedEof},
+};
+
+use super::Reader;
+use crate::nar::wire::Tag;
+
+/// Consume a little-endian [prim@u64] from the reader.
+pub fn u64(reader: &mut Reader) -> io::Result<u64> {
+    let mut buf = [0; 8];
+    reader.read_exact(&mut buf)?;
+    Ok(u64::from_le_bytes(buf))
+}
+
+/// Consume a byte string of up to `max_len` bytes from the reader.
+pub fn bytes(reader: &mut Reader, max_len: usize) -> io::Result<Vec<u8>> {
+    assert!(max_len <= isize::MAX as usize);
+
+    // read the length, and reject excessively large values
+    let len = self::u64(reader)?;
+    if len > max_len as u64 {
+        return Err(InvalidData.into());
+    }
+    // we know the length fits in a usize now
+    let len = len as usize;
+
+    // read the data and padding into a buffer
+    let buf_len = (len + 7) & !7;
+    let mut buf = vec![0; buf_len];
+    reader.read_exact(&mut buf)?;
+
+    // verify that the padding is all zeroes
+    for b in buf.drain(len..) {
+        if b != 0 {
+            return Err(InvalidData.into());
+        }
+    }
+
+    Ok(buf)
+}
+
+/// Consume a known token from the reader.
+pub fn token<const N: usize>(reader: &mut Reader, token: &[u8; N]) -> io::Result<()> {
+    let mut buf = [0u8; N];
+
+    // This implements something similar to [Read::read_exact], but verifies that
+    // the input data matches the token while we read it. These two slices respectively
+    // represent the remaining token to be verified, and the remaining input buffer.
+    let mut token = &token[..];
+    let mut buf = &mut buf[..];
+
+    while !token.is_empty() {
+        match reader.read(buf) {
+            Ok(0) => {
+                return Err(UnexpectedEof.into());
+            }
+            Ok(n) => {
+                let (t, b);
+                (t, token) = token.split_at(n);
+                (b, buf) = buf.split_at_mut(n);
+
+                if t != b {
+                    return Err(InvalidData.into());
+                }
+            }
+            Err(e) => {
+                if e.kind() != Interrupted {
+                    return Err(e);
+                }
+            }
+        }
+    }
+
+    Ok(())
+}
+
+/// Consume a [Tag] from the reader.
+pub fn tag<T: Tag>(reader: &mut Reader) -> io::Result<T> {
+    let mut buf = T::make_buf();
+    let buf = buf.as_mut();
+
+    // first read the known minimum length…
+    reader.read_exact(&mut buf[..T::MIN])?;
+
+    // then decide which tag we're expecting
+    let tag = T::from_u8(buf[T::OFF]).ok_or(InvalidData)?;
+    let (head, tail) = tag.as_bytes().split_at(T::MIN);
+
+    // make sure what we've read so far is valid
+    if buf[..T::MIN] != *head {
+        return Err(InvalidData.into());
+    }
+
+    // …then read the rest, if any
+    if !tail.is_empty() {
+        let rest = tail.len();
+        reader.read_exact(&mut buf[..rest])?;
+
+        // and make sure it's what we expect
+        if buf[..rest] != *tail {
+            return Err(InvalidData.into());
+        }
+    }
+
+    Ok(tag)
+}
diff --git a/tvix/nix-compat/src/nar/reader/test.rs b/tvix/nix-compat/src/nar/reader/test.rs
new file mode 100644
index 000000000000..fd0d6a9f5afd
--- /dev/null
+++ b/tvix/nix-compat/src/nar/reader/test.rs
@@ -0,0 +1,120 @@
+use std::io::Read;
+
+use crate::nar;
+
+#[test]
+fn symlink() {
+    let mut f = std::io::Cursor::new(include_bytes!("../tests/symlink.nar"));
+    let node = nar::reader::open(&mut f).unwrap();
+
+    match node {
+        nar::reader::Node::Symlink { target } => {
+            assert_eq!(
+                &b"/nix/store/somewhereelse"[..],
+                &target,
+                "target must match"
+            );
+        }
+        _ => panic!("unexpected type"),
+    }
+}
+
+#[test]
+fn file() {
+    let mut f = std::io::Cursor::new(include_bytes!("../tests/helloworld.nar"));
+    let node = nar::reader::open(&mut f).unwrap();
+
+    match node {
+        nar::reader::Node::File {
+            executable,
+            mut reader,
+        } => {
+            assert!(!executable);
+            let mut buf = vec![];
+            reader.read_to_end(&mut buf).expect("read must succeed");
+            assert_eq!(&b"Hello World!"[..], &buf);
+        }
+        _ => panic!("unexpected type"),
+    }
+}
+
+#[test]
+fn complicated() {
+    let mut f = std::io::Cursor::new(include_bytes!("../tests/complicated.nar"));
+    let node = nar::reader::open(&mut f).unwrap();
+
+    match node {
+        nar::reader::Node::Directory(mut dir_reader) => {
+            // first entry is .keep, an empty regular file.
+            let entry = dir_reader
+                .next()
+                .expect("next must succeed")
+                .expect("must be some");
+
+            assert_eq!(&b".keep"[..], &entry.name);
+
+            match entry.node {
+                nar::reader::Node::File {
+                    executable,
+                    mut reader,
+                } => {
+                    assert!(!executable);
+                    assert_eq!(reader.read(&mut [0]).unwrap(), 0);
+                }
+                _ => panic!("unexpected type for .keep"),
+            }
+
+            // second entry is aa, a symlink to /nix/store/somewhereelse
+            let entry = dir_reader
+                .next()
+                .expect("next must be some")
+                .expect("must be some");
+
+            assert_eq!(&b"aa"[..], &entry.name);
+
+            match entry.node {
+                nar::reader::Node::Symlink { target } => {
+                    assert_eq!(&b"/nix/store/somewhereelse"[..], &target);
+                }
+                _ => panic!("unexpected type for aa"),
+            }
+
+            // third entry is a directory called "keep"
+            let entry = dir_reader
+                .next()
+                .expect("next must be some")
+                .expect("must be some");
+
+            assert_eq!(&b"keep"[..], &entry.name);
+
+            match entry.node {
+                nar::reader::Node::Directory(mut subdir_reader) => {
+                    // first entry is .keep, an empty regular file.
+                    let entry = subdir_reader
+                        .next()
+                        .expect("next must succeed")
+                        .expect("must be some");
+
+                    // … it contains a single .keep, an empty regular file.
+                    assert_eq!(&b".keep"[..], &entry.name);
+
+                    match entry.node {
+                        nar::reader::Node::File {
+                            executable,
+                            mut reader,
+                        } => {
+                            assert!(!executable);
+                            assert_eq!(reader.read(&mut [0]).unwrap(), 0);
+                        }
+                        _ => panic!("unexpected type for keep/.keep"),
+                    }
+                }
+                _ => panic!("unexpected type for keep/.keep"),
+            }
+
+            // reading more entries yields None (and we actually must read until this)
+            assert!(dir_reader.next().expect("must succeed").is_none());
+        }
+        _ => panic!("unexpected type"),
+    }
+}
diff --git a/tvix/nix-compat/src/nar/tests/complicated.nar b/tvix/nix-compat/src/nar/tests/complicated.nar
new file mode 100644
index 000000000000..6a137f5fbb6b
--- /dev/null
+++ b/tvix/nix-compat/src/nar/tests/complicated.nar
Binary files differdiff --git a/tvix/nix-compat/src/nar/tests/helloworld.nar b/tvix/nix-compat/src/nar/tests/helloworld.nar
new file mode 100644
index 000000000000..2e1268115205
--- /dev/null
+++ b/tvix/nix-compat/src/nar/tests/helloworld.nar
Binary files differdiff --git a/tvix/nix-compat/src/nar/tests/symlink.nar b/tvix/nix-compat/src/nar/tests/symlink.nar
new file mode 100644
index 000000000000..7990e4ad5bc2
--- /dev/null
+++ b/tvix/nix-compat/src/nar/tests/symlink.nar
Binary files differdiff --git a/tvix/nix-compat/src/nar/wire/mod.rs b/tvix/nix-compat/src/nar/wire/mod.rs
new file mode 100644
index 000000000000..b9e021249543
--- /dev/null
+++ b/tvix/nix-compat/src/nar/wire/mod.rs
@@ -0,0 +1,133 @@
+//! NAR wire format, without I/O details, since those differ between
+//! the synchronous and asynchronous implementations.
+//!
+//! The wire format is an S-expression format, encoded onto the wire
+//! using simple encoding rules.
+//!
+//! # Encoding
+//!
+//! Lengths are represented as 64-bit unsigned integers in little-endian
+//! format. Byte strings, including file contents and syntactic strings
+//! part of the grammar, are prefixed by their 64-bit length, and padded
+//! to 8-byte (64-bit) alignment with zero bytes. The zero-length string
+//! is therefore encoded as eight zero bytes representing its length.
+//!
+//! # Grammar
+//!
+//! The NAR grammar is as follows:
+//! ```plain
+//! archive ::= "nix-archive-1" node
+//!
+//! node ::= "(" "type" "symlink" "target" string ")"
+//!      ||= "(" "type" "regular" ("executable" "")? "contents" string ")"
+//!      ||= "(" "type" "directory" entry* ")"
+//!
+//! entry ::= "entry" "(" "name" string "node" node ")"
+//! ```
+//!
+//! We rewrite it to pull together the purely syntactic elements into
+//! unified tokens, producing an equivalent grammar that can be parsed
+//! and serialized more elegantly:
+//! ```plain
+//! archive ::= TOK_NAR node
+//! node ::= TOK_SYM string             TOK_PAR
+//!      ||= (TOK_REG | TOK_EXE) string TOK_PAR
+//!      ||= TOK_DIR entry*             TOK_PAR
+//!
+//! entry ::= TOK_ENT string TOK_NOD node TOK_PAR
+//!
+//! TOK_NAR ::= "nix-archive-1" "(" "type"
+//! TOK_SYM ::= "symlink" "target"
+//! TOK_REG ::= "regular" "contents"
+//! TOK_EXE ::= "regular" "executable" ""
+//! TOK_DIR ::= "directory"
+//! TOK_ENT ::= "entry" "(" "name"
+//! TOK_NOD ::= "node" "(" "type"
+//! TOK_PAR ::= ")"
+//! ```
+//!
+//! # Restrictions
+//!
+//! NOTE: These restrictions are not (and cannot be) enforced by this module,
+//! but must be enforced by its consumers, [super::reader] and [super::writer].
+//!
+//! Directory entry names cannot have the reserved names `.` and `..`, nor contain
+//! forward slashes. They must appear in strictly ascending lexicographic order
+//! within a directory, and can be at most [MAX_NAME_LEN] bytes in length.
+//!
+//! Symlink targets can be at most [MAX_TARGET_LEN] bytes in length.
+//!
+//! Neither is permitted to be empty, or contain null bytes.
+
+// These values are the standard Linux length limits
+/// Maximum length of a directory entry name
+pub const MAX_NAME_LEN: usize = 255;
+/// Maximum length of a symlink target
+pub const MAX_TARGET_LEN: usize = 4095;
+
+#[cfg(test)]
+fn token(xs: &[&str]) -> Vec<u8> {
+    let mut out = vec![];
+    for x in xs {
+        let len = x.len() as u64;
+        out.extend_from_slice(&len.to_le_bytes());
+        out.extend_from_slice(x.as_bytes());
+
+        let n = x.len() & 7;
+        if n != 0 {
+            const ZERO: [u8; 8] = [0; 8];
+            out.extend_from_slice(&ZERO[n..]);
+        }
+    }
+    out
+}
+
+pub const TOK_NAR: [u8; 56] = *b"\x0d\0\0\0\0\0\0\0nix-archive-1\0\0\0\x01\0\0\0\0\0\0\0(\0\0\0\0\0\0\0\x04\0\0\0\0\0\0\0type\0\0\0\0";
+pub const TOK_SYM: [u8; 32] = *b"\x07\0\0\0\0\0\0\0symlink\0\x06\0\0\0\0\0\0\0target\0\0";
+pub const TOK_REG: [u8; 32] = *b"\x07\0\0\0\0\0\0\0regular\0\x08\0\0\0\0\0\0\0contents";
+pub const TOK_EXE: [u8; 64] = *b"\x07\0\0\0\0\0\0\0regular\0\x0a\0\0\0\0\0\0\0executable\0\0\0\0\0\0\0\0\0\0\0\0\0\0\x08\0\0\0\0\0\0\0contents";
+pub const TOK_DIR: [u8; 24] = *b"\x09\0\0\0\0\0\0\0directory\0\0\0\0\0\0\0";
+pub const TOK_ENT: [u8; 48] = *b"\x05\0\0\0\0\0\0\0entry\0\0\0\x01\0\0\0\0\0\0\0(\0\0\0\0\0\0\0\x04\0\0\0\0\0\0\0name\0\0\0\0";
+pub const TOK_NOD: [u8; 48] = *b"\x04\0\0\0\0\0\0\0node\0\0\0\0\x01\0\0\0\0\0\0\0(\0\0\0\0\0\0\0\x04\0\0\0\0\0\0\0type\0\0\0\0";
+pub const TOK_PAR: [u8; 16] = *b"\x01\0\0\0\0\0\0\0)\0\0\0\0\0\0\0";
+
+#[test]
+fn tokens() {
+    let cases: &[(&[u8], &[&str])] = &[
+        (&TOK_NAR, &["nix-archive-1", "(", "type"]),
+        (&TOK_SYM, &["symlink", "target"]),
+        (&TOK_REG, &["regular", "contents"]),
+        (&TOK_EXE, &["regular", "executable", "", "contents"]),
+        (&TOK_DIR, &["directory"]),
+        (&TOK_ENT, &["entry", "(", "name"]),
+        (&TOK_NOD, &["node", "(", "type"]),
+        (&TOK_PAR, &[")"]),
+    ];
+
+    for &(tok, xs) in cases {
+        assert_eq!(tok, token(xs));
+    }
+}
+
+pub use tag::Tag;
+mod tag;
+
+tag::make! {
+    /// These are the node tokens, succeeding [TOK_NAR] or [TOK_NOD],
+    /// and preceding the next variable-length element.
+    pub enum Node[16] {
+        Sym = TOK_SYM,
+        Reg = TOK_REG,
+        Exe = TOK_EXE,
+        Dir = TOK_DIR,
+    }
+
+    /// Directory entry or terminator
+    pub enum Entry[0] {
+        /// End of directory
+        None = TOK_PAR,
+        /// Directory entry
+        /// Followed by a name string, [TOK_NOD], and a [Node].
+        Some = TOK_ENT,
+    }
+}
diff --git a/tvix/nix-compat/src/nar/wire/tag.rs b/tvix/nix-compat/src/nar/wire/tag.rs
new file mode 100644
index 000000000000..55b93f998541
--- /dev/null
+++ b/tvix/nix-compat/src/nar/wire/tag.rs
@@ -0,0 +1,165 @@
+/// A type implementing Tag represents a static hash set of byte strings,
+/// with a very simple perfect hash function: every element has a unique
+/// discriminant at a common byte offset. The values of the type represent
+/// the members by this single discriminant byte; they are indices into the
+/// hash set.
+pub trait Tag: Sized {
+    /// Discriminant offset
+    const OFF: usize;
+    /// Minimum variant length
+    const MIN: usize;
+
+    /// Minimal suitably sized buffer for reading the wire representation
+    /// HACK: This is a workaround for const generics limitations.
+    type Buf: AsMut<[u8]> + Send;
+
+    /// Make an instance of [Self::Buf]
+    fn make_buf() -> Self::Buf;
+
+    /// Convert a discriminant into the corresponding variant
+    fn from_u8(x: u8) -> Option<Self>;
+
+    /// Convert a variant back into the wire representation
+    fn as_bytes(&self) -> &'static [u8];
+}
+
+/// Generate an enum implementing [Tag], enforcing at compile time that
+/// the discriminant values are distinct.
+macro_rules! make {
+    (
+        $(
+            $(#[doc = $doc:expr])*
+            $vis:vis enum $Enum:ident[$off:expr] {
+                $(
+                    $(#[doc = $var_doc:expr])*
+                    $Var:ident = $TOK:ident,
+                )+
+            }
+        )*
+    ) => {
+        $(
+            $(#[doc = $doc])*
+            #[derive(Debug, PartialEq, Eq)]
+            #[repr(u8)]
+            $vis enum $Enum {
+                $(
+                    $(#[doc = $var_doc])*
+                    $Var = $TOK[$Enum::OFF]
+                ),+
+            }
+
+            impl Tag for $Enum {
+                /// Discriminant offset
+                const OFF: usize = $off;
+                /// Minimum variant length
+                const MIN: usize = tag::min_of(&[$($TOK.len()),+]);
+
+                /// Minimal suitably sized buffer for reading the wire representation
+                type Buf = [u8; tag::buf_of(&[$($TOK.len()),+])];
+
+                /// Make an instance of [Self::Buf]
+                #[inline(always)]
+                fn make_buf() -> Self::Buf {
+                    [0u8; tag::buf_of(&[$($TOK.len()),+])]
+                }
+
+                /// Convert a discriminant into the corresponding variant
+                #[inline(always)]
+                fn from_u8(x: u8) -> Option<Self> {
+                    #[allow(non_upper_case_globals)]
+                    mod __variant {
+                        $(
+                            pub const $Var: u8 = super::$Enum::$Var as u8;
+                        )+
+                    }
+
+                    match x {
+                        $(__variant::$Var => Some(Self::$Var),)+
+                        _ => None
+                    }
+                }
+
+                /// Convert a variant back into the wire representation
+                #[inline(always)]
+                fn as_bytes(&self) -> &'static [u8] {
+                    match self {
+                        $(Self::$Var => &$TOK,)+
+                    }
+                }
+            }
+        )*
+    };
+}
+
+// The following functions are written somewhat unusually,
+// since they're const functions that cannot use iterators.
+
+/// Maximum element of a slice
+const fn max_of(mut xs: &[usize]) -> usize {
+    let mut y = usize::MIN;
+    while let &[x, ref tail @ ..] = xs {
+        y = if x > y { x } else { y };
+        xs = tail;
+    }
+    y
+}
+
+/// Minimum element of a slice
+pub const fn min_of(mut xs: &[usize]) -> usize {
+    let mut y = usize::MAX;
+    while let &[x, ref tail @ ..] = xs {
+        y = if x < y { x } else { y };
+        xs = tail;
+    }
+    y
+}
+
+/// Minimum buffer size to contain either of `0..Tag::MIN` and `Tag::MIN..`
+/// at a particular time, for all possible tag wire representations, given
+/// the sizes of all wire representations.
+///
+/// # Example
+///
+/// ```plain
+/// OFF = 16
+/// MIN = 24
+/// MAX = 64
+///
+/// BUF = max(MIN, MAX-MIN)
+///     = max(24, 64-24)
+///     = max(24, 40)
+///     = 40
+/// ```
+pub const fn buf_of(xs: &[usize]) -> usize {
+    max_of(&[min_of(xs), max_of(xs) - min_of(xs)])
+}
+
+pub(crate) use make;
+
+#[cfg(test)]
+mod test {
+    use super::super::tag::{self, Tag};
+
+    const TOK_A: [u8; 3] = [0xed, 0xef, 0x1c];
+    const TOK_B: [u8; 3] = [0xed, 0xf0, 0x1c];
+
+    const OFFSET: usize = 1;
+
+    make! {
+        enum Token[OFFSET] {
+            A = TOK_A,
+            B = TOK_B,
+        }
+    }
+
+    #[test]
+    fn example() {
+        assert_eq!(Token::from_u8(0xed), None);
+
+        let tag = Token::from_u8(0xef).unwrap();
+        assert_eq!(tag.as_bytes(), &TOK_A[..]);
+
+        let tag = Token::from_u8(0xf0).unwrap();
+        assert_eq!(tag.as_bytes(), &TOK_B[..]);
+    }
+}
diff --git a/tvix/nix-compat/src/nar/writer/async.rs b/tvix/nix-compat/src/nar/writer/async.rs
new file mode 100644
index 000000000000..11aefab9cba7
--- /dev/null
+++ b/tvix/nix-compat/src/nar/writer/async.rs
@@ -0,0 +1,235 @@
+//! Implements an interface for writing the Nix archive format (NAR).
+//!
+//! NAR files (and their hashed representations) are used in C++ Nix for
+//! addressing fixed-output derivations and a variety of other things.
+//!
+//! NAR files can be output to any type that implements [`AsyncWrite`], and content
+//! can be read from any type that implementes [`AsyncBufRead`].
+//!
+//! Writing a single file might look like this:
+//!
+//! ```rust
+//! # futures::executor::block_on(async {
+//! # use futures::io::BufReader;
+//! # let some_file: Vec<u8> = vec![0, 1, 2, 3, 4];
+//!
+//! // Output location to write the NAR to.
+//! let mut sink: Vec<u8> = Vec::new();
+//!
+//! // Instantiate writer for this output location.
+//! let mut nar = nix_compat::nar::writer::r#async::open(&mut sink).await?;
+//!
+//! // Acquire metadata for the single file to output, and pass it in a
+//! // `BufRead`-implementing type.
+//!
+//! let executable = false;
+//! let size = some_file.len() as u64;
+//! let mut reader = BufReader::new(some_file.as_slice());
+//! nar.file(executable, size, &mut reader).await?;
+//! # Ok::<(), std::io::Error>(())
+//! # });
+//! ```
+
+use crate::nar::wire;
+use futures_util::{AsyncBufRead, AsyncBufReadExt, AsyncWrite, AsyncWriteExt};
+use std::{
+    io::{
+        self,
+        ErrorKind::{InvalidInput, UnexpectedEof},
+    },
+    pin::Pin,
+};
+
+/// Convenience type alias for types implementing [`AsyncWrite`].
+pub type Writer<'a> = dyn AsyncWrite + Unpin + Send + 'a;
+
+/// Create a new NAR, writing the output to the specified writer.
+pub async fn open<'a, 'w: 'a>(writer: &'a mut Writer<'w>) -> io::Result<Node<'a, 'w>> {
+    let mut node = Node { writer };
+    node.write(&wire::TOK_NAR).await?;
+    Ok(node)
+}
+
+/// Single node in a NAR file.
+///
+/// A NAR can be thought of as a tree of nodes represented by this type. Each
+/// node can be a file, a symlink or a directory containing other nodes.
+pub struct Node<'a, 'w: 'a> {
+    writer: &'a mut Writer<'w>,
+}
+
+impl<'a, 'w> Node<'a, 'w> {
+    async fn write(&mut self, data: &[u8]) -> io::Result<()> {
+        self.writer.write_all(data).await
+    }
+
+    async fn pad(&mut self, n: u64) -> io::Result<()> {
+        match (n & 7) as usize {
+            0 => Ok(()),
+            n => self.write(&[0; 8][n..]).await,
+        }
+    }
+
+    /// Make this node a symlink.
+    pub async fn symlink(mut self, target: &[u8]) -> io::Result<()> {
+        debug_assert!(
+            target.len() <= wire::MAX_TARGET_LEN,
+            "target.len() > {}",
+            wire::MAX_TARGET_LEN
+        );
+        debug_assert!(!target.is_empty(), "target is empty");
+        debug_assert!(!target.contains(&0), "target contains null byte");
+
+        self.write(&wire::TOK_SYM).await?;
+        self.write(&target.len().to_le_bytes()).await?;
+        self.write(target).await?;
+        self.pad(target.len() as u64).await?;
+        self.write(&wire::TOK_PAR).await?;
+        Ok(())
+    }
+
+    /// Make this node a single file.
+    pub async fn file(
+        mut self,
+        executable: bool,
+        size: u64,
+        reader: &mut (dyn AsyncBufRead + Unpin + Send),
+    ) -> io::Result<()> {
+        self.write(if executable {
+            &wire::TOK_EXE
+        } else {
+            &wire::TOK_REG
+        })
+        .await?;
+
+        self.write(&size.to_le_bytes()).await?;
+
+        let mut need = size;
+        while need != 0 {
+            let data = reader.fill_buf().await?;
+
+            if data.is_empty() {
+                return Err(UnexpectedEof.into());
+            }
+
+            let n = need.min(data.len() as u64) as usize;
+            self.write(&data[..n]).await?;
+
+            need -= n as u64;
+            Pin::new(&mut *reader).consume(n);
+        }
+
+        // bail if there's still data left in the passed reader.
+        // This uses the same code as [BufRead::has_data_left] (unstable).
+        if reader.fill_buf().await.map(|b| !b.is_empty())? {
+            return Err(io::Error::new(
+                InvalidInput,
+                "reader contained more data than specified size",
+            ));
+        }
+
+        self.pad(size).await?;
+        self.write(&wire::TOK_PAR).await?;
+
+        Ok(())
+    }
+
+    /// Make this node a directory, the content of which is set using the
+    /// resulting [`Directory`] value.
+    ///
+    /// It is the caller's responsibility to invoke [`Directory::close`],
+    /// or invalid archives will be produced silently.
+    pub async fn directory(mut self) -> io::Result<Directory<'a, 'w>> {
+        self.write(&wire::TOK_DIR).await?;
+        Ok(Directory::new(self))
+    }
+}
+
+#[cfg(debug_assertions)]
+type Name = Vec<u8>;
+#[cfg(not(debug_assertions))]
+type Name = ();
+
+fn into_name(_name: &[u8]) -> Name {
+    #[cfg(debug_assertions)]
+    _name.to_owned()
+}
+
+/// Content of a NAR node that represents a directory.
+pub struct Directory<'a, 'w> {
+    node: Node<'a, 'w>,
+    prev_name: Option<Name>,
+}
+
+impl<'a, 'w> Directory<'a, 'w> {
+    fn new(node: Node<'a, 'w>) -> Self {
+        Self {
+            node,
+            prev_name: None,
+        }
+    }
+
+    /// Add an entry to the directory.
+    ///
+    /// The entry is simply another [`Node`], which can then be filled like the
+    /// root of a NAR (including, of course, by nesting directories).
+    ///
+    /// It is the caller's responsibility to ensure that directory entries are
+    /// written in order of ascending name. If this is not ensured, this method
+    /// may panic or silently produce invalid archives.
+    pub async fn entry(&mut self, name: &[u8]) -> io::Result<Node<'_, 'w>> {
+        debug_assert!(
+            name.len() <= wire::MAX_NAME_LEN,
+            "name.len() > {}",
+            wire::MAX_NAME_LEN
+        );
+        debug_assert!(!name.is_empty(), "name is empty");
+        debug_assert!(!name.contains(&0), "name contains null byte");
+        debug_assert!(!name.contains(&b'/'), "name contains {:?}", '/');
+        debug_assert!(name != b".", "name == {:?}", ".");
+        debug_assert!(name != b"..", "name == {:?}", "..");
+
+        match self.prev_name {
+            None => {
+                self.prev_name = Some(into_name(name));
+            }
+            Some(ref mut _prev_name) => {
+                #[cfg(debug_assertions)]
+                {
+                    use bstr::ByteSlice;
+                    assert!(
+                        &**_prev_name < name,
+                        "misordered names: {:?} >= {:?}",
+                        _prev_name.as_bstr(),
+                        name.as_bstr()
+                    );
+                    name.clone_into(_prev_name);
+                }
+                self.node.write(&wire::TOK_PAR).await?;
+            }
+        }
+
+        self.node.write(&wire::TOK_ENT).await?;
+        self.node.write(&name.len().to_le_bytes()).await?;
+        self.node.write(name).await?;
+        self.node.pad(name.len() as u64).await?;
+        self.node.write(&wire::TOK_NOD).await?;
+
+        Ok(Node {
+            writer: &mut *self.node.writer,
+        })
+    }
+
+    /// Close a directory and write terminators for the directory to the NAR.
+    ///
+    /// **Important:** This *must* be called when all entries have been written
+    /// in a directory, otherwise the resulting NAR file will be invalid.
+    pub async fn close(mut self) -> io::Result<()> {
+        if self.prev_name.is_some() {
+            self.node.write(&wire::TOK_PAR).await?;
+        }
+
+        self.node.write(&wire::TOK_PAR).await?;
+        Ok(())
+    }
+}
diff --git a/tvix/nix-compat/src/nar/writer/mod.rs b/tvix/nix-compat/src/nar/writer/mod.rs
new file mode 100644
index 000000000000..fe8ccccb3787
--- /dev/null
+++ b/tvix/nix-compat/src/nar/writer/mod.rs
@@ -0,0 +1,9 @@
+pub use sync::*;
+
+pub mod sync;
+
+#[cfg(test)]
+mod test;
+
+#[cfg(feature = "async")]
+pub mod r#async;
diff --git a/tvix/nix-compat/src/nar/writer/sync.rs b/tvix/nix-compat/src/nar/writer/sync.rs
new file mode 100644
index 000000000000..6270129028fa
--- /dev/null
+++ b/tvix/nix-compat/src/nar/writer/sync.rs
@@ -0,0 +1,224 @@
+//! Implements an interface for writing the Nix archive format (NAR).
+//!
+//! NAR files (and their hashed representations) are used in C++ Nix for
+//! addressing fixed-output derivations and a variety of other things.
+//!
+//! NAR files can be output to any type that implements [`Write`], and content
+//! can be read from any type that implementes [`BufRead`].
+//!
+//! Writing a single file might look like this:
+//!
+//! ```rust
+//! # use std::io::BufReader;
+//! # let some_file: Vec<u8> = vec![0, 1, 2, 3, 4];
+//!
+//! // Output location to write the NAR to.
+//! let mut sink: Vec<u8> = Vec::new();
+//!
+//! // Instantiate writer for this output location.
+//! let mut nar = nix_compat::nar::writer::open(&mut sink)?;
+//!
+//! // Acquire metadata for the single file to output, and pass it in a
+//! // `BufRead`-implementing type.
+//!
+//! let executable = false;
+//! let size = some_file.len() as u64;
+//! let mut reader = BufReader::new(some_file.as_slice());
+//! nar.file(executable, size, &mut reader)?;
+//! # Ok::<(), std::io::Error>(())
+//! ```
+
+use crate::nar::wire;
+use std::io::{
+    self, BufRead,
+    ErrorKind::{InvalidInput, UnexpectedEof},
+    Write,
+};
+
+/// Convenience type alias for types implementing [`Write`].
+pub type Writer<'a> = dyn Write + Send + 'a;
+
+/// Create a new NAR, writing the output to the specified writer.
+pub fn open<'a, 'w: 'a>(writer: &'a mut Writer<'w>) -> io::Result<Node<'a, 'w>> {
+    let mut node = Node { writer };
+    node.write(&wire::TOK_NAR)?;
+    Ok(node)
+}
+
+/// Single node in a NAR file.
+///
+/// A NAR can be thought of as a tree of nodes represented by this type. Each
+/// node can be a file, a symlink or a directory containing other nodes.
+pub struct Node<'a, 'w: 'a> {
+    writer: &'a mut Writer<'w>,
+}
+
+impl<'a, 'w> Node<'a, 'w> {
+    fn write(&mut self, data: &[u8]) -> io::Result<()> {
+        self.writer.write_all(data)
+    }
+
+    fn pad(&mut self, n: u64) -> io::Result<()> {
+        match (n & 7) as usize {
+            0 => Ok(()),
+            n => self.write(&[0; 8][n..]),
+        }
+    }
+
+    /// Make this node a symlink.
+    pub fn symlink(mut self, target: &[u8]) -> io::Result<()> {
+        debug_assert!(
+            target.len() <= wire::MAX_TARGET_LEN,
+            "target.len() > {}",
+            wire::MAX_TARGET_LEN
+        );
+        debug_assert!(!target.is_empty(), "target is empty");
+        debug_assert!(!target.contains(&0), "target contains null byte");
+
+        self.write(&wire::TOK_SYM)?;
+        self.write(&target.len().to_le_bytes())?;
+        self.write(target)?;
+        self.pad(target.len() as u64)?;
+        self.write(&wire::TOK_PAR)?;
+        Ok(())
+    }
+
+    /// Make this node a single file.
+    pub fn file(mut self, executable: bool, size: u64, reader: &mut dyn BufRead) -> io::Result<()> {
+        self.write(if executable {
+            &wire::TOK_EXE
+        } else {
+            &wire::TOK_REG
+        })?;
+
+        self.write(&size.to_le_bytes())?;
+
+        let mut need = size;
+        while need != 0 {
+            let data = reader.fill_buf()?;
+
+            if data.is_empty() {
+                return Err(UnexpectedEof.into());
+            }
+
+            let n = need.min(data.len() as u64) as usize;
+            self.write(&data[..n])?;
+
+            need -= n as u64;
+            reader.consume(n);
+        }
+
+        // bail if there's still data left in the passed reader.
+        // This uses the same code as [BufRead::has_data_left] (unstable).
+        if reader.fill_buf().map(|b| !b.is_empty())? {
+            return Err(io::Error::new(
+                InvalidInput,
+                "reader contained more data than specified size",
+            ));
+        }
+
+        self.pad(size)?;
+        self.write(&wire::TOK_PAR)?;
+
+        Ok(())
+    }
+
+    /// Make this node a directory, the content of which is set using the
+    /// resulting [`Directory`] value.
+    ///
+    /// It is the caller's responsibility to invoke [`Directory::close`],
+    /// or invalid archives will be produced silently.
+    pub fn directory(mut self) -> io::Result<Directory<'a, 'w>> {
+        self.write(&wire::TOK_DIR)?;
+        Ok(Directory::new(self))
+    }
+}
+
+#[cfg(debug_assertions)]
+type Name = Vec<u8>;
+#[cfg(not(debug_assertions))]
+type Name = ();
+
+fn into_name(_name: &[u8]) -> Name {
+    #[cfg(debug_assertions)]
+    _name.to_owned()
+}
+
+/// Content of a NAR node that represents a directory.
+pub struct Directory<'a, 'w> {
+    node: Node<'a, 'w>,
+    prev_name: Option<Name>,
+}
+
+impl<'a, 'w> Directory<'a, 'w> {
+    fn new(node: Node<'a, 'w>) -> Self {
+        Self {
+            node,
+            prev_name: None,
+        }
+    }
+
+    /// Add an entry to the directory.
+    ///
+    /// The entry is simply another [`Node`], which can then be filled like the
+    /// root of a NAR (including, of course, by nesting directories).
+    ///
+    /// It is the caller's responsibility to ensure that directory entries are
+    /// written in order of ascending name. If this is not ensured, this method
+    /// may panic or silently produce invalid archives.
+    pub fn entry(&mut self, name: &[u8]) -> io::Result<Node<'_, 'w>> {
+        debug_assert!(
+            name.len() <= wire::MAX_NAME_LEN,
+            "name.len() > {}",
+            wire::MAX_NAME_LEN
+        );
+        debug_assert!(!name.is_empty(), "name is empty");
+        debug_assert!(!name.contains(&0), "name contains null byte");
+        debug_assert!(!name.contains(&b'/'), "name contains {:?}", '/');
+        debug_assert!(name != b".", "name == {:?}", ".");
+        debug_assert!(name != b"..", "name == {:?}", "..");
+
+        match self.prev_name {
+            None => {
+                self.prev_name = Some(into_name(name));
+            }
+            Some(ref mut _prev_name) => {
+                #[cfg(debug_assertions)]
+                {
+                    use bstr::ByteSlice;
+                    assert!(
+                        &**_prev_name < name,
+                        "misordered names: {:?} >= {:?}",
+                        _prev_name.as_bstr(),
+                        name.as_bstr()
+                    );
+                    name.clone_into(_prev_name);
+                }
+                self.node.write(&wire::TOK_PAR)?;
+            }
+        }
+
+        self.node.write(&wire::TOK_ENT)?;
+        self.node.write(&name.len().to_le_bytes())?;
+        self.node.write(name)?;
+        self.node.pad(name.len() as u64)?;
+        self.node.write(&wire::TOK_NOD)?;
+
+        Ok(Node {
+            writer: &mut *self.node.writer,
+        })
+    }
+
+    /// Close a directory and write terminators for the directory to the NAR.
+    ///
+    /// **Important:** This *must* be called when all entries have been written
+    /// in a directory, otherwise the resulting NAR file will be invalid.
+    pub fn close(mut self) -> io::Result<()> {
+        if self.prev_name.is_some() {
+            self.node.write(&wire::TOK_PAR)?;
+        }
+
+        self.node.write(&wire::TOK_PAR)?;
+        Ok(())
+    }
+}
diff --git a/tvix/nix-compat/src/nar/writer/test.rs b/tvix/nix-compat/src/nar/writer/test.rs
new file mode 100644
index 000000000000..7b1dd1a2a96e
--- /dev/null
+++ b/tvix/nix-compat/src/nar/writer/test.rs
@@ -0,0 +1,133 @@
+use crate::nar;
+
+#[test]
+fn symlink() {
+    let mut buf = vec![];
+    let node = nar::writer::open(&mut buf).unwrap();
+
+    node.symlink("/nix/store/somewhereelse".as_bytes()).unwrap();
+
+    assert_eq!(include_bytes!("../tests/symlink.nar"), buf.as_slice());
+}
+
+#[cfg(feature = "async")]
+#[test]
+fn symlink_async() {
+    let mut buf = vec![];
+
+    futures::executor::block_on(async {
+        let node = nar::writer::r#async::open(&mut buf).await.unwrap();
+        node.symlink("/nix/store/somewhereelse".as_bytes())
+            .await
+            .unwrap();
+    });
+
+    assert_eq!(include_bytes!("../tests/symlink.nar"), buf.as_slice());
+}
+
+#[test]
+fn file() {
+    let mut buf = vec![];
+    let node = nar::writer::open(&mut buf).unwrap();
+
+    let file_contents = "Hello World!".to_string();
+    node.file(
+        false,
+        file_contents.len() as u64,
+        &mut std::io::Cursor::new(file_contents),
+    )
+    .unwrap();
+
+    assert_eq!(include_bytes!("../tests/helloworld.nar"), buf.as_slice());
+}
+
+#[cfg(feature = "async")]
+#[test]
+fn file_async() {
+    let mut buf = vec![];
+
+    futures::executor::block_on(async {
+        let node = nar::writer::r#async::open(&mut buf).await.unwrap();
+
+        let file_contents = "Hello World!".to_string();
+        node.file(
+            false,
+            file_contents.len() as u64,
+            &mut futures::io::Cursor::new(file_contents),
+        )
+        .await
+        .unwrap();
+    });
+
+    assert_eq!(include_bytes!("../tests/helloworld.nar"), buf.as_slice());
+}
+
+#[test]
+fn complicated() {
+    let mut buf = vec![];
+    let node = nar::writer::open(&mut buf).unwrap();
+
+    let mut dir_node = node.directory().unwrap();
+
+    let e = dir_node.entry(".keep".as_bytes()).unwrap();
+    e.file(false, 0, &mut std::io::Cursor::new([]))
+        .expect("read .keep must succeed");
+
+    let e = dir_node.entry("aa".as_bytes()).unwrap();
+    e.symlink("/nix/store/somewhereelse".as_bytes())
+        .expect("symlink must succeed");
+
+    let e = dir_node.entry("keep".as_bytes()).unwrap();
+    let mut subdir_node = e.directory().expect("directory must succeed");
+
+    let e_sub = subdir_node
+        .entry(".keep".as_bytes())
+        .expect("subdir entry must succeed");
+    e_sub.file(false, 0, &mut std::io::Cursor::new([])).unwrap();
+
+    // close the subdir, and then the dir, which is required.
+    subdir_node.close().unwrap();
+    dir_node.close().unwrap();
+
+    assert_eq!(include_bytes!("../tests/complicated.nar"), buf.as_slice());
+}
+
+#[cfg(feature = "async")]
+#[test]
+fn complicated_async() {
+    let mut buf = vec![];
+
+    futures::executor::block_on(async {
+        let node = nar::writer::r#async::open(&mut buf).await.unwrap();
+
+        let mut dir_node = node.directory().await.unwrap();
+
+        let e = dir_node.entry(".keep".as_bytes()).await.unwrap();
+        e.file(false, 0, &mut futures::io::Cursor::new([]))
+            .await
+            .expect("read .keep must succeed");
+
+        let e = dir_node.entry("aa".as_bytes()).await.unwrap();
+        e.symlink("/nix/store/somewhereelse".as_bytes())
+            .await
+            .expect("symlink must succeed");
+
+        let e = dir_node.entry("keep".as_bytes()).await.unwrap();
+        let mut subdir_node = e.directory().await.expect("directory must succeed");
+
+        let e_sub = subdir_node
+            .entry(".keep".as_bytes())
+            .await
+            .expect("subdir entry must succeed");
+        e_sub
+            .file(false, 0, &mut futures::io::Cursor::new([]))
+            .await
+            .unwrap();
+
+        // close the subdir, and then the dir, which is required.
+        subdir_node.close().await.unwrap();
+        dir_node.close().await.unwrap();
+    });
+
+    assert_eq!(include_bytes!("../tests/complicated.nar"), buf.as_slice());
+}
diff --git a/tvix/nix-compat/src/narinfo/fingerprint.rs b/tvix/nix-compat/src/narinfo/fingerprint.rs
new file mode 100644
index 000000000000..3e02aca57160
--- /dev/null
+++ b/tvix/nix-compat/src/narinfo/fingerprint.rs
@@ -0,0 +1,50 @@
+use crate::{nixbase32, store_path::StorePathRef};
+
+/// Computes the fingerprint string for certain fields in a [super::NarInfo].
+/// This fingerprint is signed by an ed25519 key, and in the case of a Nix HTTP
+/// Binary cache, included in the NARInfo files served from there.
+pub fn fingerprint<'a, R: Iterator<Item = &'a StorePathRef<'a>>>(
+    store_path: &StorePathRef,
+    nar_sha256: &[u8; 32],
+    nar_size: u64,
+    references: R,
+) -> String {
+    format!(
+        "1;{};sha256:{};{};{}",
+        store_path.to_absolute_path(),
+        nixbase32::encode(nar_sha256),
+        nar_size,
+        // references are absolute paths, joined with `,`.
+        references
+            .map(|r| r.to_absolute_path())
+            .collect::<Vec<String>>()
+            .join(",")
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::narinfo::NarInfo;
+
+    const NARINFO_STR: &str = r#"StorePath: /nix/store/syd87l2rxw8cbsxmxl853h0r6pdwhwjr-curl-7.82.0-bin
+URL: nar/05ra3y72i3qjri7xskf9qj8kb29r6naqy1sqpbs3azi3xcigmj56.nar.xz
+Compression: xz
+FileHash: sha256:05ra3y72i3qjri7xskf9qj8kb29r6naqy1sqpbs3azi3xcigmj56
+FileSize: 68852
+NarHash: sha256:1b4sb93wp679q4zx9k1ignby1yna3z7c4c2ri3wphylbc2dwsys0
+NarSize: 196040
+References: 0jqd0rlxzra1rs38rdxl43yh6rxchgc6-curl-7.82.0 6w8g7njm4mck5dmjxws0z1xnrxvl81xa-glibc-2.34-115 j5jxw3iy7bbz4a57fh9g2xm2gxmyal8h-zlib-1.2.12 yxvjs9drzsphm9pcf42a4byzj1kb9m7k-openssl-1.1.1n
+Deriver: 5rwxzi7pal3qhpsyfc16gzkh939q1np6-curl-7.82.0.drv
+Sig: cache.nixos.org-1:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==
+Sig: test1:519iiVLx/c4Rdt5DNt6Y2Jm6hcWE9+XY69ygiWSZCNGVcmOcyL64uVAJ3cV8vaTusIZdbTnYo9Y7vDNeTmmMBQ==
+"#;
+
+    #[test]
+    fn fingerprint() {
+        let parsed = NarInfo::parse(NARINFO_STR).expect("must parse");
+        assert_eq!(
+            "1;/nix/store/syd87l2rxw8cbsxmxl853h0r6pdwhwjr-curl-7.82.0-bin;sha256:1b4sb93wp679q4zx9k1ignby1yna3z7c4c2ri3wphylbc2dwsys0;196040;/nix/store/0jqd0rlxzra1rs38rdxl43yh6rxchgc6-curl-7.82.0,/nix/store/6w8g7njm4mck5dmjxws0z1xnrxvl81xa-glibc-2.34-115,/nix/store/j5jxw3iy7bbz4a57fh9g2xm2gxmyal8h-zlib-1.2.12,/nix/store/yxvjs9drzsphm9pcf42a4byzj1kb9m7k-openssl-1.1.1n",
+            parsed.fingerprint()
+        );
+    }
+}
diff --git a/tvix/nix-compat/src/narinfo/mod.rs b/tvix/nix-compat/src/narinfo/mod.rs
new file mode 100644
index 000000000000..b1c10bceb200
--- /dev/null
+++ b/tvix/nix-compat/src/narinfo/mod.rs
@@ -0,0 +1,527 @@
+//! NAR info files describe a store path in a traditional Nix binary cache.
+//! Over the wire, they are formatted as "Key: value" pairs separated by newlines.
+//!
+//! It contains four kinds of information:
+//! 1. the description of the store path itself
+//!    * store path prefix, digest, and name
+//!    * NAR hash and size
+//!    * references
+//! 2. authenticity information
+//!    * zero or more signatures over that description
+//!    * an optional [CAHash] for content-addressed paths (fixed outputs, sources, and derivations)
+//! 3. derivation metadata
+//!    * deriver (the derivation that produced this path)
+//!    * system (the system value of that derivation)
+//! 4. cache-specific information
+//!    * URL of the compressed NAR, relative to the NAR info file
+//!    * compression algorithm used for the NAR
+//!    * hash and size of the compressed NAR
+
+use bitflags::bitflags;
+use data_encoding::HEXLOWER;
+use std::{
+    fmt::{self, Display},
+    mem,
+};
+
+use crate::{nixbase32, nixhash::CAHash, store_path::StorePathRef};
+
+mod fingerprint;
+mod public_keys;
+mod signature;
+
+pub use fingerprint::fingerprint;
+
+pub use public_keys::{Error as PubKeyError, PubKey};
+pub use signature::{Error as SignatureError, Signature};
+
+#[derive(Debug)]
+pub struct NarInfo<'a> {
+    pub flags: Flags,
+    // core (authenticated, but unverified here)
+    /// Store path described by this [NarInfo]
+    pub store_path: StorePathRef<'a>,
+    /// SHA-256 digest of the NAR file
+    pub nar_hash: [u8; 32],
+    /// Size of the NAR file in bytes
+    pub nar_size: u64,
+    /// Store paths known to be referenced by the contents
+    pub references: Vec<StorePathRef<'a>>,
+    // authenticity
+    /// Ed25519 signature over the path fingerprint
+    pub signatures: Vec<Signature<'a>>,
+    /// Content address (for content-defined paths)
+    pub ca: Option<CAHash>,
+    // derivation metadata
+    /// Nix system triple of [NarInfo::deriver]
+    pub system: Option<&'a str>,
+    /// Store path of the derivation that produced this. The last .drv suffix is stripped.
+    pub deriver: Option<StorePathRef<'a>>,
+    // cache-specific untrusted metadata
+    /// Relative URL of the compressed NAR file
+    pub url: &'a str,
+    /// Compression method of the NAR file
+    /// `None` means `Compression: none`.
+    ///
+    /// Nix interprets a missing `Compression` field as `Some("bzip2")`,
+    /// so we do as well. We haven't found any examples of this in the
+    /// wild, not even in the cache.nixos.org dataset.
+    pub compression: Option<&'a str>,
+    /// SHA-256 digest of the file at `url`
+    pub file_hash: Option<[u8; 32]>,
+    /// Size of the file at `url` in bytes
+    pub file_size: Option<u64>,
+}
+
+bitflags! {
+    /// TODO(edef): be conscious of these when roundtripping
+    #[derive(Debug, Copy, Clone)]
+    pub struct Flags: u8 {
+        const UNKNOWN_FIELD = 1 << 0;
+        const COMPRESSION_DEFAULT = 1 << 1;
+        // Format quirks encountered in the cache.nixos.org dataset
+        const REFERENCES_OUT_OF_ORDER = 1 << 2;
+        const NAR_HASH_HEX = 1 << 3;
+    }
+}
+
+impl<'a> NarInfo<'a> {
+    pub fn parse(input: &'a str) -> Result<Self, Error> {
+        let mut flags = Flags::empty();
+        let mut store_path = None;
+        let mut url = None;
+        let mut compression = None;
+        let mut file_hash = None;
+        let mut file_size = None;
+        let mut nar_hash = None;
+        let mut nar_size = None;
+        let mut references = None;
+        let mut system = None;
+        let mut deriver = None;
+        let mut signatures = vec![];
+        let mut ca = None;
+
+        for line in input.lines() {
+            let (tag, val) = line
+                .split_once(':')
+                .ok_or_else(|| Error::InvalidLine(line.to_string()))?;
+
+            let val = val
+                .strip_prefix(' ')
+                .ok_or_else(|| Error::InvalidLine(line.to_string()))?;
+
+            match tag {
+                "StorePath" => {
+                    let val = val
+                        .strip_prefix("/nix/store/")
+                        .ok_or(Error::InvalidStorePath(
+                            crate::store_path::Error::MissingStoreDir,
+                        ))?;
+                    let val = StorePathRef::from_bytes(val.as_bytes())
+                        .map_err(Error::InvalidStorePath)?;
+
+                    if store_path.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "URL" => {
+                    if val.is_empty() {
+                        return Err(Error::EmptyField(tag.to_string()));
+                    }
+
+                    if url.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "Compression" => {
+                    if val.is_empty() {
+                        return Err(Error::EmptyField(tag.to_string()));
+                    }
+
+                    if compression.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "FileHash" => {
+                    let val = val
+                        .strip_prefix("sha256:")
+                        .ok_or_else(|| Error::MissingPrefixForHash(tag.to_string()))?;
+                    let val = nixbase32::decode_fixed::<32>(val)
+                        .map_err(|e| Error::UnableToDecodeHash(tag.to_string(), e))?;
+
+                    if file_hash.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "FileSize" => {
+                    let val = val
+                        .parse::<u64>()
+                        .map_err(|_| Error::UnableToParseSize(tag.to_string(), val.to_string()))?;
+
+                    if file_size.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "NarHash" => {
+                    let val = val
+                        .strip_prefix("sha256:")
+                        .ok_or_else(|| Error::MissingPrefixForHash(tag.to_string()))?;
+
+                    let val = if val.len() != HEXLOWER.encode_len(32) {
+                        nixbase32::decode_fixed::<32>(val)
+                    } else {
+                        flags |= Flags::NAR_HASH_HEX;
+
+                        let val = val.as_bytes();
+                        let mut buf = [0u8; 32];
+
+                        HEXLOWER
+                            .decode_mut(val, &mut buf)
+                            .map_err(|e| e.error)
+                            .map(|_| buf)
+                    };
+
+                    let val = val.map_err(|e| Error::UnableToDecodeHash(tag.to_string(), e))?;
+
+                    if nar_hash.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "NarSize" => {
+                    let val = val
+                        .parse::<u64>()
+                        .map_err(|_| Error::UnableToParseSize(tag.to_string(), val.to_string()))?;
+
+                    if nar_size.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "References" => {
+                    let val: Vec<StorePathRef> = if !val.is_empty() {
+                        let mut prev = "";
+                        val.split(' ')
+                            .enumerate()
+                            .map(|(i, s)| {
+                                // TODO(edef): track *duplicates* if this occurs
+                                if mem::replace(&mut prev, s) >= s {
+                                    flags |= Flags::REFERENCES_OUT_OF_ORDER;
+                                }
+
+                                StorePathRef::from_bytes(s.as_bytes())
+                                    .map_err(|err| Error::InvalidReference(i, err))
+                            })
+                            .collect::<Result<_, _>>()?
+                    } else {
+                        vec![]
+                    };
+
+                    if references.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "System" => {
+                    if val.is_empty() {
+                        return Err(Error::EmptyField(tag.to_string()));
+                    }
+
+                    if system.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                "Deriver" => {
+                    match val.strip_suffix(".drv") {
+                        Some(val) => {
+                            let val = StorePathRef::from_bytes(val.as_bytes())
+                                .map_err(Error::InvalidDeriverStorePath)?;
+
+                            if deriver.replace(val).is_some() {
+                                return Err(Error::DuplicateField(tag.to_string()));
+                            }
+                        }
+                        None => {
+                            return Err(Error::InvalidDeriverStorePathMissingSuffix);
+                        }
+                    };
+                }
+                "Sig" => {
+                    let val = Signature::parse(val)
+                        .map_err(|e| Error::UnableToParseSignature(signatures.len(), e))?;
+
+                    signatures.push(val);
+                }
+                "CA" => {
+                    let val = CAHash::from_nix_hex_str(val)
+                        .ok_or_else(|| Error::UnableToParseCA(val.to_string()))?;
+
+                    if ca.replace(val).is_some() {
+                        return Err(Error::DuplicateField(tag.to_string()));
+                    }
+                }
+                _ => {
+                    flags |= Flags::UNKNOWN_FIELD;
+                }
+            }
+        }
+
+        Ok(NarInfo {
+            store_path: store_path.ok_or(Error::MissingField("StorePath"))?,
+            nar_hash: nar_hash.ok_or(Error::MissingField("NarHash"))?,
+            nar_size: nar_size.ok_or(Error::MissingField("NarSize"))?,
+            references: references.ok_or(Error::MissingField("References"))?,
+            signatures,
+            ca,
+            system,
+            deriver,
+            url: url.ok_or(Error::MissingField("URL"))?,
+            compression: match compression {
+                Some("none") => None,
+                None => {
+                    flags |= Flags::COMPRESSION_DEFAULT;
+                    Some("bzip2")
+                }
+                _ => compression,
+            },
+            file_hash,
+            file_size,
+            flags,
+        })
+    }
+
+    /// Computes the fingerprint string for certain fields in this [NarInfo].
+    /// This fingerprint is signed in [self.signatures].
+    pub fn fingerprint(&self) -> String {
+        fingerprint(
+            &self.store_path,
+            &self.nar_hash,
+            self.nar_size,
+            self.references.iter(),
+        )
+    }
+}
+
+impl Display for NarInfo<'_> {
+    fn fmt(&self, w: &mut fmt::Formatter) -> fmt::Result {
+        writeln!(w, "StorePath: /nix/store/{}", self.store_path)?;
+        writeln!(w, "URL: {}", self.url)?;
+
+        if let Some(compression) = self.compression {
+            writeln!(w, "Compression: {compression}")?;
+        }
+
+        if let Some(file_hash) = self.file_hash {
+            writeln!(w, "FileHash: sha256:{}", nixbase32::encode(&file_hash),)?;
+        }
+
+        if let Some(file_size) = self.file_size {
+            writeln!(w, "FileSize: {file_size}")?;
+        }
+
+        writeln!(w, "NarHash: sha256:{}", nixbase32::encode(&self.nar_hash),)?;
+        writeln!(w, "NarSize: {}", self.nar_size)?;
+
+        write!(w, "References:")?;
+        if self.references.is_empty() {
+            write!(w, " ")?;
+        } else {
+            for path in &self.references {
+                write!(w, " {path}")?;
+            }
+        }
+        writeln!(w)?;
+
+        if let Some(deriver) = &self.deriver {
+            writeln!(w, "Deriver: {deriver}.drv")?;
+        }
+
+        if let Some(system) = self.system {
+            writeln!(w, "System: {system}")?;
+        }
+
+        for sig in &self.signatures {
+            writeln!(w, "Sig: {sig}")?;
+        }
+
+        if let Some(ca) = &self.ca {
+            writeln!(w, "CA: {}", ca.to_nix_nixbase32_string())?;
+        }
+
+        Ok(())
+    }
+}
+
+#[derive(thiserror::Error, Debug)]
+pub enum Error {
+    #[error("duplicate field: {0}")]
+    DuplicateField(String),
+
+    #[error("missing field: {0}")]
+    MissingField(&'static str),
+
+    #[error("invalid line: {0}")]
+    InvalidLine(String),
+
+    #[error("invalid StorePath: {0}")]
+    InvalidStorePath(crate::store_path::Error),
+
+    #[error("field {0} may not be empty string")]
+    EmptyField(String),
+
+    #[error("invalid {0}: {1}")]
+    UnableToParseSize(String, String),
+
+    #[error("unable to parse #{0} reference: {1}")]
+    InvalidReference(usize, crate::store_path::Error),
+
+    #[error("invalid Deriver store path: {0}")]
+    InvalidDeriverStorePath(crate::store_path::Error),
+
+    #[error("invalid Deriver store path, must end with .drv")]
+    InvalidDeriverStorePathMissingSuffix,
+
+    #[error("missing prefix for {0}")]
+    MissingPrefixForHash(String),
+
+    #[error("unable to decode {0}: {1}")]
+    UnableToDecodeHash(String, data_encoding::DecodeError),
+
+    #[error("unable to parse signature #{0}: {1}")]
+    UnableToParseSignature(usize, SignatureError),
+
+    #[error("unable to parse CA field: {0}")]
+    UnableToParseCA(String),
+}
+
+#[cfg(test)]
+mod test {
+    use hex_literal::hex;
+    use lazy_static::lazy_static;
+    use pretty_assertions::assert_eq;
+    use std::{io, str};
+
+    use crate::{
+        nixhash::{CAHash, NixHash},
+        store_path::StorePathRef,
+    };
+
+    use super::{Flags, NarInfo};
+
+    lazy_static! {
+        static ref CASES: &'static [&'static str] = {
+            let data = zstd::decode_all(io::Cursor::new(include_bytes!(
+                "../../testdata/narinfo.zst"
+            )))
+            .unwrap();
+            let data = str::from_utf8(Vec::leak(data)).unwrap();
+            Vec::leak(
+                data.split_inclusive("\n\n")
+                    .map(|s| s.strip_suffix('\n').unwrap())
+                    .collect::<Vec<_>>(),
+            )
+        };
+    }
+
+    #[test]
+    fn roundtrip() {
+        for &input in *CASES {
+            let parsed = NarInfo::parse(input).expect("should parse");
+            let output = format!("{parsed}");
+            assert_eq!(input, output, "should roundtrip");
+        }
+    }
+
+    #[test]
+    fn references_out_of_order() {
+        let parsed = NarInfo::parse(
+            r#"StorePath: /nix/store/xi429w4ddvb1r77978hm7jfb2jsn559r-gcc-3.4.6
+URL: nar/1hr09cgkyw1hcsfkv5qp5jlpmf2mqrkrqs3xj5zklq9c1h9544ff.nar.bz2
+Compression: bzip2
+FileHash: sha256:1hr09cgkyw1hcsfkv5qp5jlpmf2mqrkrqs3xj5zklq9c1h9544ff
+FileSize: 4006
+NarHash: sha256:0ik9mpqxpd9hv325hdblj2nawqj5w7951qdyy8ikxgwr6fq7m11c
+NarSize: 21264
+References: a8922c0h87iilxzzvwn2hmv8x210aqb9-glibc-2.7 7w2acjgalb0cm7b3bg8yswza4l7iil9y-binutils-2.18 mm631h09mj964hm9q04l5fd8vw12j1mm-bash-3.2-p39 nx2zs2qd6snfcpzw4a0jnh26z9m0yihz-gcc-3.4.6 xi429w4ddvb1r77978hm7jfb2jsn559r-gcc-3.4.6
+Deriver: 2dzpn70c1hawczwhg9aavqk18zp9zsva-gcc-3.4.6.drv
+Sig: cache.nixos.org-1:o1DTsjCz0PofLJ216P2RBuSulI8BAb6zHxWE4N+tzlcELk5Uk/GO2SCxWTRN5wJutLZZ+cHTMdWqOHF88KGQDg==
+"#).expect("should parse");
+
+        assert!(parsed.flags.contains(Flags::REFERENCES_OUT_OF_ORDER));
+        assert_eq!(
+            vec![
+                "a8922c0h87iilxzzvwn2hmv8x210aqb9-glibc-2.7",
+                "7w2acjgalb0cm7b3bg8yswza4l7iil9y-binutils-2.18",
+                "mm631h09mj964hm9q04l5fd8vw12j1mm-bash-3.2-p39",
+                "nx2zs2qd6snfcpzw4a0jnh26z9m0yihz-gcc-3.4.6",
+                "xi429w4ddvb1r77978hm7jfb2jsn559r-gcc-3.4.6"
+            ],
+            parsed
+                .references
+                .iter()
+                .map(StorePathRef::to_string)
+                .collect::<Vec<_>>(),
+        );
+    }
+
+    #[test]
+    fn ca_nar_hash_sha1() {
+        let parsed = NarInfo::parse(
+            r#"StorePath: /nix/store/k20pahypzvr49fy82cw5sx72hdfg3qcr-texlive-hyphenex-37354
+URL: nar/0i5biw0g01514llhfswxy6xfav8lxxdq1xg6ik7hgsqbpw0f06yi.nar.xz
+Compression: xz
+FileHash: sha256:0i5biw0g01514llhfswxy6xfav8lxxdq1xg6ik7hgsqbpw0f06yi
+FileSize: 7120
+NarHash: sha256:0h1bm4sj1cnfkxgyhvgi8df1qavnnv94sd0v09wcrm971602shfg
+NarSize: 22552
+References: 
+Sig: cache.nixos.org-1:u01BybwQhyI5H1bW1EIWXssMDhDDIvXOG5uh8Qzgdyjz6U1qg6DHhMAvXZOUStIj6X5t4/ufFgR8i3fjf0bMAw==
+CA: fixed:r:sha1:1ak1ymbmsfx7z8kh09jzkr3a4dvkrfjw
+"#).expect("should parse");
+
+        assert_eq!(
+            parsed.ca,
+            Some(CAHash::Nar(NixHash::Sha1(hex!(
+                "5cba3c77236ae4f9650270a27fbad375551fa60a"
+            ))))
+        );
+    }
+
+    #[test]
+    fn compression_default() {
+        // This doesn't exist as such in cache.nixos.org.
+        // We explicitly removed the compression field for the sake of this test.
+        let parsed = NarInfo::parse(r#"StorePath: /nix/store/a1jjalr4csx9hcga7fnm122aqabrjnch-digikam-2.6.0
+URL: nar/1fzimfnvq2k8b40n4g54abmncpx2ddckh6qlb77pgq6xiysyil69.nar.bz2
+FileHash: sha256:1fzimfnvq2k8b40n4g54abmncpx2ddckh6qlb77pgq6xiysyil69
+FileSize: 43503778
+NarHash: sha256:0zpbbwipqzr5p8mlpag9wrsp5hlaxkq7gax5jj0hg3vvdziypcw5
+NarSize: 100658640
+References: 0izkyk7bq2ag9393nvnhgm87p75cq09w-liblqr-1-0.4.1 1cslpgyb7vb30inj3210jv6agqv42jxz-qca-2.0.3 1sya3bwjxkzpkmwn67gfzp4gz4g62l36-libXrandr-1.3.1 26yxdaa9z0ma5sgw02i670rsqnl57crs-glib-2.30.3 27lnjh99236kmhbpc5747599zcymfzmg-qt-4.8.2 2v6x378vcfvyxilkvihs60zha54z2x2y-qjson-0.7.1 45hgr3fbnr45n795hn2x7hsymp0h2j2m-libjpeg-8c 4kw1b212s80ap2iyibxrimcqb5imhfj7-libkexiv2-4.7.4 7dvylm5crlc0sfafcc0n46mb5ch67q0j-glibc-2.13 a05cbh1awjbl1rbyb2ynyf4k42v5a9a7-boost-1.47.0 a1jjalr4csx9hcga7fnm122aqabrjnch-digikam-2.6.0 aav5ffg8wlnilgnvdb2jnrv2aam4zmmz-perl-5.14.2 ab0m9h30nsr13w48qriv0k350kmwx567-kdelibs-4.7.4 avffkd49cqvpwdkzry8bn69dkbw4cy29-lensfun-0.2.5 cy8rl8h4yp2j3h8987vkklg328q3wmjz-gcc-4.6.3 dmmh5ihyg1r2dm4azgsfj2kprj92czlg-libSM-1.2.0 fl56j5n4shfw9c0r6vs2i4f1h9zx5kac-soprano-2.7.6 g15cmvh15ggdjcwapskngv20q4yhix40-jasper-1.900.1 i04maxd0din6v92rnqcwl9yra0kl2vk5-marble-4.7.4 kqjjb3m26rdddwwwkk8v45821aps877k-libICE-1.0.7 lxz9r135wkndvi642z4bjgmvyypsgirb-libtiff-3.9.4 m9c8i0a6cl30lcqp654dqkbag3wjmd00-libX11-1.4.1 mpnj4k2ijrgyfkh48fg96nzcmklfh5pl-coreutils-8.15 nppljblap477s0893c151lyq7r7n5v1q-zlib-1.2.7 nw9mdbyp8kyn3v4vkdzq0gsnqbc4mnx3-expat-2.0.1 p1a0dn931mzdkvj6h5yzshbmgxba5r0z-libgphoto2-2.4.11 pvjj07xa1cfkad3gwk376nzdrgknbcqm-mesa-7.11.2 pzcxag98jqccp9ycbxknyh0w95pgnsk4-lcms-1.19 qfi5pgds33kg6vlnxsmj0hyl74vcmyiz-libpng-1.5.10 scm6bj86s3qh3s3x0b9ayjp6755p4q86-mysql-5.1.54 sd23qspcyg385va0lr35xgz3hvlqphg6-libkipi-4.7.4 svmbrhc6kzfzakv20a7zrfl6kbr5mfpq-kdepimlibs-4.7.4 v7kh3h7xfwjz4hgffg3gwrfzjff9bw9d-bash-4.2-p24 vi17f22064djgpk0w248da348q8gxkww-libkdcraw-4.7.4 wkjdzmj3z4dcbsc9f833zs6krdgg2krk-phonon-4.6.0 xf3i3awqi0035ixy2qyb6hk4c92r3vrn-opencv-2.4.2 y1vr0nz8i59x59501020nh2k1dw3bhwq-libusb-0.1.12 yf3hin2hb6i08n7zrk8g3acy54rhg9bp-libXext-1.2.0
+Deriver: la77dr44phk5m5jnl4dvk01cwpykyw9s-digikam-2.6.0.drv
+System: i686-linux
+Sig: cache.nixos.org-1:92fl0i5q7EyegCj5Yf4L0bENkWuVAtgveiRcTEEUH0P6HvCE1xFcPbz/0Pf6Np+K1LPzHK+s5RHOmVoxRsvsDg==
+"#).expect("should parse");
+
+        assert!(parsed.flags.contains(Flags::COMPRESSION_DEFAULT));
+        assert_eq!(parsed.compression, Some("bzip2"));
+    }
+
+    #[test]
+    fn nar_hash_hex() {
+        let parsed = NarInfo::parse(r#"StorePath: /nix/store/0vpqfxbkx0ffrnhbws6g9qwhmliksz7f-perl-HTTP-Cookies-6.01
+URL: nar/1rv1m9inydm1r4krw8hmwg1hs86d0nxddd1pbhihx7l7fycjvfk3.nar.xz
+Compression: xz
+FileHash: sha256:1rv1m9inydm1r4krw8hmwg1hs86d0nxddd1pbhihx7l7fycjvfk3
+FileSize: 19912
+NarHash: sha256:60adfd293a4d81ad7cd7e47263cbb3fc846309ef91b154a08ba672b558f94ff3
+NarSize: 45840
+References: 0vpqfxbkx0ffrnhbws6g9qwhmliksz7f-perl-HTTP-Cookies-6.01 9vrhbib2lxd9pjlg6fnl5b82gblidrcr-perl-HTTP-Message-6.06 wy20zslqxzxxfpzzk0rajh41d7a6mlnf-perl-HTTP-Date-6.02
+Deriver: fb4ihlq3psnsjq95mvvs49rwpplpc8zj-perl-HTTP-Cookies-6.01.drv
+Sig: cache.nixos.org-1:HhaiY36Uk3XV1JGe9d9xHnzAapqJXprU1YZZzSzxE97jCuO5RR7vlG2kF7MSC5thwRyxAtdghdSz3AqFi+QSCw==
+"#).expect("should parse");
+
+        assert!(parsed.flags.contains(Flags::NAR_HASH_HEX));
+        assert_eq!(
+            hex!("60adfd293a4d81ad7cd7e47263cbb3fc846309ef91b154a08ba672b558f94ff3"),
+            parsed.nar_hash,
+        );
+    }
+}
diff --git a/tvix/nix-compat/src/narinfo/public_keys.rs b/tvix/nix-compat/src/narinfo/public_keys.rs
new file mode 100644
index 000000000000..ced05cadb133
--- /dev/null
+++ b/tvix/nix-compat/src/narinfo/public_keys.rs
@@ -0,0 +1,149 @@
+//! This module defines data structures and parsers for the public key format
+//! used inside Nix to verify signatures on .narinfo files.
+
+use std::fmt::Display;
+
+use data_encoding::BASE64;
+use ed25519_dalek::{VerifyingKey, PUBLIC_KEY_LENGTH};
+
+use super::Signature;
+
+/// This represents a ed25519 public key and "name".
+/// These are normally passed in the `trusted-public-keys` Nix config option,
+/// and consist of a name and base64-encoded ed25519 pubkey, separated by a `:`.
+#[derive(Debug)]
+pub struct PubKey {
+    name: String,
+    verifying_key: VerifyingKey,
+}
+
+impl PubKey {
+    pub fn new(name: String, verifying_key: VerifyingKey) -> Self {
+        Self {
+            name,
+            verifying_key,
+        }
+    }
+
+    pub fn parse(input: &str) -> Result<Self, Error> {
+        let (name, bytes64) = input.split_once(':').ok_or(Error::MissingSeparator)?;
+
+        if name.is_empty()
+            || !name
+                .chars()
+                .all(|c| char::is_alphanumeric(c) || c == '-' || c == '.')
+        {
+            return Err(Error::InvalidName(name.to_string()));
+        }
+
+        if bytes64.len() != BASE64.encode_len(PUBLIC_KEY_LENGTH) {
+            return Err(Error::InvalidPubKeyLen(bytes64.len()));
+        }
+
+        let mut buf = [0; PUBLIC_KEY_LENGTH + 1];
+        let mut bytes = [0; PUBLIC_KEY_LENGTH];
+        match BASE64.decode_mut(bytes64.as_bytes(), &mut buf) {
+            Ok(PUBLIC_KEY_LENGTH) => {
+                bytes.copy_from_slice(&buf[..PUBLIC_KEY_LENGTH]);
+            }
+            Ok(_) => unreachable!(),
+            // keeping DecodePartial gets annoying lifetime-wise
+            Err(_) => return Err(Error::DecodeError(input.to_string())),
+        }
+
+        let verifying_key = VerifyingKey::from_bytes(&bytes).map_err(Error::InvalidVerifyingKey)?;
+
+        Ok(Self {
+            name: name.to_string(),
+            verifying_key,
+        })
+    }
+
+    pub fn name(&self) -> &str {
+        &self.name
+    }
+
+    /// Verify the passed in signature is a correct signature for the passed in fingerprint and is signed
+    /// by the key material referred to by [Self],
+    /// which means the name in the signature has to match,
+    /// and the signature bytes themselves need to be a valid signature made by
+    /// the signing key identified by [Self::verifying key].
+    pub fn verify(&self, fingerprint: &str, signature: &Signature) -> bool {
+        if self.name() != signature.name() {
+            return false;
+        }
+
+        return signature.verify(fingerprint.as_bytes(), &self.verifying_key);
+    }
+}
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("Invalid name: {0}")]
+    InvalidName(String),
+    #[error("Missing separator")]
+    MissingSeparator,
+    #[error("Invalid pubkey len: {0}")]
+    InvalidPubKeyLen(usize),
+    #[error("VerifyingKey error: {0}")]
+    InvalidVerifyingKey(ed25519_dalek::SignatureError),
+    #[error("Unable to base64-decode pubkey: {0}")]
+    DecodeError(String),
+}
+
+impl Display for PubKey {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}:{}",
+            self.name,
+            BASE64.encode(self.verifying_key.as_bytes())
+        )
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use data_encoding::BASE64;
+    use ed25519_dalek::PUBLIC_KEY_LENGTH;
+    use test_case::test_case;
+
+    use crate::narinfo::Signature;
+
+    use super::PubKey;
+    const FINGERPRINT: &str = "1;/nix/store/syd87l2rxw8cbsxmxl853h0r6pdwhwjr-curl-7.82.0-bin;sha256:1b4sb93wp679q4zx9k1ignby1yna3z7c4c2ri3wphylbc2dwsys0;196040;/nix/store/0jqd0rlxzra1rs38rdxl43yh6rxchgc6-curl-7.82.0,/nix/store/6w8g7njm4mck5dmjxws0z1xnrxvl81xa-glibc-2.34-115,/nix/store/j5jxw3iy7bbz4a57fh9g2xm2gxmyal8h-zlib-1.2.12,/nix/store/yxvjs9drzsphm9pcf42a4byzj1kb9m7k-openssl-1.1.1n";
+
+    #[test_case("cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=", "cache.nixos.org-1", BASE64.decode(b"6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=").unwrap()[..].try_into().unwrap(); "cache.nixos.org")]
+    #[test_case("cheesecake:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=", "cheesecake", BASE64.decode(b"6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=").unwrap()[..].try_into().unwrap(); "cache.nixos.org different name")]
+    #[test_case("test1:tLAEn+EeaBUJYqEpTd2yeerr7Ic6+0vWe+aXL/vYUpE=", "test1", BASE64.decode(b"tLAEn+EeaBUJYqEpTd2yeerr7Ic6+0vWe+aXL/vYUpE=").unwrap()[..].try_into().unwrap(); "test-1")]
+    fn parse(
+        input: &'static str,
+        exp_name: &'static str,
+        exp_verifying_key_bytes: &[u8; PUBLIC_KEY_LENGTH],
+    ) {
+        let pubkey = PubKey::parse(input).expect("must parse");
+        assert_eq!(exp_name, pubkey.name());
+        assert_eq!(exp_verifying_key_bytes, pubkey.verifying_key.as_bytes());
+    }
+
+    #[test_case("6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY="; "empty name")]
+    #[test_case("cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY"; "missing padding")]
+    #[test_case("cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDS"; "wrong length")]
+    fn parse_fail(input: &'static str) {
+        PubKey::parse(input).expect_err("must fail");
+    }
+
+    #[test_case("cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=", FINGERPRINT, "cache.nixos.org-1:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", true; "correct cache.nixos.org")]
+    #[test_case("cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=", FINGERPRINT, "cache.nixos.org:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", false; "wrong name mismatch")]
+    fn verify(
+        pubkey_str: &'static str,
+        fingerprint: &'static str,
+        signature_str: &'static str,
+        expected: bool,
+    ) {
+        let pubkey = PubKey::parse(pubkey_str).expect("must parse");
+        let signature = Signature::parse(signature_str).expect("must parse");
+
+        assert_eq!(expected, pubkey.verify(fingerprint, &signature));
+    }
+}
diff --git a/tvix/nix-compat/src/narinfo/signature.rs b/tvix/nix-compat/src/narinfo/signature.rs
new file mode 100644
index 000000000000..1f4a90c26c2e
--- /dev/null
+++ b/tvix/nix-compat/src/narinfo/signature.rs
@@ -0,0 +1,132 @@
+use std::fmt::{self, Display};
+
+use data_encoding::BASE64;
+use ed25519_dalek::SIGNATURE_LENGTH;
+
+#[derive(Debug)]
+pub struct Signature<'a> {
+    name: &'a str,
+    bytes: [u8; SIGNATURE_LENGTH],
+}
+
+impl<'a> Signature<'a> {
+    pub fn new(name: &'a str, bytes: [u8; SIGNATURE_LENGTH]) -> Self {
+        Self { name, bytes }
+    }
+
+    pub fn parse(input: &'a str) -> Result<Self, Error> {
+        let (name, bytes64) = input.split_once(':').ok_or(Error::MissingSeparator)?;
+
+        if name.is_empty()
+            || !name
+                .chars()
+                .all(|c| char::is_alphanumeric(c) || c == '-' || c == '.')
+        {
+            return Err(Error::InvalidName(name.to_string()));
+        }
+
+        if bytes64.len() != BASE64.encode_len(SIGNATURE_LENGTH) {
+            return Err(Error::InvalidSignatureLen(bytes64.len()));
+        }
+
+        let mut bytes = [0; SIGNATURE_LENGTH];
+        let mut buf = [0; SIGNATURE_LENGTH + 2];
+        match BASE64.decode_mut(bytes64.as_bytes(), &mut buf) {
+            Ok(SIGNATURE_LENGTH) => bytes.copy_from_slice(&buf[..SIGNATURE_LENGTH]),
+            Ok(_) => unreachable!(),
+            // keeping DecodePartial gets annoying lifetime-wise
+            Err(_) => return Err(Error::DecodeError(input.to_string())),
+        }
+
+        Ok(Signature { name, bytes })
+    }
+
+    pub fn name(&self) -> &'a str {
+        self.name
+    }
+
+    pub fn bytes(&self) -> &[u8; SIGNATURE_LENGTH] {
+        &self.bytes
+    }
+
+    /// For a given fingerprint and ed25519 verifying key, ensure if the signature is valid.
+    pub fn verify(&self, fingerprint: &[u8], verifying_key: &ed25519_dalek::VerifyingKey) -> bool {
+        let signature = ed25519_dalek::Signature::from_bytes(self.bytes());
+
+        verifying_key.verify_strict(fingerprint, &signature).is_ok()
+    }
+}
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("Invalid name: {0}")]
+    InvalidName(String),
+    #[error("Missing separator")]
+    MissingSeparator,
+    #[error("Invalid signature len: (expected {} b64-encoded, got {}", BASE64.encode_len(SIGNATURE_LENGTH), .0)]
+    InvalidSignatureLen(usize),
+    #[error("Unable to base64-decode signature: {0}")]
+    DecodeError(String),
+}
+
+impl Display for Signature<'_> {
+    fn fmt(&self, w: &mut fmt::Formatter) -> fmt::Result {
+        write!(w, "{}:{}", self.name, BASE64.encode(&self.bytes))
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use data_encoding::BASE64;
+    use ed25519_dalek::VerifyingKey;
+    use lazy_static::lazy_static;
+
+    use super::Signature;
+    use test_case::test_case;
+
+    const FINGERPRINT: &str = "1;/nix/store/syd87l2rxw8cbsxmxl853h0r6pdwhwjr-curl-7.82.0-bin;sha256:1b4sb93wp679q4zx9k1ignby1yna3z7c4c2ri3wphylbc2dwsys0;196040;/nix/store/0jqd0rlxzra1rs38rdxl43yh6rxchgc6-curl-7.82.0,/nix/store/6w8g7njm4mck5dmjxws0z1xnrxvl81xa-glibc-2.34-115,/nix/store/j5jxw3iy7bbz4a57fh9g2xm2gxmyal8h-zlib-1.2.12,/nix/store/yxvjs9drzsphm9pcf42a4byzj1kb9m7k-openssl-1.1.1n";
+
+    // The signing key labelled as `cache.nixos.org-1`,
+    lazy_static! {
+        static ref PUB_CACHE_NIXOS_ORG_1: VerifyingKey = ed25519_dalek::VerifyingKey::from_bytes(
+            BASE64
+                .decode(b"6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=")
+                .unwrap()[..]
+                .try_into()
+                .unwrap()
+        )
+        .unwrap();
+        static ref PUB_TEST_1: VerifyingKey = ed25519_dalek::VerifyingKey::from_bytes(
+            BASE64
+                .decode(b"tLAEn+EeaBUJYqEpTd2yeerr7Ic6+0vWe+aXL/vYUpE=")
+                .unwrap()[..]
+                .try_into()
+                .unwrap()
+        )
+        .unwrap();
+    }
+
+    #[test_case(&PUB_CACHE_NIXOS_ORG_1, &"cache.nixos.org-1:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", FINGERPRINT, true; "valid cache.nixos.org-1")]
+    #[test_case(&PUB_CACHE_NIXOS_ORG_1, &"cache.nixos.org-1:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", FINGERPRINT, true; "valid test1")]
+    #[test_case(&PUB_CACHE_NIXOS_ORG_1, &"cache.nixos.org-2:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", FINGERPRINT, true; "valid cache.nixos.org different name")]
+    #[test_case(&PUB_CACHE_NIXOS_ORG_1, &"cache.nixos.org-1:TsTTb000000000000000000000000ytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", FINGERPRINT, false; "fail invalid cache.nixos.org-1 signature")]
+    #[test_case(&PUB_CACHE_NIXOS_ORG_1, &"cache.nixos.org-1:TsTTb3WGTZKphvYdBHXwo6weVILmTytUjLB+vcX89fOjjRicCHmKA4RCPMVLkj6TMJ4GMX3HPVWRdD1hkeKZBQ==", &FINGERPRINT[0..5], false; "fail valid sig but wrong fp cache.nixos.org-1")]
+    fn verify_sigs(
+        verifying_key: &VerifyingKey,
+        sig_str: &'static str,
+        fp: &str,
+        expect_valid: bool,
+    ) {
+        let sig = Signature::parse(sig_str).expect("must parse");
+        assert_eq!(expect_valid, sig.verify(fp.as_bytes(), verifying_key));
+    }
+
+    #[test_case("cache.nixos.org-1:o1DTsjCz0PofLJ216P2RBuSulI8BAb6zHxWE4N+tzlcELk5Uk/GO2SCxWTRN5wJutLZZ+cHTMdWqOHF8"; "wrong length")]
+    #[test_case("test\n:u01BybwQhyI5H1bW1EIWXssMDhDDIvXOG5uh8Qzgdyjz6U1qg6DHhMAvXZOUStIj6X5t4/ufFgR8i3fjf0bMAw=="; "wrong name newline")]
+    #[test_case("test :u01BybwQhyI5H1bW1EIWXssMDhDDIvXOG5uh8Qzgdyjz6U1qg6DHhMAvXZOUStIj6X5t4/ufFgR8i3fjf0bMAw=="; "wrong name space")]
+    #[test_case(":u01BybwQhyI5H1bW1EIWXssMDhDDIvXOG5uh8Qzgdyjz6U1qg6DHhMAvXZOUStIj6X5t4/ufFgR8i3fjf0bMAw=="; "empty name")]
+    #[test_case("u01BybwQhyI5H1bW1EIWXssMDhDDIvXOG5uh8Qzgdyjz6U1qg6DHhMAvXZOUStIj6X5t4/ufFgR8i3fjf0bMAw=="; "b64 only")]
+    fn parse_fail(input: &'static str) {
+        Signature::parse(input).expect_err("must fail");
+    }
+}
diff --git a/tvix/nix-compat/src/nixbase32.rs b/tvix/nix-compat/src/nixbase32.rs
new file mode 100644
index 000000000000..c4c2f2b96761
--- /dev/null
+++ b/tvix/nix-compat/src/nixbase32.rs
@@ -0,0 +1,203 @@
+//! Implements the slightly odd "base32" encoding that's used in Nix.
+//!
+//! Nix uses a custom alphabet. Contrary to other implementations (RFC4648),
+//! encoding to "nix base32" doesn't use any padding, and reads in characters
+//! in reverse order.
+//!
+//! This is also the main reason why we can't use `data_encoding::Encoding` -
+//! it gets things wrong if there normally would be a need for padding.
+
+use std::fmt::Write;
+
+use data_encoding::{DecodeError, DecodeKind};
+
+const ALPHABET: &[u8; 32] = b"0123456789abcdfghijklmnpqrsvwxyz";
+
+/// Returns encoded input
+pub fn encode(input: &[u8]) -> String {
+    let output_len = encode_len(input.len());
+    let mut output = String::with_capacity(output_len);
+
+    for n in (0..output_len).rev() {
+        let b = n * 5; // bit offset within the entire input
+        let i = b / 8; // input byte index
+        let j = b % 8; // bit offset within that input byte
+
+        // 5-bit words aren't aligned to bytes
+        // we can only read byte-aligned units
+        // read 16 bits then shift and mask to 5
+        let c = {
+            let mut word = input[i] as u16;
+            if let Some(&msb) = input.get(i + 1) {
+                word |= (msb as u16) << 8;
+            }
+            (word >> j) & 0x1f
+        };
+
+        output.write_char(ALPHABET[c as usize] as char).unwrap();
+    }
+
+    output
+}
+
+/// This maps a nixbase32-encoded character to its binary representation, which
+/// is also the index of the character in the alphabet. Invalid characters are
+/// mapped to 0xFF, which is itself an invalid value.
+const BASE32_ORD: [u8; 256] = {
+    let mut ord = [0xFF; 256];
+    let mut alphabet = ALPHABET.as_slice();
+    let mut i = 0;
+
+    while let &[c, ref tail @ ..] = alphabet {
+        ord[c as usize] = i;
+        alphabet = tail;
+        i += 1;
+    }
+
+    ord
+};
+
+/// Returns decoded input
+pub fn decode(input: impl AsRef<[u8]>) -> Result<Vec<u8>, DecodeError> {
+    let input = input.as_ref();
+
+    let output_len = decode_len(input.len());
+    let mut output: Vec<u8> = vec![0x00; output_len];
+
+    decode_inner(input, &mut output)?;
+    Ok(output)
+}
+
+pub fn decode_fixed<const K: usize>(input: impl AsRef<[u8]>) -> Result<[u8; K], DecodeError> {
+    let input = input.as_ref();
+
+    if input.len() != encode_len(K) {
+        return Err(DecodeError {
+            position: input.len().min(encode_len(K)),
+            kind: DecodeKind::Length,
+        });
+    }
+
+    let mut output = [0; K];
+    decode_inner(input, &mut output)?;
+    Ok(output)
+}
+
+fn decode_inner(input: &[u8], output: &mut [u8]) -> Result<(), DecodeError> {
+    // loop over all characters in reverse, and keep the iteration count in n.
+    let mut carry = 0;
+    let mut mask = 0;
+    for (n, &c) in input.iter().rev().enumerate() {
+        let b = n * 5;
+        let i = b / 8;
+        let j = b % 8;
+
+        let digit = BASE32_ORD[c as usize];
+        let value = (digit as u16) << j;
+        output[i] |= value as u8 | carry;
+        carry = (value >> 8) as u8;
+
+        mask |= digit;
+    }
+
+    if mask == 0xFF {
+        return Err(DecodeError {
+            position: find_invalid(input),
+            kind: DecodeKind::Symbol,
+        });
+    }
+
+    // if we're at the end, but have a nonzero carry, the encoding is invalid.
+    if carry != 0 {
+        return Err(DecodeError {
+            position: 0,
+            kind: DecodeKind::Trailing,
+        });
+    }
+
+    Ok(())
+}
+
+fn find_invalid(input: &[u8]) -> usize {
+    for (i, &c) in input.iter().enumerate() {
+        if !ALPHABET.contains(&c) {
+            return i;
+        }
+    }
+
+    unreachable!()
+}
+
+/// Returns the decoded length of an input of length len.
+pub const fn decode_len(len: usize) -> usize {
+    (len * 5) / 8
+}
+
+/// Returns the encoded length of an input of length len
+pub const fn encode_len(len: usize) -> usize {
+    (len * 8 + 4) / 5
+}
+
+#[cfg(test)]
+mod tests {
+    use hex_literal::hex;
+    use test_case::test_case;
+
+    #[test_case("", &[]; "empty bytes")]
+    #[test_case("0z", &hex!("1f"); "one byte")]
+    #[test_case("00bgd045z0d4icpbc2yyz4gx48ak44la", &hex!("8a12321522fd91efbd60ebb2481af88580f61600"); "store path")]
+    #[test_case("0c5b8vw40dy178xlpddw65q9gf1h2186jcc3p4swinwggbllv8mk", &hex!("b3a24de97a8fdbc835b9833169501030b8977031bcb54b3b3ac13740f846ab30"); "sha256")]
+    fn encode(enc: &str, dec: &[u8]) {
+        assert_eq!(enc, super::encode(dec));
+    }
+
+    #[test_case("", Some(&[]) ; "empty bytes")]
+    #[test_case("0z", Some(&hex!("1f")); "one byte")]
+    #[test_case("00bgd045z0d4icpbc2yyz4gx48ak44la", Some(&hex!("8a12321522fd91efbd60ebb2481af88580f61600")); "store path")]
+    #[test_case("0c5b8vw40dy178xlpddw65q9gf1h2186jcc3p4swinwggbllv8mk", Some(&hex!("b3a24de97a8fdbc835b9833169501030b8977031bcb54b3b3ac13740f846ab30")); "sha256")]
+    // this is invalid encoding, because it encodes 10 1-bits, so the carry
+    // would be 2 1-bits
+    #[test_case("zz", None; "invalid encoding-1")]
+    // this is an even more specific example - it'd decode as 00000000 11
+    #[test_case("c0", None; "invalid encoding-2")]
+
+    fn decode(enc: &str, dec: Option<&[u8]>) {
+        match dec {
+            Some(dec) => {
+                // The decode needs to match what's passed in dec
+                assert_eq!(dec, super::decode(enc).unwrap());
+            }
+            None => {
+                // the decode needs to be an error
+                assert!(super::decode(enc).is_err());
+            }
+        }
+    }
+
+    #[test]
+    fn decode_fixed() {
+        assert_eq!(
+            super::decode_fixed("00bgd045z0d4icpbc2yyz4gx48ak44la").unwrap(),
+            hex!("8a12321522fd91efbd60ebb2481af88580f61600")
+        );
+        assert_eq!(
+            super::decode_fixed::<32>("00").unwrap_err(),
+            super::DecodeError {
+                position: 2,
+                kind: super::DecodeKind::Length
+            }
+        );
+    }
+
+    #[test]
+    fn encode_len() {
+        assert_eq!(super::encode_len(0), 0);
+        assert_eq!(super::encode_len(20), 32);
+    }
+
+    #[test]
+    fn decode_len() {
+        assert_eq!(super::decode_len(0), 0);
+        assert_eq!(super::decode_len(32), 20);
+    }
+}
diff --git a/tvix/nix-compat/src/nixhash/algos.rs b/tvix/nix-compat/src/nixhash/algos.rs
new file mode 100644
index 000000000000..f189eba09db6
--- /dev/null
+++ b/tvix/nix-compat/src/nixhash/algos.rs
@@ -0,0 +1,56 @@
+use std::fmt::Display;
+
+use serde::{Deserialize, Serialize};
+
+use crate::nixhash::Error;
+
+/// This are the hash algorithms supported by cppnix.
+#[derive(Clone, Copy, Debug, Eq, PartialEq, Serialize, Deserialize)]
+pub enum HashAlgo {
+    Md5,
+    Sha1,
+    Sha256,
+    Sha512,
+}
+
+impl HashAlgo {
+    // return the number of bytes in the digest of the given hash algo.
+    pub fn digest_length(&self) -> usize {
+        match self {
+            HashAlgo::Sha1 => 20,
+            HashAlgo::Sha256 => 32,
+            HashAlgo::Sha512 => 64,
+            HashAlgo::Md5 => 16,
+        }
+    }
+}
+
+impl Display for HashAlgo {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match &self {
+            HashAlgo::Md5 => write!(f, "md5"),
+            HashAlgo::Sha1 => write!(f, "sha1"),
+            HashAlgo::Sha256 => write!(f, "sha256"),
+            HashAlgo::Sha512 => write!(f, "sha512"),
+        }
+    }
+}
+
+/// TODO(Raito): this could be automated via macros, I suppose.
+/// But this may be more expensive than just doing it by hand
+/// and ensuring that is kept in sync.
+pub const SUPPORTED_ALGOS: [&str; 4] = ["md5", "sha1", "sha256", "sha512"];
+
+impl TryFrom<&str> for HashAlgo {
+    type Error = Error;
+
+    fn try_from(algo_str: &str) -> Result<Self, Self::Error> {
+        match algo_str {
+            "md5" => Ok(Self::Md5),
+            "sha1" => Ok(Self::Sha1),
+            "sha256" => Ok(Self::Sha256),
+            "sha512" => Ok(Self::Sha512),
+            _ => Err(Error::InvalidAlgo(algo_str.to_string())),
+        }
+    }
+}
diff --git a/tvix/nix-compat/src/nixhash/ca_hash.rs b/tvix/nix-compat/src/nixhash/ca_hash.rs
new file mode 100644
index 000000000000..0d6c68d7fa7f
--- /dev/null
+++ b/tvix/nix-compat/src/nixhash/ca_hash.rs
@@ -0,0 +1,213 @@
+use crate::nixbase32;
+use crate::nixhash::{self, HashAlgo, NixHash};
+use serde::de::Unexpected;
+use serde::ser::SerializeMap;
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
+use serde_json::{Map, Value};
+use std::borrow::Cow;
+
+use super::algos::SUPPORTED_ALGOS;
+use super::from_algo_and_digest;
+
+/// A Nix CAHash describes a content-addressed hash of a path.
+///
+/// The way Nix prints it as a string is a bit confusing, but there's essentially
+/// three modes, `Flat`, `Nar` and `Text`.
+/// `Flat` and `Nar` support all 4 algos that [NixHash] supports
+/// (sha1, md5, sha256, sha512), `Text` only supports sha256.
+#[derive(Clone, Debug, Eq, PartialEq)]
+pub enum CAHash {
+    Flat(NixHash),  // "fixed flat"
+    Nar(NixHash),   // "fixed recursive"
+    Text([u8; 32]), // "text", only supports sha256
+}
+
+impl CAHash {
+    pub fn digest(&self) -> Cow<NixHash> {
+        match *self {
+            CAHash::Flat(ref digest) => Cow::Borrowed(digest),
+            CAHash::Nar(ref digest) => Cow::Borrowed(digest),
+            CAHash::Text(digest) => Cow::Owned(NixHash::Sha256(digest)),
+        }
+    }
+
+    /// Constructs a [CAHash] from the textual representation,
+    /// which is one of the three:
+    /// - `text:sha256:$nixbase32sha256digest`
+    /// - `fixed:r:$algo:$nixbase32digest`
+    /// - `fixed:$algo:$nixbase32digest`
+    /// which is the format that's used in the NARInfo for example.
+    pub fn from_nix_hex_str(s: &str) -> Option<Self> {
+        let (tag, s) = s.split_once(':')?;
+
+        match tag {
+            "text" => {
+                let digest = s.strip_prefix("sha256:")?;
+                let digest = nixbase32::decode_fixed(digest).ok()?;
+                Some(CAHash::Text(digest))
+            }
+            "fixed" => {
+                if let Some(s) = s.strip_prefix("r:") {
+                    NixHash::from_nix_hex_str(s).map(CAHash::Nar)
+                } else {
+                    NixHash::from_nix_hex_str(s).map(CAHash::Flat)
+                }
+            }
+            _ => None,
+        }
+    }
+
+    /// Formats a [CAHash] in the Nix default hash format, which is the format
+    /// that's used in NARInfos for example.
+    pub fn to_nix_nixbase32_string(&self) -> String {
+        match self {
+            CAHash::Flat(nh) => format!("fixed:{}", nh.to_nix_nixbase32_string()),
+            CAHash::Nar(nh) => format!("fixed:r:{}", nh.to_nix_nixbase32_string()),
+            CAHash::Text(digest) => {
+                format!("text:sha256:{}", nixbase32::encode(digest))
+            }
+        }
+    }
+
+    /// This takes a serde_json::Map and turns it into this structure. This is necessary to do such
+    /// shenigans because we have external consumers, like the Derivation parser, who would like to
+    /// know whether we have a invalid or a missing NixHashWithMode structure in another structure,
+    /// e.g. Output.
+    /// This means we have this combinatorial situation:
+    /// - no hash, no hashAlgo: no [CAHash] so we return Ok(None).
+    /// - present hash, missing hashAlgo: invalid, we will return missing_field
+    /// - missing hash, present hashAlgo: same
+    /// - present hash, present hashAlgo: either we return ourselves or a type/value validation
+    /// error.
+    /// This function is for internal consumption regarding those needs until we have a better
+    /// solution. Now this is said, let's explain how this works.
+    ///
+    /// We want to map the serde data model into a [CAHash].
+    ///
+    /// The serde data model has a `hash` field (containing a digest in nixbase32),
+    /// and a `hashAlgo` field, containing the stringified hash algo.
+    /// In case the hash is recursive, hashAlgo also has a `r:` prefix.
+    ///
+    /// This is to match how `nix show-derivation` command shows them in JSON
+    /// representation.
+    pub(crate) fn from_map<'de, D>(map: &Map<String, Value>) -> Result<Option<Self>, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        // If we don't have hash neither hashAlgo, let's just return None.
+        if !map.contains_key("hash") && !map.contains_key("hashAlgo") {
+            return Ok(None);
+        }
+
+        let digest: Vec<u8> = {
+            if let Some(v) = map.get("hash") {
+                if let Some(s) = v.as_str() {
+                    data_encoding::HEXLOWER
+                        .decode(s.as_bytes())
+                        .map_err(|e| serde::de::Error::custom(e.to_string()))?
+                } else {
+                    return Err(serde::de::Error::invalid_type(
+                        Unexpected::Other(&v.to_string()),
+                        &"a string",
+                    ));
+                }
+            } else {
+                return Err(serde::de::Error::missing_field(
+                    "couldn't extract `hash` key but `hashAlgo` key present",
+                ));
+            }
+        };
+
+        if let Some(v) = map.get("hashAlgo") {
+            if let Some(s) = v.as_str() {
+                match s.strip_prefix("r:") {
+                    Some(rest) => Ok(Some(Self::Nar(
+                        from_algo_and_digest(
+                            HashAlgo::try_from(rest).map_err(|e| {
+                                serde::de::Error::invalid_value(
+                                    Unexpected::Other(&e.to_string()),
+                                    &format!("one of {}", SUPPORTED_ALGOS.join(",")).as_str(),
+                                )
+                            })?,
+                            &digest,
+                        )
+                        .map_err(|e: nixhash::Error| {
+                            serde::de::Error::invalid_value(
+                                Unexpected::Other(&e.to_string()),
+                                &"a digest with right length",
+                            )
+                        })?,
+                    ))),
+                    None => Ok(Some(Self::Flat(
+                        from_algo_and_digest(
+                            HashAlgo::try_from(s).map_err(|e| {
+                                serde::de::Error::invalid_value(
+                                    Unexpected::Other(&e.to_string()),
+                                    &format!("one of {}", SUPPORTED_ALGOS.join(",")).as_str(),
+                                )
+                            })?,
+                            &digest,
+                        )
+                        .map_err(|e: nixhash::Error| {
+                            serde::de::Error::invalid_value(
+                                Unexpected::Other(&e.to_string()),
+                                &"a digest with right length",
+                            )
+                        })?,
+                    ))),
+                }
+            } else {
+                Err(serde::de::Error::invalid_type(
+                    Unexpected::Other(&v.to_string()),
+                    &"a string",
+                ))
+            }
+        } else {
+            Err(serde::de::Error::missing_field(
+                "couldn't extract `hashAlgo` key, but `hash` key present",
+            ))
+        }
+    }
+}
+
+impl Serialize for CAHash {
+    /// map a CAHash into the serde data model.
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let mut map = serializer.serialize_map(Some(2))?;
+        match self {
+            CAHash::Flat(h) => {
+                map.serialize_entry("hash", &nixbase32::encode(h.digest_as_bytes()))?;
+                map.serialize_entry("hashAlgo", &h.algo())?;
+            }
+            CAHash::Nar(h) => {
+                map.serialize_entry("hash", &nixbase32::encode(h.digest_as_bytes()))?;
+                map.serialize_entry("hashAlgo", &format!("r:{}", &h.algo()))?;
+            }
+            // It is not legal for derivations to use this (which is where
+            // we're currently exercising [Serialize] mostly,
+            // but it's still good to be able to serialize other CA hashes too.
+            CAHash::Text(h) => {
+                map.serialize_entry("hash", &nixbase32::encode(h.as_ref()))?;
+                map.serialize_entry("hashAlgo", "text")?;
+            }
+        };
+        map.end()
+    }
+}
+
+impl<'de> Deserialize<'de> for CAHash {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let value = Self::from_map::<D>(&Map::deserialize(deserializer)?)?;
+
+        match value {
+            None => Err(serde::de::Error::custom("couldn't parse as map")),
+            Some(v) => Ok(v),
+        }
+    }
+}
diff --git a/tvix/nix-compat/src/nixhash/mod.rs b/tvix/nix-compat/src/nixhash/mod.rs
new file mode 100644
index 000000000000..e148eb72ee9c
--- /dev/null
+++ b/tvix/nix-compat/src/nixhash/mod.rs
@@ -0,0 +1,530 @@
+use crate::nixbase32;
+use data_encoding::{BASE64, BASE64_NOPAD, HEXLOWER};
+use thiserror;
+
+mod algos;
+mod ca_hash;
+
+pub use algos::HashAlgo;
+pub use ca_hash::CAHash;
+
+/// NixHash represents hashes known by Nix.
+#[derive(Clone, Debug, Eq, PartialEq)]
+pub enum NixHash {
+    Md5([u8; 16]),
+    Sha1([u8; 20]),
+    Sha256([u8; 32]),
+    Sha512(Box<[u8; 64]>),
+}
+
+/// convenience Result type for all nixhash parsing Results.
+pub type Result<V> = std::result::Result<V, Error>;
+
+impl NixHash {
+    /// returns the algo as [HashAlgo].
+    pub fn algo(&self) -> HashAlgo {
+        match self {
+            NixHash::Md5(_) => HashAlgo::Md5,
+            NixHash::Sha1(_) => HashAlgo::Sha1,
+            NixHash::Sha256(_) => HashAlgo::Sha256,
+            NixHash::Sha512(_) => HashAlgo::Sha512,
+        }
+    }
+
+    /// returns the digest as variable-length byte slice.
+    pub fn digest_as_bytes(&self) -> &[u8] {
+        match self {
+            NixHash::Md5(digest) => digest,
+            NixHash::Sha1(digest) => digest,
+            NixHash::Sha256(digest) => digest,
+            NixHash::Sha512(digest) => digest.as_ref(),
+        }
+    }
+
+    /// Constructs a [NixHash] from the Nix default hash format,
+    /// the inverse of [Self::to_nix_hex_string].
+    pub fn from_nix_hex_str(s: &str) -> Option<Self> {
+        let (tag, digest) = s.split_once(':')?;
+
+        (match tag {
+            "md5" => nixbase32::decode_fixed(digest).map(NixHash::Md5),
+            "sha1" => nixbase32::decode_fixed(digest).map(NixHash::Sha1),
+            "sha256" => nixbase32::decode_fixed(digest).map(NixHash::Sha256),
+            "sha512" => nixbase32::decode_fixed(digest)
+                .map(Box::new)
+                .map(NixHash::Sha512),
+            _ => return None,
+        })
+        .ok()
+    }
+
+    /// Formats a [NixHash] in the Nix default hash format,
+    /// which is the algo, followed by a colon, then the lower hex encoded digest.
+    pub fn to_nix_hex_string(&self) -> String {
+        format!(
+            "{}:{}",
+            self.algo(),
+            HEXLOWER.encode(self.digest_as_bytes())
+        )
+    }
+
+    /// Formats a [NixHash] in the format that's used inside CAHash,
+    /// which is the algo, followed by a colon, then the nixbase32-encoded digest.
+    pub(crate) fn to_nix_nixbase32_string(&self) -> String {
+        format!(
+            "{}:{}",
+            self.algo(),
+            nixbase32::encode(self.digest_as_bytes())
+        )
+    }
+}
+
+impl TryFrom<(HashAlgo, &[u8])> for NixHash {
+    type Error = Error;
+
+    /// Constructs a new [NixHash] by specifying [HashAlgo] and digest.
+    /// It can fail if the passed digest length doesn't match what's expected for
+    /// the passed algo.
+    fn try_from(value: (HashAlgo, &[u8])) -> Result<Self> {
+        let (algo, digest) = value;
+        from_algo_and_digest(algo, digest)
+    }
+}
+
+/// Constructs a new [NixHash] by specifying [HashAlgo] and digest.
+/// It can fail if the passed digest length doesn't match what's expected for
+/// the passed algo.
+pub fn from_algo_and_digest(algo: HashAlgo, digest: &[u8]) -> Result<NixHash> {
+    if digest.len() != algo.digest_length() {
+        return Err(Error::InvalidEncodedDigestLength(digest.len(), algo));
+    }
+
+    Ok(match algo {
+        HashAlgo::Md5 => NixHash::Md5(digest.try_into().unwrap()),
+        HashAlgo::Sha1 => NixHash::Sha1(digest.try_into().unwrap()),
+        HashAlgo::Sha256 => NixHash::Sha256(digest.try_into().unwrap()),
+        HashAlgo::Sha512 => NixHash::Sha512(Box::new(digest.try_into().unwrap())),
+    })
+}
+
+/// Errors related to NixHash construction.
+#[derive(Debug, Eq, PartialEq, thiserror::Error)]
+pub enum Error {
+    #[error("invalid hash algo: {0}")]
+    InvalidAlgo(String),
+    #[error("invalid SRI string: {0}")]
+    InvalidSRI(String),
+    #[error("invalid encoded digest length '{0}' for algo {1}")]
+    InvalidEncodedDigestLength(usize, HashAlgo),
+    #[error("invalid base16 encoding: {0}")]
+    InvalidBase16Encoding(data_encoding::DecodeError),
+    #[error("invalid base32 encoding: {0}")]
+    InvalidBase32Encoding(data_encoding::DecodeError),
+    #[error("invalid base64 encoding: {0}")]
+    InvalidBase64Encoding(data_encoding::DecodeError),
+    #[error("conflicting hash algo: {0} (hash_algo) vs {1} (inline)")]
+    ConflictingHashAlgos(HashAlgo, HashAlgo),
+    #[error("missing inline hash algo, but no externally-specified algo: {0}")]
+    MissingInlineHashAlgo(String),
+}
+
+/// Nix allows specifying hashes in various encodings, and magically just
+/// derives the encoding.
+/// This function parses strings to a NixHash.
+///
+/// Hashes can be:
+/// - Nix hash strings
+/// - SRI hashes
+/// - bare digests
+///
+/// Encoding for Nix hash strings or bare digests can be:
+/// - base16 (lowerhex),
+/// - nixbase32,
+/// - base64 (StdEncoding)
+/// - sri string
+///
+/// The encoding is derived from the length of the string and the hash type.
+/// The hash is communicated out-of-band, but might also be in-band (in the
+/// case of a nix hash string or SRI), in which it needs to be consistent with the
+/// one communicated out-of-band.
+pub fn from_str(s: &str, algo_str: Option<&str>) -> Result<NixHash> {
+    // if algo_str is some, parse or bail out
+    let algo: Option<HashAlgo> = if let Some(algo_str) = algo_str {
+        Some(algo_str.try_into()?)
+    } else {
+        None
+    };
+
+    // Peek at the beginning of the string to detect SRI hashes.
+    if s.starts_with("sha1-")
+        || s.starts_with("sha256-")
+        || s.starts_with("sha512-")
+        || s.starts_with("md5-")
+    {
+        let parsed_nixhash = from_sri_str(s)?;
+
+        // ensure the algo matches with what has been passed externally, if so.
+        if let Some(algo) = algo {
+            if algo != parsed_nixhash.algo() {
+                return Err(Error::ConflictingHashAlgos(algo, parsed_nixhash.algo()));
+            }
+        }
+        return Ok(parsed_nixhash);
+    }
+
+    // Peek at the beginning again to see if it's a Nix Hash
+    if s.starts_with("sha1:")
+        || s.starts_with("sha256:")
+        || s.starts_with("sha512:")
+        || s.starts_with("md5:")
+    {
+        let parsed_nixhash = from_nix_str(s)?;
+        // ensure the algo matches with what has been passed externally, if so.
+        if let Some(algo) = algo {
+            if algo != parsed_nixhash.algo() {
+                return Err(Error::ConflictingHashAlgos(algo, parsed_nixhash.algo()));
+            }
+        }
+        return Ok(parsed_nixhash);
+    }
+
+    // Neither of these, assume a bare digest, so there MUST be an externally-passed algo.
+    match algo {
+        // Fail if there isn't.
+        None => Err(Error::MissingInlineHashAlgo(s.to_string())),
+        Some(algo) => decode_digest(s.as_bytes(), algo),
+    }
+}
+
+/// Parses a Nix hash string ($algo:$digest) to a NixHash.
+pub fn from_nix_str(s: &str) -> Result<NixHash> {
+    if let Some(rest) = s.strip_prefix("sha1:") {
+        decode_digest(rest.as_bytes(), HashAlgo::Sha1)
+    } else if let Some(rest) = s.strip_prefix("sha256:") {
+        decode_digest(rest.as_bytes(), HashAlgo::Sha256)
+    } else if let Some(rest) = s.strip_prefix("sha512:") {
+        decode_digest(rest.as_bytes(), HashAlgo::Sha512)
+    } else if let Some(rest) = s.strip_prefix("md5:") {
+        decode_digest(rest.as_bytes(), HashAlgo::Md5)
+    } else {
+        Err(Error::InvalidAlgo(s.to_string()))
+    }
+}
+
+/// Parses a Nix SRI string to a NixHash.
+/// Contrary to the SRI spec, Nix doesn't support SRI strings with multiple hashes,
+/// only supports sha256 and sha512 from the spec, and supports sha1 and md5
+/// additionally.
+/// It's also dealing with padding in a very funny way - accepting SRI strings
+/// with an arbitrary amount of padding at the end - including *more* padding
+/// characters.
+pub fn from_sri_str(s: &str) -> Result<NixHash> {
+    // try to find the first occurence of "-"
+    let idx = s.as_bytes().iter().position(|&e| e == b'-');
+
+    if idx.is_none() {
+        return Err(Error::InvalidSRI(s.to_string()));
+    }
+
+    let idx = idx.unwrap();
+
+    // try to map the part before that `-` to a supported hash algo:
+    let algo: HashAlgo = s[..idx].try_into()?;
+
+    // the rest should be the digest (as Nix doesn't support more than one hash in an SRI string).
+    let encoded_digest = &s[idx + 1..];
+
+    // strip all padding characters.
+    let encoded_digest = encoded_digest.trim_end_matches('=');
+
+    // If we are using BASE64_NOPAD, we must also disable the trailing bit checking otherwise we
+    // are bound to get invalid length for our inputs.
+    // See the `weird_sha256` example below.
+    let mut spec = BASE64_NOPAD.specification();
+    spec.check_trailing_bits = false;
+    let encoder = spec
+        .encoding()
+        .expect("Tvix bug: failed to get the special base64 encoder for Nix SRI hashes");
+
+    if encoded_digest.len() == encoder.encode_len(algo.digest_length()) {
+        let digest = encoder
+            .decode(encoded_digest.as_bytes())
+            .map_err(Error::InvalidBase64Encoding)?;
+
+        Ok(from_algo_and_digest(algo, &digest).unwrap())
+    } else {
+        Err(Error::InvalidEncodedDigestLength(idx, algo))?
+    }
+}
+
+/// Decode a plain digest depending on the hash algo specified externally.
+/// hexlower, nixbase32 and base64 encodings are supported - the encoding is
+/// inferred from the input length.
+fn decode_digest(s: &[u8], algo: HashAlgo) -> Result<NixHash> {
+    // for the chosen hash algo, calculate the expected (decoded) digest length
+    // (as bytes)
+    let digest = if s.len() == HEXLOWER.encode_len(algo.digest_length()) {
+        HEXLOWER
+            .decode(s.as_ref())
+            .map_err(Error::InvalidBase16Encoding)?
+    } else if s.len() == nixbase32::encode_len(algo.digest_length()) {
+        nixbase32::decode(s).map_err(Error::InvalidBase32Encoding)?
+    } else if s.len() == BASE64.encode_len(algo.digest_length()) {
+        BASE64
+            .decode(s.as_ref())
+            .map_err(Error::InvalidBase64Encoding)?
+    } else {
+        Err(Error::InvalidEncodedDigestLength(s.len(), algo))?
+    };
+
+    Ok(from_algo_and_digest(algo, &digest).unwrap())
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::{
+        nixbase32,
+        nixhash::{self, HashAlgo, NixHash},
+    };
+    use data_encoding::{BASE64, BASE64_NOPAD, HEXLOWER};
+    use hex_literal::hex;
+    use test_case::test_case;
+
+    const DIGEST_SHA1: [u8; 20] = hex!("6016777997c30ab02413cf5095622cd7924283ac");
+    const DIGEST_SHA256: [u8; 32] =
+        hex!("a5ce9c155ed09397614646c9717fc7cd94b1023d7b76b618d409e4fefd6e9d39");
+    const DIGEST_SHA512: [u8; 64] = hex!("ab40d0be3541f0774bba7815d13d10b03252e96e95f7dbb4ee99a3b431c21662fd6971a020160e39848aa5f305b9be0f78727b2b0789e39f124d21e92b8f39ef");
+    const DIGEST_MD5: [u8; 16] = hex!("c4874a8897440b393d862d8fd459073f");
+
+    fn to_base16(digest: &[u8]) -> String {
+        HEXLOWER.encode(digest)
+    }
+
+    fn to_nixbase32(digest: &[u8]) -> String {
+        nixbase32::encode(digest)
+    }
+
+    fn to_base64(digest: &[u8]) -> String {
+        BASE64.encode(digest)
+    }
+
+    fn to_base64_nopad(digest: &[u8]) -> String {
+        BASE64_NOPAD.encode(digest)
+    }
+
+    // TODO
+    fn make_nixhash(algo: &HashAlgo, digest_encoded: String) -> String {
+        format!("{}:{}", algo, digest_encoded)
+    }
+    fn make_sri_string(algo: &HashAlgo, digest_encoded: String) -> String {
+        format!("{}-{}", algo, digest_encoded)
+    }
+
+    /// Test parsing a hash string in various formats, and also when/how the out-of-band algo is needed.
+    #[test_case(&NixHash::Sha1(DIGEST_SHA1); "sha1")]
+    #[test_case(&NixHash::Sha256(DIGEST_SHA256); "sha256")]
+    #[test_case(&NixHash::Sha512(Box::new(DIGEST_SHA512)); "sha512")]
+    #[test_case(&NixHash::Md5(DIGEST_MD5); "md5")]
+    fn from_str(expected_hash: &NixHash) {
+        let algo = &expected_hash.algo();
+        let digest = expected_hash.digest_as_bytes();
+        // parse SRI
+        {
+            // base64 without out-of-band algo
+            let s = make_sri_string(algo, to_base64(digest));
+            let h = nixhash::from_str(&s, None).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+
+            // base64 with out-of-band-algo
+            let s = make_sri_string(algo, to_base64(digest));
+            let h = nixhash::from_str(&s, Some(&expected_hash.algo().to_string()))
+                .expect("must succeed");
+            assert_eq!(expected_hash, &h);
+
+            // base64_nopad without out-of-band algo
+            let s = make_sri_string(algo, to_base64_nopad(digest));
+            let h = nixhash::from_str(&s, None).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+
+            // base64_nopad with out-of-band-algo
+            let s = make_sri_string(algo, to_base64_nopad(digest));
+            let h = nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+        }
+
+        // parse plain base16. should succeed with algo out-of-band, but fail without.
+        {
+            let s = to_base16(digest);
+            nixhash::from_str(&s, None).expect_err("must fail");
+            let h = nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+        }
+
+        // parse plain nixbase32. should succeed with algo out-of-band, but fail without.
+        {
+            let s = to_nixbase32(digest);
+            nixhash::from_str(&s, None).expect_err("must fail");
+            let h = nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+        }
+
+        // parse plain base64. should succeed with algo out-of-band, but fail without.
+        {
+            let s = to_base64(digest);
+            nixhash::from_str(&s, None).expect_err("must fail");
+            let h = nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed");
+            assert_eq!(expected_hash, &h);
+        }
+
+        // parse Nix hash strings
+        {
+            // base16. should succeed with both algo out-of-band and in-band.
+            {
+                let s = make_nixhash(algo, to_base16(digest));
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, None).expect("must succeed")
+                );
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed")
+                );
+            }
+            // nixbase32. should succeed with both algo out-of-band and in-band.
+            {
+                let s = make_nixhash(algo, to_nixbase32(digest));
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, None).expect("must succeed")
+                );
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed")
+                );
+            }
+            // base64. should succeed with both algo out-of-band and in-band.
+            {
+                let s = make_nixhash(algo, to_base64(digest));
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, None).expect("must succeed")
+                );
+                assert_eq!(
+                    expected_hash,
+                    &nixhash::from_str(&s, Some(&algo.to_string())).expect("must succeed")
+                );
+            }
+        }
+    }
+
+    /// Test parsing an SRI hash via the [nixhash::from_sri_str] method.
+    #[test]
+    fn from_sri_str() {
+        let nix_hash = nixhash::from_sri_str("sha256-pc6cFV7Qk5dhRkbJcX/HzZSxAj17drYY1Ank/v1unTk=")
+            .expect("must succeed");
+
+        assert_eq!(HashAlgo::Sha256, nix_hash.algo());
+        assert_eq!(
+            &hex!("a5ce9c155ed09397614646c9717fc7cd94b1023d7b76b618d409e4fefd6e9d39"),
+            nix_hash.digest_as_bytes()
+        )
+    }
+
+    /// Test parsing sha512 SRI hash with various paddings, Nix accepts all of them.
+    #[test_case("sha512-7g91TBvYoYQorRTqo+rYD/i5YnWvUBLnqDhPHxBJDaBW7smuPMeRp6E6JOFuVN9bzN0QnH1ToUU0u9c2CjALEQ"; "no padding")]
+    #[test_case("sha512-7g91TBvYoYQorRTqo+rYD/i5YnWvUBLnqDhPHxBJDaBW7smuPMeRp6E6JOFuVN9bzN0QnH1ToUU0u9c2CjALEQ="; "wrong padding")]
+    #[test_case("sha512-7g91TBvYoYQorRTqo+rYD/i5YnWvUBLnqDhPHxBJDaBW7smuPMeRp6E6JOFuVN9bzN0QnH1ToUU0u9c2CjALEQ="; "correct padding")]
+    #[test_case("sha512-7g91TBvYoYQorRTqo+rYD/i5YnWvUBLnqDhPHxBJDaBW7smuPMeRp6E6JOFuVN9bzN0QnH1ToUU0u9c2CjALEQ="; "too much padding")]
+    fn from_sri_str_sha512_paddings(sri_str: &str) {
+        let nix_hash = nixhash::from_sri_str(sri_str).expect("must succeed");
+
+        assert_eq!(HashAlgo::Sha512, nix_hash.algo());
+        assert_eq!(
+            &hex!("ee0f754c1bd8a18428ad14eaa3ead80ff8b96275af5012e7a8384f1f10490da056eec9ae3cc791a7a13a24e16e54df5bccdd109c7d53a14534bbd7360a300b11"),
+            nix_hash.digest_as_bytes()
+        )
+    }
+
+    /// Ensure we detect truncated base64 digests, where the digest size
+    /// doesn't match what's expected from that hash function.
+    #[test]
+    fn from_sri_str_truncated() {
+        nixhash::from_sri_str("sha256-pc6cFV7Qk5dhRkbJcX/HzZSxAj17drYY1Ank")
+            .expect_err("must fail");
+    }
+
+    /// Ensure we fail on SRI hashes that Nix doesn't support.
+    #[test]
+    fn from_sri_str_unsupported() {
+        nixhash::from_sri_str(
+            "sha384-o4UVSl89mIB0sFUK+3jQbG+C9Zc9dRlV/Xd3KAvXEbhqxu0J5OAdg6b6VHKHwQ7U",
+        )
+        .expect_err("must fail");
+    }
+
+    /// Ensure we reject invalid base64 encoding
+    #[test]
+    fn from_sri_str_invalid_base64() {
+        nixhash::from_sri_str("sha256-invalid=base64").expect_err("must fail");
+    }
+
+    /// Ensure we reject SRI strings with multiple hashes, as Nix doesn't support that.
+    #[test]
+    fn from_sri_str_unsupported_multiple() {
+        nixhash::from_sri_str("sha256-ngth6szLtC1IJIYyz3lhftzL8SkrJkqPyPve+dGqa1Y= sha512-q0DQvjVB8HdLungV0T0QsDJS6W6V99u07pmjtDHCFmL9aXGgIBYOOYSKpfMFub4PeHJ7KweJ458STSHpK4857w==").expect_err("must fail");
+    }
+
+    /// Nix also accepts SRI strings with missing padding, but only in case the
+    /// string is expressed as SRI, so it still needs to have a `sha256-` prefix.
+    ///
+    /// This both seems to work if it is passed with and without specifying the
+    /// hash algo out-of-band (hash = "sha256-…" or sha256 = "sha256-…")
+    ///
+    /// Passing the same broken base64 string, but not as SRI, while passing
+    /// the hash algo out-of-band does not work.
+    #[test]
+    fn sha256_broken_padding() {
+        let broken_base64 = "fgIr3TyFGDAXP5+qoAaiMKDg/a1MlT6Fv/S/DaA24S8";
+        // if padded with a trailing '='
+        let expected_digest =
+            hex!("7e022bdd3c851830173f9faaa006a230a0e0fdad4c953e85bff4bf0da036e12f");
+
+        // passing hash algo out of band should succeed
+        let nix_hash = nixhash::from_str(&format!("sha256-{}", &broken_base64), Some("sha256"))
+            .expect("must succeed");
+        assert_eq!(&expected_digest, &nix_hash.digest_as_bytes());
+
+        // not passing hash algo out of band should succeed
+        let nix_hash =
+            nixhash::from_str(&format!("sha256-{}", &broken_base64), None).expect("must succeed");
+        assert_eq!(&expected_digest, &nix_hash.digest_as_bytes());
+
+        // not passing SRI, but hash algo out of band should fail
+        nixhash::from_str(broken_base64, Some("sha256")).expect_err("must fail");
+    }
+
+    /// As we decided to pass our hashes by trimming `=` completely,
+    /// we need to take into account hashes with padding requirements which
+    /// contains trailing bits which would be checked by `BASE64_NOPAD` and would
+    /// make the verification crash.
+    ///
+    /// This base64 has a trailing non-zero bit at bit 42.
+    #[test]
+    fn sha256_weird_base64() {
+        let weird_base64 = "syceJMUEknBDCHK8eGs6rUU3IQn+HnQfURfCrDxYPa9=";
+        let expected_digest =
+            hex!("b3271e24c5049270430872bc786b3aad45372109fe1e741f5117c2ac3c583daf");
+
+        let nix_hash = nixhash::from_str(&format!("sha256-{}", &weird_base64), Some("sha256"))
+            .expect("must succeed");
+        assert_eq!(&expected_digest, &nix_hash.digest_as_bytes());
+
+        // not passing hash algo out of band should succeed
+        let nix_hash =
+            nixhash::from_str(&format!("sha256-{}", &weird_base64), None).expect("must succeed");
+        assert_eq!(&expected_digest, &nix_hash.digest_as_bytes());
+
+        // not passing SRI, but hash algo out of band should fail
+        nixhash::from_str(weird_base64, Some("sha256")).expect_err("must fail");
+    }
+}
diff --git a/tvix/nix-compat/src/store_path/mod.rs b/tvix/nix-compat/src/store_path/mod.rs
new file mode 100644
index 000000000000..e23002bfb97e
--- /dev/null
+++ b/tvix/nix-compat/src/store_path/mod.rs
@@ -0,0 +1,430 @@
+use crate::nixbase32;
+use data_encoding::{DecodeError, BASE64};
+use std::{
+    fmt,
+    path::PathBuf,
+    str::{self, FromStr},
+};
+use thiserror;
+
+#[cfg(target_family = "unix")]
+use std::os::unix::ffi::OsStringExt;
+
+mod utils;
+
+pub use utils::*;
+
+pub const DIGEST_SIZE: usize = 20;
+pub const ENCODED_DIGEST_SIZE: usize = nixbase32::encode_len(DIGEST_SIZE);
+
+// The store dir prefix, without trailing slash.
+// That's usually where the Nix store is mounted at.
+pub const STORE_DIR: &str = "/nix/store";
+pub const STORE_DIR_WITH_SLASH: &str = "/nix/store/";
+
+/// Errors that can occur when parsing a literal store path
+#[derive(Debug, PartialEq, Eq, thiserror::Error)]
+pub enum Error {
+    #[error("Dash is missing between hash and name")]
+    MissingDash,
+    #[error("Hash encoding is invalid: {0}")]
+    InvalidHashEncoding(DecodeError),
+    #[error("Invalid length")]
+    InvalidLength,
+    #[error(
+        "Invalid name: \"{}\", character at position {} is invalid",
+        std::str::from_utf8(.0).unwrap_or(&BASE64.encode(.0)),
+        .1,
+    )]
+    InvalidName(Vec<u8>, u8),
+    #[error("Tried to parse an absolute path which was missing the store dir prefix.")]
+    MissingStoreDir,
+}
+
+/// Represents a path in the Nix store (a direct child of [STORE_DIR]).
+///
+/// It consists of a digest (20 bytes), and a name, which is a string.
+/// The name may only contain ASCII alphanumeric, or one of the following
+/// characters: `-`, `_`, `.`, `+`, `?`, `=`.
+/// The name is usually used to describe the pname and version of a package.
+/// Derivation paths can also be represented as store paths, their names just
+/// end with the `.drv` prefix.
+///
+/// A [StorePath] does not encode any additional subpath "inside" the store
+/// path.
+#[derive(Clone, Debug, PartialEq, Eq, Hash)]
+pub struct StorePath {
+    digest: [u8; DIGEST_SIZE],
+    name: String,
+}
+
+impl StorePath {
+    pub fn digest(&self) -> &[u8; DIGEST_SIZE] {
+        &self.digest
+    }
+
+    pub fn name(&self) -> &str {
+        self.name.as_ref()
+    }
+}
+
+impl PartialOrd for StorePath {
+    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl Ord for StorePath {
+    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+        self.digest.cmp(&other.digest)
+    }
+}
+
+impl FromStr for StorePath {
+    type Err = Error;
+
+    /// Construct a [StorePath] by passing the `$digest-$name` string
+    /// that comes after [STORE_DIR_WITH_SLASH].
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        Self::from_bytes(s.as_bytes())
+    }
+}
+
+impl StorePath {
+    /// Construct a [StorePath] by passing the `$digest-$name` string
+    /// that comes after [STORE_DIR_WITH_SLASH].
+    pub fn from_bytes(s: &[u8]) -> Result<StorePath, Error> {
+        Ok(StorePathRef::from_bytes(s)?.to_owned())
+    }
+
+    /// Decompose a string into a [StorePath] and a [PathBuf] containing the
+    /// rest of the path, or an error.
+    #[cfg(target_family = "unix")]
+    pub fn from_absolute_path_full(s: &str) -> Result<(StorePath, PathBuf), Error> {
+        // strip [STORE_DIR_WITH_SLASH] from s
+        match s.strip_prefix(STORE_DIR_WITH_SLASH) {
+            None => Err(Error::MissingStoreDir),
+            Some(rest) => {
+                // put rest in a PathBuf
+                let mut p = PathBuf::new();
+                p.push(rest);
+
+                let mut it = p.components();
+
+                // The first component of the rest must be parse-able as a [StorePath]
+                if let Some(first_component) = it.next() {
+                    // convert first component to StorePath
+                    let first_component_bytes = first_component.as_os_str().to_owned().into_vec();
+                    let store_path = StorePath::from_bytes(&first_component_bytes)?;
+                    // collect rest
+                    let rest_buf: PathBuf = it.collect();
+                    Ok((store_path, rest_buf))
+                } else {
+                    Err(Error::InvalidLength) // Well, or missing "/"?
+                }
+            }
+        }
+    }
+
+    /// Returns an absolute store path string.
+    /// That is just the string representation, prefixed with the store prefix
+    /// ([STORE_DIR_WITH_SLASH]),
+    pub fn to_absolute_path(&self) -> String {
+        let sp_ref: StorePathRef = self.into();
+        sp_ref.to_absolute_path()
+    }
+}
+
+/// Like [StorePath], but without a heap allocation for the name.
+/// Used by [StorePath] for parsing.
+///
+#[derive(Debug, Eq, PartialEq)]
+pub struct StorePathRef<'a> {
+    digest: [u8; DIGEST_SIZE],
+    name: &'a str,
+}
+
+impl<'a> From<&'a StorePath> for StorePathRef<'a> {
+    fn from(&StorePath { digest, ref name }: &'a StorePath) -> Self {
+        StorePathRef {
+            digest,
+            name: name.as_ref(),
+        }
+    }
+}
+
+impl<'a> StorePathRef<'a> {
+    pub fn digest(&self) -> &[u8; DIGEST_SIZE] {
+        &self.digest
+    }
+
+    pub fn name(&self) -> &'a str {
+        self.name
+    }
+
+    pub fn to_owned(&self) -> StorePath {
+        StorePath {
+            digest: self.digest,
+            name: self.name.to_owned(),
+        }
+    }
+
+    /// Construct a [StorePathRef] from a name and digest.
+    pub fn from_name_and_digest(name: &'a str, digest: &[u8]) -> Result<Self, Error> {
+        Ok(Self {
+            name: validate_name(name.as_bytes())?,
+            digest: digest.try_into().map_err(|_| Error::InvalidLength)?,
+        })
+    }
+
+    /// Construct a [StorePathRef] from an absolute store path string.
+    /// This is equivalent to calling [StorePathRef::from_bytes], but stripping
+    /// the [STORE_DIR_WITH_SLASH] prefix before.
+    pub fn from_absolute_path(s: &'a [u8]) -> Result<Self, Error> {
+        match s.strip_prefix(STORE_DIR_WITH_SLASH.as_bytes()) {
+            Some(s_stripped) => Self::from_bytes(s_stripped),
+            None => Err(Error::MissingStoreDir),
+        }
+    }
+
+    /// Construct a [StorePathRef] by passing the `$digest-$name` string
+    /// that comes after [STORE_DIR_WITH_SLASH].
+    pub fn from_bytes(s: &'a [u8]) -> Result<Self, Error> {
+        // the whole string needs to be at least:
+        //
+        // - 32 characters (encoded hash)
+        // - 1 dash
+        // - 1 character for the name
+        if s.len() < ENCODED_DIGEST_SIZE + 2 {
+            Err(Error::InvalidLength)?
+        }
+
+        let digest = nixbase32::decode_fixed(&s[..ENCODED_DIGEST_SIZE])
+            .map_err(Error::InvalidHashEncoding)?;
+
+        if s[ENCODED_DIGEST_SIZE] != b'-' {
+            return Err(Error::MissingDash);
+        }
+
+        Ok(StorePathRef {
+            digest,
+            name: validate_name(&s[ENCODED_DIGEST_SIZE + 1..])?,
+        })
+    }
+
+    /// Returns an absolute store path string.
+    /// That is just the string representation, prefixed with the store prefix
+    /// ([STORE_DIR_WITH_SLASH]),
+    pub fn to_absolute_path(&self) -> String {
+        format!("{}{}", STORE_DIR_WITH_SLASH, self)
+    }
+}
+
+/// NAME_CHARS contains `true` for bytes that are valid in store path names,
+/// not accounting for '.' being permitted only past the first character.
+static NAME_CHARS: [bool; 256] = {
+    let mut tbl = [false; 256];
+    let mut c = 0;
+
+    loop {
+        tbl[c as usize] = matches!(c, b'a'..=b'z' | b'A'..=b'Z' | b'0'..=b'9' | b'+' | b'-' | b'_' | b'?' | b'=' | b'.');
+
+        if c == u8::MAX {
+            break;
+        }
+
+        c += 1;
+    }
+
+    tbl
+};
+
+/// Checks a given &[u8] to match the restrictions for [StorePath::name], and
+/// returns the name as string if successful.
+pub(crate) fn validate_name(s: &(impl AsRef<[u8]> + ?Sized)) -> Result<&str, Error> {
+    let s = s.as_ref();
+
+    // Empty or excessively long names are not allowed.
+    if s.is_empty() || s.len() > 211 {
+        return Err(Error::InvalidLength);
+    }
+
+    if s[0] == b'.' {
+        return Err(Error::InvalidName(s.to_vec(), 0));
+    }
+
+    let mut valid = true;
+    for &c in s {
+        valid = valid && NAME_CHARS[c as usize];
+    }
+
+    if !valid {
+        for (i, &c) in s.iter().enumerate() {
+            if !NAME_CHARS[c as usize] {
+                return Err(Error::InvalidName(s.to_vec(), i as u8));
+            }
+        }
+
+        unreachable!();
+    }
+
+    // SAFETY: We permit a subset of ASCII, which guarantees valid UTF-8.
+    Ok(unsafe { str::from_utf8_unchecked(s) })
+}
+
+impl fmt::Display for StorePath {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        StorePathRef::from(self).fmt(f)
+    }
+}
+
+impl fmt::Display for StorePathRef<'_> {
+    /// The string representation of a store path starts with a digest (20
+    /// bytes), [crate::nixbase32]-encoded, followed by a `-`,
+    /// and ends with the name.
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{}-{}", nixbase32::encode(&self.digest), self.name)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::path::PathBuf;
+
+    use crate::store_path::{StorePathRef, DIGEST_SIZE};
+    use hex_literal::hex;
+    use test_case::test_case;
+
+    use super::{Error, StorePath};
+
+    #[test]
+    fn happy_path() {
+        let example_nix_path_str =
+            "00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432";
+        let nixpath = StorePath::from_bytes(example_nix_path_str.as_bytes())
+            .expect("Error parsing example string");
+
+        let expected_digest: [u8; DIGEST_SIZE] = hex!("8a12321522fd91efbd60ebb2481af88580f61600");
+
+        assert_eq!("net-tools-1.60_p20170221182432", nixpath.name);
+        assert_eq!(nixpath.digest, expected_digest);
+
+        assert_eq!(example_nix_path_str, nixpath.to_string())
+    }
+
+    /// This is the store path rejected when `nix-store --add`'ing an
+    /// empty `.gitignore` file.
+    ///
+    /// Nix 2.4 accidentally dropped this behaviour, but this is considered a bug.
+    /// See https://github.com/NixOS/nix/pull/9095.
+    #[test]
+    fn starts_with_dot() {
+        StorePath::from_bytes(b"fli4bwscgna7lpm7v5xgnjxrxh0yc7ra-.gitignore")
+            .expect_err("must fail");
+    }
+
+    #[test]
+    fn empty_name() {
+        StorePath::from_bytes(b"00bgd045z0d4icpbc2yy-").expect_err("must fail");
+    }
+
+    #[test]
+    fn excessive_length() {
+        StorePath::from_bytes(b"00bgd045z0d4icpbc2yy-aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")
+            .expect_err("must fail");
+    }
+
+    #[test]
+    fn invalid_hash_length() {
+        StorePath::from_bytes(b"00bgd045z0d4icpbc2yy-net-tools-1.60_p20170221182432")
+            .expect_err("must fail");
+    }
+
+    #[test]
+    fn invalid_encoding_hash() {
+        StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48aku4la-net-tools-1.60_p20170221182432")
+            .expect_err("must fail");
+    }
+
+    #[test]
+    fn more_than_just_the_bare_nix_store_path() {
+        StorePath::from_bytes(
+            b"00bgd045z0d4icpbc2yyz4gx48aku4la-net-tools-1.60_p20170221182432/bin/arp",
+        )
+        .expect_err("must fail");
+    }
+
+    #[test]
+    fn no_dash_between_hash_and_name() {
+        StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48ak44lanet-tools-1.60_p20170221182432")
+            .expect_err("must fail");
+    }
+
+    #[test]
+    fn absolute_path() {
+        let example_nix_path_str =
+            "00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432";
+        let nixpath_expected =
+            StorePathRef::from_bytes(example_nix_path_str.as_bytes()).expect("must parse");
+
+        let nixpath_actual = StorePathRef::from_absolute_path(
+            "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432".as_bytes(),
+        )
+        .expect("must parse");
+
+        assert_eq!(nixpath_expected, nixpath_actual);
+
+        assert_eq!(
+            "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432",
+            nixpath_actual.to_absolute_path(),
+        );
+    }
+
+    #[test]
+    fn absolute_path_missing_prefix() {
+        assert_eq!(
+            Error::MissingStoreDir,
+            StorePathRef::from_absolute_path(b"foobar-123").expect_err("must fail")
+        );
+    }
+
+    #[test_case(
+        "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432",
+        (StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432").unwrap(), PathBuf::new())
+    ; "without prefix")]
+    #[test_case(
+        "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432/",
+        (StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432").unwrap(), PathBuf::new())
+    ; "without prefix, but trailing slash")]
+    #[test_case(
+        "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432/bin/arp",
+        (StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432").unwrap(), PathBuf::from("bin/arp"))
+    ; "with prefix")]
+    #[test_case(
+        "/nix/store/00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432/bin/arp/",
+        (StorePath::from_bytes(b"00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432").unwrap(), PathBuf::from("bin/arp/"))
+    ; "with prefix and trailing slash")]
+    fn from_absolute_path_full(s: &str, expected: (StorePath, PathBuf)) {
+        let actual = StorePath::from_absolute_path_full(s).expect("must succeed");
+        assert_eq!(expected, actual);
+    }
+
+    #[test]
+    fn from_absolute_path_errors() {
+        assert_eq!(
+            Error::InvalidLength,
+            StorePath::from_absolute_path_full("/nix/store/").expect_err("must fail")
+        );
+        assert_eq!(
+            Error::InvalidLength,
+            StorePath::from_absolute_path_full("/nix/store/foo").expect_err("must fail")
+        );
+        assert_eq!(
+            Error::MissingStoreDir,
+            StorePath::from_absolute_path_full(
+                "00bgd045z0d4icpbc2yyz4gx48ak44la-net-tools-1.60_p20170221182432"
+            )
+            .expect_err("must fail")
+        );
+    }
+}
diff --git a/tvix/nix-compat/src/store_path/utils.rs b/tvix/nix-compat/src/store_path/utils.rs
new file mode 100644
index 000000000000..0b75ef50647c
--- /dev/null
+++ b/tvix/nix-compat/src/store_path/utils.rs
@@ -0,0 +1,296 @@
+use crate::nixbase32;
+use crate::nixhash::{CAHash, NixHash};
+use crate::store_path::{Error, StorePathRef, DIGEST_SIZE, STORE_DIR};
+use sha2::{Digest, Sha256};
+use std::io::Write;
+use thiserror;
+
+/// Errors that can occur when creating a content-addressed store path.
+///
+/// This wraps the main [crate::store_path::Error]..
+#[derive(Debug, PartialEq, Eq, thiserror::Error)]
+pub enum BuildStorePathError {
+    #[error("Invalid Store Path: {0}")]
+    InvalidStorePath(Error),
+    /// This error occurs when we have references outside the SHA-256 +
+    /// Recursive case. The restriction comes from upstream Nix. It may be
+    /// lifted at some point but there isn't a pressing need to anticipate that.
+    #[error("References were not supported as much as requested")]
+    InvalidReference(),
+}
+
+/// compress_hash takes an arbitrarily long sequence of bytes (usually
+/// a hash digest), and returns a sequence of bytes of length
+/// OUTPUT_SIZE.
+///
+/// It's calculated by rotating through the bytes in the output buffer
+/// (zero- initialized), and XOR'ing with each byte of the passed
+/// input. It consumes 1 byte at a time, and XOR's it with the current
+/// value in the output buffer.
+///
+/// This mimics equivalent functionality in C++ Nix.
+pub fn compress_hash<const OUTPUT_SIZE: usize>(input: &[u8]) -> [u8; OUTPUT_SIZE] {
+    let mut output = [0; OUTPUT_SIZE];
+
+    for (ii, ch) in input.iter().enumerate() {
+        output[ii % OUTPUT_SIZE] ^= ch;
+    }
+
+    output
+}
+
+/// This builds a store path, by calculating the text_hash_string of either a
+/// derivation or a literal text file that may contain references.
+/// If you don't want to have to pass the entire contents, you might want to use
+/// [build_ca_path] instead.
+pub fn build_text_path<S: AsRef<str>, I: IntoIterator<Item = S>, C: AsRef<[u8]>>(
+    name: &str,
+    content: C,
+    references: I,
+) -> Result<StorePathRef<'_>, BuildStorePathError> {
+    // produce the sha256 digest of the contents
+    let content_digest = Sha256::new_with_prefix(content).finalize().into();
+
+    build_ca_path(name, &CAHash::Text(content_digest), references, false)
+}
+
+/// This builds a store path from a [CAHash] and a list of references.
+pub fn build_ca_path<'a, S: AsRef<str>, I: IntoIterator<Item = S>>(
+    name: &'a str,
+    ca_hash: &CAHash,
+    references: I,
+    self_reference: bool,
+) -> Result<StorePathRef<'a>, BuildStorePathError> {
+    // self references are only allowed for CAHash::Nar(NixHash::Sha256(_)).
+    if self_reference {
+        let CAHash::Nar(NixHash::Sha256(_)) = ca_hash else {
+            return Err(BuildStorePathError::InvalidReference());
+        };
+    }
+
+    let (ty, hash) = match &ca_hash {
+        CAHash::Text(ref digest) => (
+            make_references_string("text", references, false),
+            NixHash::Sha256(*digest),
+        ),
+        CAHash::Nar(NixHash::Sha256(ref digest)) => (
+            make_references_string("source", references, self_reference),
+            NixHash::Sha256(*digest),
+        ),
+
+        // for all other CAHash::Nar, another custom scheme is used.
+        CAHash::Nar(ref hash) => {
+            if references.into_iter().next().is_some() {
+                return Err(BuildStorePathError::InvalidReference());
+            }
+
+            (
+                "output:out".to_string(),
+                NixHash::Sha256(
+                    Sha256::new_with_prefix(format!("fixed:out:r:{}:", hash.to_nix_hex_string()))
+                        .finalize()
+                        .into(),
+                ),
+            )
+        }
+        // CaHash::Flat is using something very similar, except the `r:` prefix.
+        CAHash::Flat(ref hash) => {
+            if references.into_iter().next().is_some() {
+                return Err(BuildStorePathError::InvalidReference());
+            }
+
+            (
+                "output:out".to_string(),
+                NixHash::Sha256(
+                    Sha256::new_with_prefix(format!("fixed:out:{}:", hash.to_nix_hex_string()))
+                        .finalize()
+                        .into(),
+                ),
+            )
+        }
+    };
+
+    build_store_path_from_fingerprint_parts(&ty, &hash, name)
+        .map_err(BuildStorePathError::InvalidStorePath)
+}
+
+/// For given NAR sha256 digest and name, return the new [StorePathRef] this
+/// would have, or an error, in case the name is invalid.
+pub fn build_nar_based_store_path<'a>(
+    nar_sha256_digest: &[u8; 32],
+    name: &'a str,
+) -> Result<StorePathRef<'a>, BuildStorePathError> {
+    let nar_hash_with_mode = CAHash::Nar(NixHash::Sha256(nar_sha256_digest.to_owned()));
+
+    build_ca_path(name, &nar_hash_with_mode, Vec::<String>::new(), false)
+}
+
+/// This builds an input-addressed store path.
+///
+/// Input-addresed store paths are always derivation outputs, the "input" in question is the
+/// derivation and its closure.
+pub fn build_output_path<'a>(
+    drv_hash: &NixHash,
+    output_name: &str,
+    output_path_name: &'a str,
+) -> Result<StorePathRef<'a>, Error> {
+    build_store_path_from_fingerprint_parts(
+        &(String::from("output:") + output_name),
+        drv_hash,
+        output_path_name,
+    )
+}
+
+/// This builds a store path from fingerprint parts.
+/// Usually, that function is used from [build_text_path] and
+/// passed a "text hash string" (starting with "text:" as fingerprint),
+/// but other fingerprints starting with "output:" are also used in Derivation
+/// output path calculation.
+///
+/// The fingerprint is hashed with sha256, its digest is compressed to 20 bytes,
+/// and nixbase32-encoded (32 characters).
+fn build_store_path_from_fingerprint_parts<'a>(
+    ty: &str,
+    hash: &NixHash,
+    name: &'a str,
+) -> Result<StorePathRef<'a>, Error> {
+    let digest: [u8; DIGEST_SIZE] = compress_hash(&{
+        let mut h = Sha256::new();
+        write!(h, "{ty}:{}:{STORE_DIR}:{name}", hash.to_nix_hex_string()).unwrap();
+        h.finalize()
+    });
+
+    // name validation happens in here.
+    StorePathRef::from_name_and_digest(name, &digest)
+}
+
+/// This contains the Nix logic to create "text hash strings", which are used
+/// in `builtins.toFile`, as well as in Derivation Path calculation.
+///
+/// A text hash is calculated by concatenating the following fields, separated by a `:`:
+///
+///  - text
+///  - references, individually joined by `:`
+///  - the nix_hash_string representation of the sha256 digest of some contents
+///  - the value of `storeDir`
+///  - the name
+fn make_references_string<S: AsRef<str>, I: IntoIterator<Item = S>>(
+    ty: &str,
+    references: I,
+    self_ref: bool,
+) -> String {
+    let mut s = String::from(ty);
+
+    for reference in references {
+        s.push(':');
+        s.push_str(reference.as_ref());
+    }
+
+    if self_ref {
+        s.push_str(":self");
+    }
+
+    s
+}
+
+/// Nix placeholders (i.e. values returned by `builtins.placeholder`)
+/// are used to populate outputs with paths that must be
+/// string-replaced with the actual placeholders later, at runtime.
+///
+/// The actual placeholder is basically just a SHA256 hash encoded in
+/// cppnix format.
+pub fn hash_placeholder(name: &str) -> String {
+    let digest = Sha256::new_with_prefix(format!("nix-output:{}", name)).finalize();
+
+    format!("/{}", nixbase32::encode(&digest))
+}
+
+#[cfg(test)]
+mod test {
+    use hex_literal::hex;
+
+    use super::*;
+    use crate::nixhash::{CAHash, NixHash};
+
+    #[test]
+    fn build_text_path_with_zero_references() {
+        // This hash should match `builtins.toFile`, e.g.:
+        //
+        // nix-repl> builtins.toFile "foo" "bar"
+        // "/nix/store/vxjiwkjkn7x4079qvh1jkl5pn05j2aw0-foo"
+
+        let store_path = build_text_path("foo", "bar", Vec::<String>::new())
+            .expect("build_store_path() should succeed");
+
+        assert_eq!(
+            store_path.to_absolute_path().as_str(),
+            "/nix/store/vxjiwkjkn7x4079qvh1jkl5pn05j2aw0-foo"
+        );
+    }
+
+    #[test]
+    fn build_text_path_with_non_zero_references() {
+        // This hash should match:
+        //
+        // nix-repl> builtins.toFile "baz" "${builtins.toFile "foo" "bar"}"
+        // "/nix/store/5xd714cbfnkz02h2vbsj4fm03x3f15nf-baz"
+
+        let inner = build_text_path("foo", "bar", Vec::<String>::new())
+            .expect("path_with_references() should succeed");
+        let inner_path = inner.to_absolute_path();
+
+        let outer = build_text_path("baz", &inner_path, vec![inner_path.as_str()])
+            .expect("path_with_references() should succeed");
+
+        assert_eq!(
+            outer.to_absolute_path().as_str(),
+            "/nix/store/5xd714cbfnkz02h2vbsj4fm03x3f15nf-baz"
+        );
+    }
+
+    #[test]
+    fn build_sha1_path() {
+        let outer = build_ca_path(
+            "bar",
+            &CAHash::Nar(NixHash::Sha1(hex!(
+                "0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33"
+            ))),
+            Vec::<String>::new(),
+            false,
+        )
+        .expect("path_with_references() should succeed");
+
+        assert_eq!(
+            outer.to_absolute_path().as_str(),
+            "/nix/store/mp57d33657rf34lzvlbpfa1gjfv5gmpg-bar"
+        );
+    }
+
+    #[test]
+    fn build_store_path_with_non_zero_references() {
+        // This hash should match:
+        //
+        // nix-repl> builtins.toFile "baz" "${builtins.toFile "foo" "bar"}"
+        // "/nix/store/5xd714cbfnkz02h2vbsj4fm03x3f15nf-baz"
+        //
+        // $ nix store make-content-addressed /nix/store/5xd714cbfnkz02h2vbsj4fm03x3f15nf-baz
+        // rewrote '/nix/store/5xd714cbfnkz02h2vbsj4fm03x3f15nf-baz' to '/nix/store/s89y431zzhmdn3k8r96rvakryddkpv2v-baz'
+        let outer = build_ca_path(
+            "baz",
+            &CAHash::Nar(NixHash::Sha256(
+                nixbase32::decode(b"1xqkzcb3909fp07qngljr4wcdnrh1gdam1m2n29i6hhrxlmkgkv1")
+                    .expect("nixbase32 should decode")
+                    .try_into()
+                    .expect("should have right len"),
+            )),
+            vec!["/nix/store/dxwkwjzdaq7ka55pkk252gh32bgpmql4-foo"],
+            false,
+        )
+        .expect("path_with_references() should succeed");
+
+        assert_eq!(
+            outer.to_absolute_path().as_str(),
+            "/nix/store/s89y431zzhmdn3k8r96rvakryddkpv2v-baz"
+        );
+    }
+}
diff --git a/tvix/nix-compat/testdata/narinfo.zst b/tvix/nix-compat/testdata/narinfo.zst
new file mode 100644
index 000000000000..361a422da86e
--- /dev/null
+++ b/tvix/nix-compat/testdata/narinfo.zst
Binary files differ