From 57cdebff8e975d00154257d83d0848a486f40452 Mon Sep 17 00:00:00 2001 From: Vincent Ambo Date: Mon, 12 Feb 2024 00:04:04 +0300 Subject: docs(web/tvl/blog): editing pass over 2024-02 tvix update Change-Id: I62f74a0b53dd655029006052b4a5e00254c4a381 Reviewed-on: https://cl.tvl.fyi/c/depot/+/10805 Tested-by: BuildkiteCI Reviewed-by: flokli --- web/tvl/blog/2024-02-tvix-update.md | 209 ++++++++++++++++++++---------------- 1 file changed, 116 insertions(+), 93 deletions(-) (limited to 'web/tvl') diff --git a/web/tvl/blog/2024-02-tvix-update.md b/web/tvl/blog/2024-02-tvix-update.md index 3b286a17e2f4..6f7de626410c 100644 --- a/web/tvl/blog/2024-02-tvix-update.md +++ b/web/tvl/blog/2024-02-tvix-update.md @@ -1,58 +1,74 @@ -It's already been way too long since the last update here. It doesn't mean -nothing has moved forward since, in fact a lot of things happened, but while -there's been a talk about Tvix at [NixCon 2023][nixcon2023], as well as a -[Nix Developer Dialogue Interview][nix-dev-dialogues-tvix], we never found the -time to write something down - time to change that :-) +We've now been working on our rewrite of Nix, [Tvix][], for a little more than +two years. + +Our last written update was in September 2023, and although we did publish a +couple of things in the meantime (flokli's talk on Tvix at [NixCon +2023][nixcon2023], our interview at the [Nix Developer +Dialogues][nix-dev-dialogues-tvix], or tazjin's [talk on +tvix-eval][tvix-eval-ru] (in Russian)), we never found the time to write +something down. + +In the meantime a lot of stuff has happened though, so it's time to change that +:-) + +Note: This blog post is intended for a technical audience that is already +intimately familiar with Nix, and knows what things like derivations or store +paths are. If you're new to Nix, this will not make a lot of sense to you! ## Evaluation regression testing -Most of the Evaluator work has been driven by evaluating nixpkgs, and ensuring -we "evaluate it the same", which ultimately means we produce the same builds and -end up with the same store paths. - -We don't build things yet, but at least for nixpkgs (which doesn't do any IFD), -it's possible to just peek at the `outPath` (and `drvPath`) of a package, and -compare the calculated store path(s) with what Nix produces to determine if they -use the same build recipe (for the package as well as all of its dependencies) -[^1]. - -We added some "integration tests" to our CI ensuring we keep getting the same -output and drv paths as Nix, and already have quite complicated expressions in -there and passing, like firefox, which exercises some of the more hairy bits in -nixpkgs (like cross-compilation infrastructure for WASM). Yay! - -Even though we're not getting into very fine-grained optimization until we're -sure Tvix evaluates nixpkgs in a compatible fashion, we still want to have an -idea about evaluation performance, and how it changes over time. -To do this, we extended our benchmarks and wired them into -[Windtunnel][windtunnel], which now regularly runs benchmarks and provides -graphs of how the benchmark results change over time, from commit to commit. - -In the future, we plan to extend this to also run this as a part of code review, -before changes are applied to our main branch. This will make it easier to see -changes in performance right in the web interface, without having -to do a manual benchmark locally before and after the change. - -## `builtins.derivation[Strict]`, ATerms and output path calculation -These two are obviously needed to compare any derivation. As an interesting side -note, in Nixcpp, the former is defined by a piece of -[bundled-in-Nix `.nix` code][nixcpp-builtins-derivation], that massages some -parameters around and then calls the "native" `derivationStrict` - so -implementing them required having the necessary tooling in Tvix to have builtins -defined in `.nix` source code). - -`builtins.derivation[Strict]` returns an attribute set with the previously -mentioned `outPath` and `drvPath` paths. Implementing that correctly required -implementing output path calculation the same way as Nix does (bit-by-bit). - -Very little of how the output path calculation works is documented anywhere -in Nixlang. It uses a subset of [ATerm][aterm] internally, produces -"fingerprints" containing hashes of these ATerms, which are then hashed again. -The intermediate hashes are not printed out anywhere (except if you [patch nix] -[nixcpp-patch-hashes] to do so). + +Most of the evaluator work has been driven by evaluating `nixpkgs`, and ensuring +that we produce the same derivations, and that their build results end up in the +same store paths. + +Builds are not hooked up all the way to the evaluator yet, but for Nix code +without IFD (such as `nixpkgs`!) we can verify this property without building. +An evaluated Nix derivation's `outPath` (and `drvPath`) can be compared with +what C++ Nix produces for the same code, to determine whether we evaluated the +package (and all of its dependencies!) correctly [^1]. + +We added integration tests in CI that ensure that the paths we calculate match +C++ Nix, and are successfully evaluating fairly complicated expressions in them. +For example, we test against the Firefox derivation, which exercises some of the +more hairy bits in `nixpkgs` (like WASM cross-compilation infrastructure). Yay! + +Although we're avoiding fine-grained optimization until we're sure Tvix +evaluates all of `nixpkgs` correctly, we still want to have an idea about +evaluation performance and how our work affects it over time. + +For this we extended our benchmark suite and integrated it with +[Windtunnel][windtunnel], which now regularly runs benchmarks and provides a +view into how the timings change from commit to commit. + +In the future, we plan to run this as a part of code review, before changes are +applied to our canonical branch, to provide this as an additional signal to +authors and reviewers without having to run the benchmarks manually. + +## ATerms, output path calculation, and `builtins.derivation` + +We've implemented all of these features, which comprise the components needed to +construct derivations in the Nix language, and to allow us to perform the path +comparisons we mentioned before. + +As an interesting side note, in C++ Nix `builtins.derivation` is not actually a +builtin! It is a piece of [bundled Nix code][nixcpp-builtins-derivation], that +massages some parameters and then calls the *actual* builtin: +`derivationStrict`. We've decided to keep this setup, and implemented support in +in Tvix to have builtins defined in `.nix` source code. + +These builtins return attribute sets with the previously mentioned `outPath` and +`drvPath` fields. Implementing them correctly meant that we needed to implement +output path calculation *exactly* the same way as Nix does (bit-by-bit). + +Very little of how this output path calculation works is documented anywhere in +C++ Nix. It uses a subset of [ATerm][aterm] internally, produces "fingerprints" +containing hashes of these ATerms, which are then hashed again. The intermediate +hashes are not printed out anywhere (except if you [patch +Nix][nixcpp-patch-hashes] to do so). We already did parts of this correctly while starting this work on [go-nix][go-nix-outpath] some while ago, but found some more edge cases and -ultimately came up with a nicer interface than there. +ultimately came up with a nicer interface for Tvix. All the Derivation internal data model, ATerm serialization and output path calculation have been sliced out into a more general-purpose @@ -166,7 +182,7 @@ that deal with path access in `tvix-eval`, like: - `builtins.readFile` We also recently started working on more complicated builtins like -`builtins.filterSource` and `builtins.path`, which are also used in nixpkgs. +`builtins.filterSource` and `builtins.path`, which are also used in `nixpkgs`. Both import a path into the store, and allow passing a Nix expression that's used as a filter function for each path. `builtins.path` can also ensuring the @@ -181,6 +197,7 @@ Getting the abstractions right there required some back-and-forth, but the remaining changes should land quite soon. ## Catchables / tryEval + As you may know, Nix has a limited exception system for dealing with user-generated errors: `builtins.tryEval` can be used to detect if an expression fails (if `builtins.throw` or `assert` are used to generate it). This feature @@ -202,7 +219,7 @@ the (in this case) VM runtime is not guaranteed. Previously, `builtins.tryEval` (which is implemented in Rust and can access VM internals) just allowed the VM to recover from certain kinds of errors. This proved to be insufficient as it [blew up as soon as a `builtins.tryEval`-ed thunk -is forced again][tryeval-infrec]—extra bookkeeping was needed. As a +is forced again][tryeval-infrec]–extra bookkeeping was needed. As a solution, we now store thunk evaluation errors that can be recovered from as `Value::Catchable` which mitigates this problem. @@ -215,31 +232,34 @@ regressions if or rather when we touch this system again. We already have some ideas for replacing the `Catchable` value type with a cleaner representation. ## String contexts -For a long time, we had the [working theory][refscan-string-contexts] -of being able to get away with not implementing string contexts, but -instead do reference scanning on a set of "known paths" (and not implement + +For a long time, we had the [working theory][refscan-string-contexts] that we +could get away with not implementing string contexts, and instead do reference +scanning on a set of "known paths" (and not implement `builtins.unsafeDiscardStringContext`). -Unfortunately, we ultimately discovered we won't be able to do that, mostly due to a -[bug in Nix][string-contexts-nix-bug] that's worked around in the nixpkgs' -`stdenv.mkDerivation` implementation, but impossible to fix in Nix without -breaking all hashes. +Unfortunately, we discovered that while this is *conceptually* true, due to a +[bug in Nix][string-contexts-nix-bug] that's worked around in the +`stdenv.mkDerivation` implementation, we can't currently do this and calculate +the same hashes. -So we recently added support for string contexts into our `NixString` -implementation, implemented the (`unsafeDiscardStringContext`, `getContext`) -builtins, as well as some more unit tests that introspect string context -behaviour of various builtins. +Because hash compatibility is important for us at this point, we bit the bullet +and added support for string contexts into our `NixString` implementation, +implemented the context-related builtins, and added more unit tests that verify +string context behaviour of various builtins. ## Strings as bstr -C++ nix uses C-style zero-terminated char pointers to represent strings -internally - however, until recently, Tvix has used Rust `String` and `str` for -string values. Since those are required to be valid utf-8, we haven't been able -to properly represent all the string values that Nix supports. -We recently converted our internal representation to `BString`, which allows -treating a `Vec` as a "morally-string-like" value. +C++ Nix uses C-style zero-terminated strings internally - however, until +recently, Tvix has used Rust `String` and `str` for string values. Since those +are required to be valid utf-8, we haven't been able to properly represent all +the string values that Nix supports. + +We recently converted our internal representation to byte strings, which allows +us to treat a `Vec` as a "string-like" value. ## JSON/TOML/XML + We added support for the `toJSON`, `toXML`, `fromJSON` and `fromTOML` builtins. `toXML` is particularly exciting, as it's the only format that allows expressing @@ -247,38 +267,40 @@ We added support for the `toJSON`, `toXML`, `fromJSON` and `fromTOML` builtins. we can now include these in our unit test suite (and pass, yay!). ## Builder protocol, drv->builder -Some work went into the Builder protocol, and how Tvix represents builds -internally. -Nix uses Derivations (in A-Term) as nodes in its build graph, but it refers to -other store paths used in that build simply by these store paths *only*, and -doesn't encode their expected *content hashes* anywhere. +We've been working on the builder protocol, and Tvix's internal build +representation. -In Nix, this poses a big problem as soon as these builds are scheduled on remote -builders: Builds scheduled to a builder are not truly hermetic, but rely on that -referred store path to actually have the same contents as the machine -orchestrating the build (or at least very similar). +Nix uses derivations (encoded in ATerm) as nodes in its build graph, but it +refers to other store paths used in that build by these store paths *only*. As +mentioned before, store paths only address the inputs - and not the content. -If a package is not binary reproducible, this can lead to so-called +This poses a big problem in Nix as soon as builds are scheduled on remote +builders: There is no guarantee that files at the same store path on the remote +builder actually have the same contents as on the machine orchestrating the +build. If a package is not binary reproducible, this can lead to so-called [frankenbuilds][frankenbuild]. -Even ignoring this problem, we still rely on state (the exact contents of that -output in the remote builders' Nix Store), and making them appear there (if they -can't be substituted) requires the one scheduling the build to copy them over, -and be a trusted user. -We wanted to eliminate this hermiticity problem in the internal data model -that Tvix uses to manage builds, by not relying on external state, and instead -explicitly tracking both build inputs as well as produced outputs by their -contents, too. +This also introduces a dependency on the state that's present on the remote +builder machine: Whatever is in its store and matches the paths will be used, +even if it was maliciously placed there. + +To eliminate this hermiticity problem and increase the integrity of builds, +we've decided to use content-addressing in the builder protocol. + +We're currently hacking on this at [Thaigersprint](https://thaigersprint.org/) +and might have some more news to share soon! + +-------------- ---- +That's it for now, try out Tvix and hit us up on IRC or on our mailing list if +you run into any snags, or have any questions. -That's it for now! +เจอกันนะ :) ---- -[^1]: Why this means it's using the same build recipes is due to how the output - path calculation for input-addressed paths works, more of that in the - next section. +[^1]: We know that we calculated all dependencies correctly because of how their + hashes are included in the hashes of their dependents, and so on. More on + path calculation and input-addressed paths in the next section! [^2]: That's the same reason why `builtins.derivation[Strict]` also lives in `tvix-glue`, not in `tvix-eval`. [^3]: See [nix-casync](https://discourse.nixos.org/t/nix-casync-a-more-efficient-way-to-store-and-substitute-nix-store-paths/16539) @@ -296,6 +318,7 @@ That's it for now! [nix-compat-narinfo]: https://docs.tvix.dev/rust/nix_compat/narinfo/index.html [nix-dev-dialogues-tvix]: https://www.youtube.com/watch?v=ZYG3T4l8RU8 [nixcon2023]: https://www.youtube.com/watch?v=j67prAPYScY +[tvix-eval-ru]: https://tazj.in/blog/tvix-eval-talk-2023 [nixcpp-builtins-derivation]: https://github.com/NixOS/nix/blob/49cf090cb2f51d6935756a6cf94d568cab063f81/src/libexpr/primops/derivation.nix#L4 [nixcpp-patch-hashes]: https://github.com/adisbladis/nix/tree/hash-tracing [refscan-string-contexts]: https://inbox.tvl.su/depot/20230316120039.j4fkp3puzrtbjcpi@tp/T/#t -- cgit 1.4.1