Add support for passing structured data to builders

Previously, all derivation attributes had to be coerced into strings
so that they could be passed via the environment. This is lossy
(e.g. lists get flattened, necessitating configureFlags
vs. configureFlagsArray, of which the latter cannot be specified as an
attribute), doesn't support attribute sets at all, and has size
limitations (necessitating hacks like passAsFile).

This patch adds a new mode for passing attributes to builders, namely
encoded as a JSON file ".attrs.json" in the current directory of the
builder. This mode is activated via the special attribute

  __structuredAttrs = true;

(The idea is that one day we can set this in stdenv.mkDerivation.)

For example,

  stdenv.mkDerivation {
    __structuredAttrs = true;
    name = "foo";
    buildInputs = [ pkgs.hello pkgs.cowsay ];
    doCheck = true;
    hardening.format = false;
  }

results in a ".attrs.json" file containing (sans the indentation):

  {
    "buildInputs": [],
    "builder": "/nix/store/ygl61ycpr2vjqrx775l1r2mw1g2rb754-bash-4.3-p48/bin/bash",
    "configureFlags": [
      "--with-foo",
      "--with-bar=1 2"
    ],
    "doCheck": true,
    "hardening": {
      "format": false
    },
    "name": "foo",
    "nativeBuildInputs": [
      "/nix/store/10h6li26i7g6z3mdpvra09yyf10mmzdr-hello-2.10",
      "/nix/store/4jnvjin0r6wp6cv1hdm5jbkx3vinlcvk-cowsay-3.03"
    ],
    "propagatedBuildInputs": [],
    "propagatedNativeBuildInputs": [],
    "stdenv": "/nix/store/f3hw3p8armnzy6xhd4h8s7anfjrs15n2-stdenv",
    "system": "x86_64-linux"
  }

"passAsFile" is ignored in this mode because it's not needed - large
strings are included directly in the JSON representation.

It is up to the builder to do something with the JSON
representation. For example, in bash-based builders, lists/attrsets of
string values could be mapped to bash (associative) arrays.

4 files changed, 130 insertions, 58 deletions

diff --git a/src/libexpr/eval.cc b/src/libexpr/eval.cc
index 5f9f5bac1ed6..d418ab4e43aa 100644
--- a/src/libexpr/eval.cc
+++ b/src/libexpr/eval.cc
@@ -291,6 +291,8 @@ EvalState::EvalState(const Strings & _searchPath, ref<Store> store)
     , sToString(symbols.create("__toString"))
     , sRight(symbols.create("right"))
     , sWrong(symbols.create("wrong"))
+    , sStructuredAttrs(symbols.create("__structuredAttrs"))
+    , sBuilder(symbols.create("builder"))
     , store(store)
     , baseEnv(allocEnv(128))
     , staticBaseEnv(false, 0)
diff --git a/src/libexpr/eval.hh b/src/libexpr/eval.hh
index 195cb0db3acc..46d5a1cc866c 100644
--- a/src/libexpr/eval.hh
+++ b/src/libexpr/eval.hh
@@ -68,7 +68,7 @@ public:
     const Symbol sWith, sOutPath, sDrvPath, sType, sMeta, sName, sValue,
         sSystem, sOverrides, sOutputs, sOutputName, sIgnoreNulls,
         sFile, sLine, sColumn, sFunctor, sToString,
-        sRight, sWrong;
+        sRight, sWrong, sStructuredAttrs, sBuilder;
     Symbol sDerivationNix;
 
     /* If set, force copying files to the Nix store even if they
diff --git a/src/libexpr/primops.cc b/src/libexpr/primops.cc
index d8dbedcaf89e..5a570cefb2fa 100644
--- a/src/libexpr/primops.cc
+++ b/src/libexpr/primops.cc
@@ -8,6 +8,7 @@
 #include "names.hh"
 #include "store-api.hh"
 #include "util.hh"
+#include "json.hh"
 #include "value-to-json.hh"
 #include "value-to-xml.hh"
 #include "primops.hh"
@@ -474,6 +475,13 @@ static void prim_derivationStrict(EvalState & state, const Pos & pos, Value * *
         throw;
     }
 
+    /* Check whether attributes should be passed as a JSON file. */
+    std::ostringstream jsonBuf;
+    std::unique_ptr<JSONObject> jsonObject;
+    attr = args[0]->attrs->find(state.sStructuredAttrs);
+    if (attr != args[0]->attrs->end() && state.forceBool(*attr->value, pos))
+        jsonObject = std::make_unique<JSONObject>(jsonBuf);
+
     /* Check whether null attributes should be ignored. */
     bool ignoreNulls = false;
     attr = args[0]->attrs->find(state.sIgnoreNulls);
@@ -491,24 +499,48 @@ static void prim_derivationStrict(EvalState & state, const Pos & pos, Value * *
     StringSet outputs;
     outputs.insert("out");
 
-    for (auto & i : *args[0]->attrs) {
-        if (i.name == state.sIgnoreNulls) continue;
-        string key = i.name;
+    for (auto & i : args[0]->attrs->lexicographicOrder()) {
+        if (i->name == state.sIgnoreNulls) continue;
+        string key = i->name;
         Activity act(*logger, lvlVomit, format("processing attribute ‘%1%’") % key);
 
+        auto handleHashMode = [&](const std::string & s) {
+            if (s == "recursive") outputHashRecursive = true;
+            else if (s == "flat") outputHashRecursive = false;
+            else throw EvalError("invalid value ‘%s’ for ‘outputHashMode’ attribute, at %s", s, posDrvName);
+        };
+
+        auto handleOutputs = [&](const Strings & ss) {
+            outputs.clear();
+            for (auto & j : ss) {
+                if (outputs.find(j) != outputs.end())
+                    throw EvalError(format("duplicate derivation output ‘%1%’, at %2%") % j % posDrvName);
+                /* !!! Check whether j is a valid attribute
+                   name. */
+                /* Derivations cannot be named ‘drv’, because
+                   then we'd have an attribute ‘drvPath’ in
+                   the resulting set. */
+                if (j == "drv")
+                    throw EvalError(format("invalid derivation output name ‘drv’, at %1%") % posDrvName);
+                outputs.insert(j);
+            }
+            if (outputs.empty())
+                throw EvalError(format("derivation cannot have an empty set of outputs, at %1%") % posDrvName);
+        };
+
         try {
 
             if (ignoreNulls) {
-                state.forceValue(*i.value);
-                if (i.value->type == tNull) continue;
+                state.forceValue(*i->value);
+                if (i->value->type == tNull) continue;
             }
 
             /* The `args' attribute is special: it supplies the
                command-line arguments to the builder. */
             if (key == "args") {
-                state.forceList(*i.value, pos);
-                for (unsigned int n = 0; n < i.value->listSize(); ++n) {
-                    string s = state.coerceToString(posDrvName, *i.value->listElems()[n], context, true);
+                state.forceList(*i->value, pos);
+                for (unsigned int n = 0; n < i->value->listSize(); ++n) {
+                    string s = state.coerceToString(posDrvName, *i->value->listElems()[n], context, true);
                     drv.args.push_back(s);
                 }
             }
@@ -516,39 +548,51 @@ static void prim_derivationStrict(EvalState & state, const Pos & pos, Value * *
             /* All other attributes are passed to the builder through
                the environment. */
             else {
-                string s = state.coerceToString(posDrvName, *i.value, context, true);
-                drv.env[key] = s;
-                if (key == "builder") drv.builder = s;
-                else if (i.name == state.sSystem) drv.platform = s;
-                else if (i.name == state.sName) {
-                    drvName = s;
-                    printMsg(lvlVomit, format("derivation name is ‘%1%’") % drvName);
-                }
-                else if (key == "outputHash") outputHash = s;
-                else if (key == "outputHashAlgo") outputHashAlgo = s;
-                else if (key == "outputHashMode") {
-                    if (s == "recursive") outputHashRecursive = true;
-                    else if (s == "flat") outputHashRecursive = false;
-                    else throw EvalError(format("invalid value ‘%1%’ for ‘outputHashMode’ attribute, at %2%") % s % posDrvName);
-                }
-                else if (key == "outputs") {
-                    Strings tmp = tokenizeString<Strings>(s);
-                    outputs.clear();
-                    for (auto & j : tmp) {
-                        if (outputs.find(j) != outputs.end())
-                            throw EvalError(format("duplicate derivation output ‘%1%’, at %2%") % j % posDrvName);
-                        /* !!! Check whether j is a valid attribute
-                           name. */
-                        /* Derivations cannot be named ‘drv’, because
-                           then we'd have an attribute ‘drvPath’ in
-                           the resulting set. */
-                        if (j == "drv")
-                            throw EvalError(format("invalid derivation output name ‘drv’, at %1%") % posDrvName);
-                        outputs.insert(j);
+
+                if (jsonObject) {
+
+                    if (i->name == state.sStructuredAttrs) continue;
+
+                    auto placeholder(jsonObject->placeholder(key));
+                    printValueAsJSON(state, true, *i->value, placeholder, context);
+
+                    if (i->name == state.sBuilder)
+                        drv.builder = state.forceString(*i->value, context, posDrvName);
+                    else if (i->name == state.sSystem)
+                        drv.platform = state.forceStringNoCtx(*i->value, posDrvName);
+                    else if (i->name == state.sName)
+                        drvName = state.forceStringNoCtx(*i->value, posDrvName);
+                    else if (key == "outputHash")
+                        outputHash = state.forceStringNoCtx(*i->value, posDrvName);
+                    else if (key == "outputHashAlgo")
+                        outputHashAlgo = state.forceStringNoCtx(*i->value, posDrvName);
+                    else if (key == "outputHashMode")
+                        handleHashMode(state.forceStringNoCtx(*i->value, posDrvName));
+                    else if (key == "outputs") {
+                        /* Require ‘outputs’ to be a list of strings. */
+                        state.forceList(*i->value, posDrvName);
+                        Strings ss;
+                        for (unsigned int n = 0; n < i->value->listSize(); ++n)
+                            ss.emplace_back(state.forceStringNoCtx(*i->value->listElems()[n], posDrvName));
+                        handleOutputs(ss);
+                    }
+
+                } else {
+                    auto s = state.coerceToString(posDrvName, *i->value, context, true);
+                    drv.env.emplace(key, s);
+                    if (i->name == state.sBuilder) drv.builder = s;
+                    else if (i->name == state.sSystem) drv.platform = s;
+                    else if (i->name == state.sName) {
+                        drvName = s;
+                        printMsg(lvlVomit, format("derivation name is ‘%1%’") % drvName);
                     }
-                    if (outputs.empty())
-                        throw EvalError(format("derivation cannot have an empty set of outputs, at %1%") % posDrvName);
+                    else if (key == "outputHash") outputHash = s;
+                    else if (key == "outputHashAlgo") outputHashAlgo = s;
+                    else if (key == "outputHashMode") handleHashMode(s);
+                    else if (key == "outputs")
+                        handleOutputs(tokenizeString<Strings>(s));
                 }
+
             }
 
         } catch (Error & e) {
@@ -558,6 +602,11 @@ static void prim_derivationStrict(EvalState & state, const Pos & pos, Value * *
         }
     }
 
+    if (jsonObject) {
+        jsonObject.reset();
+        drv.env.emplace("__json", jsonBuf.str());
+    }
+
     /* Everything in the context of the strings in the derivation
        attributes should be added as dependencies of the resulting
        derivation. */
diff --git a/src/libstore/build.cc b/src/libstore/build.cc
index 6250de13cbb2..d76c8d1727fc 100644
--- a/src/libstore/build.cc
+++ b/src/libstore/build.cc
@@ -865,6 +865,9 @@ private:
     /* Fill in the environment for the builder. */
     void initEnv();
 
+    /* Write a JSON file containing the derivation attributes. */
+    void writeStructuredAttrs();
+
     /* Make a file owned by the builder. */
     void chownToBuilder(const Path & path);
 
@@ -1726,13 +1729,15 @@ void DerivationGoal::startBuilder()
     tmpDirInSandbox = useChroot ? canonPath("/tmp", true) + "/nix-build-" + drvName + "-0" : tmpDir;
     chownToBuilder(tmpDir);
 
-    /* Construct the environment passed to the builder. */
-    initEnv();
-
     /* Substitute output placeholders with the actual output paths. */
     for (auto & output : drv->outputs)
         inputRewrites[hashPlaceholder(output.first)] = output.second.path;
 
+    /* Construct the environment passed to the builder. */
+    initEnv();
+
+    writeStructuredAttrs();
+
     /* Handle exportReferencesGraph(), if set. */
     doExportReferencesGraph();
 
@@ -2148,22 +2153,29 @@ void DerivationGoal::initEnv()
     /* The maximum number of cores to utilize for parallel building. */
     env["NIX_BUILD_CORES"] = (format("%d") % settings.buildCores).str();
 
-    /* Add all bindings specified in the derivation via the
-       environments, except those listed in the passAsFile
-       attribute. Those are passed as file names pointing to
-       temporary files containing the contents. */
-    StringSet passAsFile = tokenizeString<StringSet>(get(drv->env, "passAsFile"));
-    int fileNr = 0;
-    for (auto & i : drv->env) {
-        if (passAsFile.find(i.first) == passAsFile.end()) {
-            env[i.first] = i.second;
-        } else {
-            string fn = ".attr-" + std::to_string(fileNr++);
-            Path p = tmpDir + "/" + fn;
-            writeFile(p, i.second);
-            chownToBuilder(p);
-            env[i.first + "Path"] = tmpDirInSandbox + "/" + fn;
+    /* In non-structured mode, add all bindings specified in the
+       derivation via the environments, except those listed in the
+       passAsFile attribute. Those are passed as file names pointing
+       to temporary files containing the contents. Note that
+       passAsFile is ignored in structure mode because it's not
+       needed (attributes are not passed through the environment, so
+       there is no size constraint). */
+    if (!drv->env.count("__json")) {
+
+        StringSet passAsFile = tokenizeString<StringSet>(get(drv->env, "passAsFile"));
+        int fileNr = 0;
+        for (auto & i : drv->env) {
+            if (passAsFile.find(i.first) == passAsFile.end()) {
+                env[i.first] = i.second;
+            } else {
+                string fn = ".attr-" + std::to_string(fileNr++);
+                Path p = tmpDir + "/" + fn;
+                writeFile(p, i.second);
+                chownToBuilder(p);
+                env[i.first + "Path"] = tmpDirInSandbox + "/" + fn;
+            }
         }
+
     }
 
     /* For convenience, set an environment pointing to the top build
@@ -2201,6 +2213,15 @@ void DerivationGoal::initEnv()
 }
 
 
+void DerivationGoal::writeStructuredAttrs()
+{
+    auto json = drv->env.find("__json");
+    if (json == drv->env.end()) return;
+
+    writeFile(tmpDir + "/.attrs.json", rewriteStrings(json->second, inputRewrites));
+}
+
+
 void DerivationGoal::chownToBuilder(const Path & path)
 {
     if (!buildUser) return;