WIP strip strings at compile time

And also render the docs nicely.

I would like to use a markdown AST for this, but to avoid new deps
(lowdown's AST doesn't suffice) I am just doing crude string
manipulations for now.

.github/workflows/ci.yml

@@ -174,7 +174,7 @@ jobs:
         echo "installer-url=file://$GITHUB_WORKSPACE/out" >> "$GITHUB_OUTPUT"
         TARBALL_PATH="$(find "$GITHUB_WORKSPACE/out" -name 'nix*.tar.xz' -print | head -n 1)"
         echo "tarball-path=file://$TARBALL_PATH" >> "$GITHUB_OUTPUT"
-    - uses: cachix/install-nix-action@456688f15bc354bef6d396e4a35f4f89d40bf2b7 # v31.8.2
+    - uses: cachix/install-nix-action@7ec16f2c061ab07b235a7245e06ed46fe9a1cab6 # v31.8.3
       if: ${{ !matrix.experimental-installer }}
       with:
         install_url: ${{ format('{0}/install', steps.installer-tarball-url.outputs.installer-url) }}

doc/manual/package.nix

@@ -58,8 +58,8 @@ mkMesonDerivation (finalAttrs: {
     "man"
   ];
 
-  # Hack for sake of the dev shell
-  passthru.externalNativeBuildInputs = [
+  nativeBuildInputs = [
+    nix-cli
     meson
     ninja
     (lib.getBin lowdown-unsandboxed)
@@ -78,10 +78,6 @@ mkMesonDerivation (finalAttrs: {
     changelog-d
   ];
 
-  nativeBuildInputs = finalAttrs.passthru.externalNativeBuildInputs ++ [
-    nix-cli
-  ];
-
   preConfigure = ''
     chmod u+w ./.version
     echo ${finalAttrs.version} > ./.version

doc/manual/rl-next/json-format-changes.md

@@ -22,7 +22,15 @@ The store path info JSON format has been updated from version 1 to version 2:
   - New: `"ca": {"method": "nar", "hash": {"algorithm": "sha256", "format": "base64", "hash": "EMIJ+giQ..."}}`
   - Still `null` values for input-addressed store objects
 
-Version 1 format is still accepted when reading for backward compatibility.
+- **Structured hash fields**:
+
+  Hash values (`narHash` and `downloadHash`) are now structured JSON objects instead of strings:
+
+  - Old: `"narHash": "sha256:FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="`
+  - New: `"narHash": {"algorithm": "sha256", "format": "base64", "hash": "FePFYIlM..."}`
+  - Same structure applies to `downloadHash` in NAR info contexts
+
+Nix currently only produces, and doesn't consume this format.
 
 **Affected command**: `nix path-info --json`
 

doc/manual/rl-next/s3-storage-class.md

@@ -0,0 +1,21 @@
+---
+synopsis: "S3 binary cache stores now support storage class configuration"
+prs: [14464]
+issues: [7015]
+---
+
+S3 binary cache stores now support configuring the storage class for uploaded objects via the `storage-class` parameter. This allows users to optimize costs by selecting appropriate storage tiers based on access patterns.
+
+Example usage:
+
+```bash
+# Use Glacier storage for long-term archival
+nix copy --to 's3://my-bucket?storage-class=GLACIER' /nix/store/...
+
+# Use Intelligent Tiering for automatic cost optimization
+nix copy --to 's3://my-bucket?storage-class=INTELLIGENT_TIERING' /nix/store/...
+```
+
+The storage class applies to both regular uploads and multipart uploads. When not specified, objects use the bucket's default storage class.
+
+See the [S3 storage classes documentation](https://docs.aws.amazon.com/AmazonS3/latest/userguide/storage-class-intro.html) for available storage classes and their characteristics.

doc/manual/source/SUMMARY.md.in

@@ -29,6 +29,7 @@
   - [Build Trace](store/build-trace.md)
   - [Derivation Resolution](store/resolution.md)
   - [Building](store/building.md)
+  - [Secrets](store/secrets.md)
   - [Store Types](store/types/index.md)
 {{#include ./store/types/SUMMARY.md}}
   - [Appendix: Math notation](store/math-notation.md)

doc/manual/source/protocols/json/schema/store-object-info-v2.yaml

@@ -71,7 +71,7 @@ $defs:
           Note: This field may not be present in all contexts, such as when the path is used as the key and the the store object info the value in map.
 
       narHash:
-        type: string
+        "$ref": "./hash-v1.yaml"
         title: NAR Hash
         description: |
           Hash of the [file system object](@docroot@/store/file-system-object.md) part of the store object when serialized as a [Nix Archive](@docroot@/store/file-system-object/content-address.md#serial-nix-archive).
@@ -229,7 +229,7 @@ $defs:
           > This is an impure "`.narinfo`" field that may not be included in certain contexts.
 
       downloadHash:
-        type: string
+        "$ref": "./hash-v1.yaml"
         title: Download Hash
         description: |
           A digest for the compressed archive itself, as opposed to the data contained within.

doc/manual/source/protocols/nix-archive/index.md

@@ -24,7 +24,7 @@ nar-obj-inner
   | str("type"), str("directory") directory
   ;
 
-regular = [ str("executable") ], str("contents"), str(contents);
+regular = [ str("executable"), str("") ], str("contents"), str(contents);
 
 symlink = str("target"), str(target);
 
@@ -52,4 +52,4 @@ The Nix Archive (NAR) format is also formally described using [Kaitai Struct](ht
 {{#include nar.ksy}}
 ```
 
-The source of the spec can be found [here](https://github.com/nixos/nix/blob/master/src/nix-manual/source/protocols/nix-archive/nar.ksy). Contributions and improvements to the spec are welcomed.
+The source of the spec can be found [here](https://github.com/nixos/nix/blob/master/src/nix-manual/source/protocols/nix-archive/nar.ksy). Contributions and improvements to the spec are welcomed.

doc/manual/source/store/secrets.md

@@ -0,0 +1,20 @@
+# Secrets
+
+The store is readable to all users on the system. For this reason, it
+is generally discouraged to allow secrets to make it into the store.
+
+Even on a single-user system, separate system users isolate services
+from each other and having secrets that all local users can read
+weakens that isolation. When using external store caches the secrets
+may end up there, and on multi-user systems the secrets will be
+available to all those users.
+
+Organize your derivations so that secrets are read from the filesystem
+(with appropriate access controls) at run time. Place the secrets on
+the filesystem manually or use a scheme that includes the secret in
+the store in encrypted form, and decrypts it adding the relevant
+access control on system activation.
+Several such schemes for NixOS can in the
+[comparison of secret managing schemes] on the wiki.
+
+[comparison of secret managing schemes]: https://wiki.nixos.org/wiki/Comparison_of_secret_managing_schemes

meson.build

@@ -60,4 +60,9 @@ if get_option('unit-tests')
   subproject('libflake-tests')
 endif
 subproject('nix-functional-tests')
-subproject('json-schema-checks')
+if get_option('json-schema-checks')
+  subproject('json-schema-checks')
+endif
+if get_option('kaitai-struct-checks')
+  subproject('kaitai-struct-checks')
+endif

meson.options

@@ -27,3 +27,17 @@ option(
   value : false,
   description : 'Build benchmarks (requires gbenchmark)',
 )
+
+option(
+  'kaitai-struct-checks',
+  type : 'boolean',
+  value : true,
+  description : 'Check the Kaitai Struct specifications (requires Kaitai Struct)',
+)
+
+option(
+  'json-schema-checks',
+  type : 'boolean',
+  value : true,
+  description : 'Check JSON schema validity of schemas and examples (requires jv)',
+)

packaging/dev-shell.nix

@@ -3,10 +3,118 @@
   devFlake,
 }:
 
+let
+  # Some helper functions
+
+  /**
+    Compute a filtered closure of build inputs.
+
+    Specifically, `buildInputsClosure cond startSet` computes the closure formed
+    by recursive application of `p: filter cond p.buildInputs ++ filter cond p.propagatedBuildInputs`
+    to `startSet`.
+
+    Example:
+    ```nix
+    builtInputsClosure isInternal [ pkg1 pkg2 ]
+    => [ pkg1 pkg3 pkg2 pkg10 ]
+    ```
+
+    Note: order tbd
+
+    Note: `startSet` is *NOT* filtered.
+  */
+  buildInputsClosureCond =
+    cond: startSet:
+    let
+      closure = builtins.genericClosure {
+        startSet = map (d: {
+          key = d.drvPath;
+          value = d;
+        }) startSet;
+        operator =
+          d:
+          let
+            r =
+              map
+                (d': {
+                  key = d'.drvPath;
+                  value = d';
+                })
+                (
+                  lib.filter cond d.value.buildInputs or [ ] ++ lib.filter cond d.value.propagatedBuildInputs or [ ]
+                );
+          in
+          r;
+      };
+    in
+    map (item: item.value) closure;
+
+  /**
+    `[ pkg1 pkg2 ]` -> `{ "...-pkg2.drv" = null; "...-pkg1.drv" = null }`
+
+    Note: fairly arbitrary order (hash based). Use for efficient set membership test only.
+  */
+  byDrvPath =
+    l:
+    lib.listToAttrs (
+      map (c: {
+        name =
+          # Just a lookup key
+          builtins.unsafeDiscardStringContext c.drvPath;
+        value = null;
+      }) l
+    );
+
+  /**
+    Stable dedup.
+
+    Unlike `listToAttrs` -> `attrValues`, this preserves the input ordering,
+    which is more predictable ("deterministic") than e.g. sorting store paths,
+    whose hashes affect the ordering on every change.
+  */
+  # TODO: add to Nixpkgs lib, refer from uniqueStrings
+  dedupByString =
+    key: l:
+    let
+      r =
+        lib.foldl'
+          (
+            a@{ list, set }:
+            elem:
+            let
+              k = builtins.unsafeDiscardStringContext (key elem);
+            in
+            if set ? ${k} then
+              a
+            else
+              let
+                # Note: O(n²) copying. Use linkedLists to concat them in one go at the end.
+                # https://github.com/NixOS/nixpkgs/pull/452088
+                newList = [ elem ] ++ list;
+                newSet = set // {
+                  ${k} = null;
+                };
+              in
+              builtins.seq newList builtins.seq newSet {
+                list = newList;
+                set = newSet;
+              }
+          )
+          {
+            list = [ ];
+            set = { };
+          }
+          l;
+    in
+    r.list;
+
+in
+
 { pkgs }:
 
+# TODO: don't use nix-util for this?
 pkgs.nixComponents2.nix-util.overrideAttrs (
-  attrs:
+  finalAttrs: prevAttrs:
 
   let
     stdenv = pkgs.nixDependencies2.stdenv;
@@ -21,13 +129,89 @@ pkgs.nixComponents2.nix-util.overrideAttrs (
       "-D${prefix}:${rest}";
     havePerl = stdenv.buildPlatform == stdenv.hostPlatform && stdenv.hostPlatform.isUnix;
     ignoreCrossFile = flags: builtins.filter (flag: !(lib.strings.hasInfix "cross-file" flag)) flags;
+
+    activeComponents = buildInputsClosureCond isInternal (
+      lib.attrValues (finalAttrs.passthru.config.getComponents allComponents)
+    );
+
+    allComponents = lib.filterAttrs (k: v: lib.isDerivation v) pkgs.nixComponents2;
+    internalDrvs = byDrvPath (
+      # Drop the attr names (not present in buildInputs anyway)
+      lib.attrValues allComponents
+      ++ lib.concatMap (c: lib.attrValues c.tests or { }) (lib.attrValues allComponents)
+    );
+
+    isInternal =
+      dep: internalDrvs ? ${builtins.unsafeDiscardStringContext dep.drvPath or "_non-existent_"};
+
   in
   {
-    pname = "shell-for-" + attrs.pname;
+    pname = "shell-for-nix";
+
+    passthru = {
+      inherit activeComponents;
+
+      # We use this attribute to store non-derivation values like functions and
+      # perhaps other things that are primarily for overriding and not the shell.
+      config = {
+        # Default getComponents
+        getComponents =
+          c:
+          builtins.removeAttrs c (
+            lib.optionals (!havePerl) [ "nix-perl-bindings" ]
+            ++ lib.optionals (!buildCanExecuteHost) [ "nix-manual" ]
+          );
+      };
+
+      /**
+        Produce a devShell for a given set of nix components
+
+        Example:
+
+        ```nix
+        shell.withActiveComponents (c: {
+          inherit (c) nix-util;
+        })
+        ```
+      */
+      withActiveComponents =
+        f2:
+        finalAttrs.finalPackage.overrideAttrs (
+          finalAttrs: prevAttrs: {
+            passthru = prevAttrs.passthru // {
+              config = prevAttrs.passthru.config // {
+                getComponents = f2;
+              };
+            };
+          }
+        );
+
+      small =
+        (finalAttrs.finalPackage.withActiveComponents (c: {
+          inherit (c)
+            nix-cli
+            nix-util-tests
+            nix-store-tests
+            nix-expr-tests
+            nix-fetchers-tests
+            nix-flake-tests
+            nix-functional-tests
+            # Currently required
+            nix-perl-bindings
+            ;
+        })).overrideAttrs
+          (o: {
+            mesonFlags = o.mesonFlags ++ [
+              # TODO: infer from activeComponents or vice versa
+              "-Dkaitai-struct-checks=false"
+              "-Djson-schema-checks=false"
+            ];
+          });
+    };
 
     # Remove the version suffix to avoid unnecessary attempts to substitute in nix develop
     version = lib.fileContents ../.version;
-    name = attrs.pname;
+    name = finalAttrs.pname;
 
     installFlags = "sysconfdir=$(out)/etc";
     shellHook = ''
@@ -98,17 +282,9 @@ pkgs.nixComponents2.nix-util.overrideAttrs (
     nativeBuildInputs =
       let
         inputs =
-          attrs.nativeBuildInputs or [ ]
-          ++ pkgs.nixComponents2.nix-util.nativeBuildInputs
-          ++ pkgs.nixComponents2.nix-store.nativeBuildInputs
-          ++ pkgs.nixComponents2.nix-fetchers.nativeBuildInputs
-          ++ pkgs.nixComponents2.nix-expr.nativeBuildInputs
-          ++ lib.optionals havePerl pkgs.nixComponents2.nix-perl-bindings.nativeBuildInputs
-          ++ lib.optionals buildCanExecuteHost pkgs.nixComponents2.nix-manual.externalNativeBuildInputs
-          ++ pkgs.nixComponents2.nix-internal-api-docs.nativeBuildInputs
-          ++ pkgs.nixComponents2.nix-external-api-docs.nativeBuildInputs
-          ++ pkgs.nixComponents2.nix-functional-tests.externalNativeBuildInputs
-          ++ pkgs.nixComponents2.nix-json-schema-checks.externalNativeBuildInputs
+          dedupByString (v: "${v}") (
+            lib.filter (x: !isInternal x) (lib.lists.concatMap (c: c.nativeBuildInputs) activeComponents)
+          )
           ++ lib.optional (
             !buildCanExecuteHost
             # Hack around https://github.com/nixos/nixpkgs/commit/bf7ad8cfbfa102a90463433e2c5027573b462479
@@ -117,9 +293,7 @@ pkgs.nixComponents2.nix-util.overrideAttrs (
             && lib.meta.availableOn stdenv.buildPlatform (stdenv.hostPlatform.emulator pkgs.buildPackages)
           ) pkgs.buildPackages.mesonEmulatorHook
           ++ [
-            pkgs.buildPackages.cmake
             pkgs.buildPackages.gnused
-            pkgs.buildPackages.changelog-d
             modular.pre-commit.settings.package
             (pkgs.writeScriptBin "pre-commit-hooks-install" modular.pre-commit.settings.installationScript)
             pkgs.buildPackages.nixfmt-rfc-style
@@ -136,18 +310,22 @@ pkgs.nixComponents2.nix-util.overrideAttrs (
       # from making its way into NIX_CFLAGS_COMPILE.
       lib.filter (p: !lib.hasInfix "separate-debug-info" p) inputs;
 
+    propagatedNativeBuildInputs = dedupByString (v: "${v}") (
+      lib.filter (x: !isInternal x) (
+        lib.lists.concatMap (c: c.propagatedNativeBuildInputs) activeComponents
+      )
+    );
+
     buildInputs = [
       pkgs.gbenchmark
     ]
-    ++ attrs.buildInputs or [ ]
-    ++ pkgs.nixComponents2.nix-util.buildInputs
-    ++ pkgs.nixComponents2.nix-store.buildInputs
-    ++ pkgs.nixComponents2.nix-store-tests.externalBuildInputs
-    ++ pkgs.nixComponents2.nix-fetchers.buildInputs
-    ++ pkgs.nixComponents2.nix-expr.buildInputs
-    ++ pkgs.nixComponents2.nix-expr.externalPropagatedBuildInputs
-    ++ pkgs.nixComponents2.nix-cmd.buildInputs
-    ++ lib.optionals havePerl pkgs.nixComponents2.nix-perl-bindings.externalBuildInputs
+    ++ dedupByString (v: "${v}") (
+      lib.filter (x: !isInternal x) (lib.lists.concatMap (c: c.buildInputs) activeComponents)
+    )
     ++ lib.optional havePerl pkgs.perl;
+
+    propagatedBuildInputs = dedupByString (v: "${v}") (
+      lib.filter (x: !isInternal x) (lib.lists.concatMap (c: c.propagatedBuildInputs) activeComponents)
+    );
   }
 )

src/json-schema-checks/package.nix

@@ -34,15 +34,11 @@ mkMesonDerivation (finalAttrs: {
 
   outputs = [ "out" ];
 
-  passthru.externalNativeBuildInputs = [
-    jsonschema
-  ];
-
   nativeBuildInputs = [
     meson
     ninja
-  ]
-  ++ finalAttrs.passthru.externalNativeBuildInputs;
+    jsonschema
+  ];
 
   doCheck = true;
 

src/kaitai-struct-checks/package.nix

@@ -37,7 +37,15 @@ mkMesonDerivation (finalAttrs: {
 
   outputs = [ "out" ];
 
-  passthru.externalNativeBuildInputs = [
+  buildInputs = [
+    gtest
+    kaitai-struct-cpp-stl-runtime
+  ];
+
+  nativeBuildInputs = [
+    meson
+    ninja
+    pkg-config
     # This can go away when we bump up to 25.11
     (kaitai-struct-compiler.overrideAttrs (finalAttrs: {
       version = "0.11";
@@ -48,20 +56,6 @@ mkMesonDerivation (finalAttrs: {
     }))
   ];
 
-  passthru.externalBuildInputs = [
-    gtest
-    kaitai-struct-cpp-stl-runtime
-  ];
-
-  buildInputs = finalAttrs.passthru.externalBuildInputs;
-
-  nativeBuildInputs = [
-    meson
-    ninja
-    pkg-config
-  ]
-  ++ finalAttrs.passthru.externalNativeBuildInputs;
-
   doCheck = true;
 
   mesonCheckFlags = [ "--print-errorlogs" ];

src/libcmd/command.cc

@@ -2,6 +2,7 @@
 #include <nlohmann/json.hpp>
 
 #include "nix/cmd/command.hh"
+#include "nix/cmd/legacy.hh"
 #include "nix/cmd/markdown.hh"
 #include "nix/store/store-open.hh"
 #include "nix/store/local-fs-store.hh"
@@ -14,6 +15,18 @@
 
 namespace nix {
 
+RegisterCommand::Commands & RegisterCommand::commands()
+{
+    static RegisterCommand::Commands commands;
+    return commands;
+}
+
+RegisterLegacyCommand::Commands & RegisterLegacyCommand::commands()
+{
+    static RegisterLegacyCommand::Commands commands;
+    return commands;
+}
+
 nix::Commands RegisterCommand::getCommandsFor(const std::vector<std::string> & prefix)
 {
     nix::Commands res;

src/libcmd/common-eval-args.cc

@@ -33,7 +33,8 @@ EvalSettings evalSettings{
                 // FIXME `parseFlakeRef` should take a `std::string_view`.
                 auto flakeRef = parseFlakeRef(fetchSettings, std::string{rest}, {}, true, false);
                 debug("fetching flake search path element '%s''", rest);
-                auto [accessor, lockedRef] = flakeRef.resolve(state.store).lazyFetch(state.store);
+                auto [accessor, lockedRef] =
+                    flakeRef.resolve(fetchSettings, state.store).lazyFetch(fetchSettings, state.store);
                 auto storePath = nix::fetchToStore(
                     state.fetchSettings,
                     *state.store,
@@ -131,7 +132,7 @@ MixEvalArgs::MixEvalArgs()
             fetchers::Attrs extraAttrs;
             if (to.subdir != "")
                 extraAttrs["dir"] = to.subdir;
-            fetchers::overrideRegistry(from.input, to.input, extraAttrs);
+            fetchers::overrideRegistry(fetchSettings, from.input, to.input, extraAttrs);
         }},
         .completer = {[&](AddCompletions & completions, size_t, std::string_view prefix) {
             completeFlakeRef(completions, openStore(), prefix);
@@ -187,7 +188,7 @@ SourcePath lookupFileArg(EvalState & state, std::string_view s, const Path * bas
     else if (hasPrefix(s, "flake:")) {
         experimentalFeatureSettings.require(Xp::Flakes);
         auto flakeRef = parseFlakeRef(fetchSettings, std::string(s.substr(6)), {}, true, false);
-        auto [accessor, lockedRef] = flakeRef.resolve(state.store).lazyFetch(state.store);
+        auto [accessor, lockedRef] = flakeRef.resolve(fetchSettings, state.store).lazyFetch(fetchSettings, state.store);
         auto storePath = nix::fetchToStore(
             state.fetchSettings, *state.store, SourcePath(accessor), FetchMode::Copy, lockedRef.input.getName());
         state.allowPath(storePath);

src/libcmd/include/nix/cmd/command.hh

@@ -286,11 +286,7 @@ struct RegisterCommand
 {
     typedef std::map<std::vector<std::string>, std::function<ref<Command>()>> Commands;
 
-    static Commands & commands()
-    {
-        static Commands commands;
-        return commands;
-    }
+    static Commands & commands();
 
     RegisterCommand(std::vector<std::string> && name, std::function<ref<Command>()> command)
     {

src/libcmd/include/nix/cmd/legacy.hh

@@ -13,11 +13,7 @@ struct RegisterLegacyCommand
 {
     typedef std::map<std::string, MainFunction> Commands;
 
-    static Commands & commands()
-    {
-        static Commands commands;
-        return commands;
-    }
+    static Commands & commands();
 
     RegisterLegacyCommand(const std::string & name, MainFunction fun)
     {

src/libcmd/installables.cc

@@ -185,6 +185,7 @@ MixFlakeOptions::MixFlakeOptions()
                     }
 
                     overrideRegistry(
+                        fetchSettings,
                         fetchers::Input::fromAttrs(fetchSettings, {{"type", "indirect"}, {"id", inputName}}),
                         input3->lockedRef.input,
                         extraAttrs);

src/libcmd/repl.cc

@@ -738,7 +738,7 @@ void NixRepl::loadFlake(const std::string & flakeRefS)
     }
 
     auto flakeRef = parseFlakeRef(fetchSettings, flakeRefS, cwd.string(), true);
-    if (evalSettings.pureEval && !flakeRef.input.isLocked())
+    if (evalSettings.pureEval && !flakeRef.input.isLocked(fetchSettings))
         throw Error("cannot use ':load-flake' on locked flake reference '%s' (use --impure to override)", flakeRefS);
 
     Value v;

src/libexpr-c/nix_api_expr.cc

@@ -137,6 +137,8 @@ nix_eval_state_builder * nix_eval_state_builder_new(nix_c_context * context, Sto
 
 void nix_eval_state_builder_free(nix_eval_state_builder * builder)
 {
+    if (builder)
+        builder->~nix_eval_state_builder();
     operator delete(builder, static_cast<std::align_val_t>(alignof(nix_eval_state_builder)));
 }
 
@@ -203,6 +205,8 @@ EvalState * nix_state_create(nix_c_context * context, const char ** lookupPath_c
 
 void nix_state_free(EvalState * state)
 {
+    if (state)
+        state->~EvalState();
     operator delete(state, static_cast<std::align_val_t>(alignof(EvalState)));
 }
 

src/libexpr-test-support/include/nix/expr/tests/libexpr.hh

@@ -104,6 +104,7 @@ MATCHER(IsAttrs, "")
 MATCHER_P(IsStringEq, s, fmt("The string is equal to \"%1%\"", s))
 {
     if (arg.type() != nString) {
+        *result_listener << "Expected a string got " << arg.type();
         return false;
     }
     return arg.string_view() == s;

src/libexpr-tests/json.cc

@@ -1,5 +1,6 @@
 #include "nix/expr/tests/libexpr.hh"
 #include "nix/expr/value-to-json.hh"
+#include "nix/expr/static-string-data.hh"
 
 namespace nix {
 // Testing the conversion to JSON
@@ -54,7 +55,7 @@ TEST_F(JSONValueTest, IntNegative)
 TEST_F(JSONValueTest, String)
 {
     Value v;
-    v.mkStringNoCopy("test");
+    v.mkStringNoCopy("test"_sds);
     ASSERT_EQ(getJSONValue(v), "\"test\"");
 }
 
@@ -62,7 +63,7 @@ TEST_F(JSONValueTest, StringQuotes)
 {
     Value v;
 
-    v.mkStringNoCopy("test\"");
+    v.mkStringNoCopy("test\""_sds);
     ASSERT_EQ(getJSONValue(v), "\"test\\\"\"");
 }
 

src/libexpr-tests/nix_api_expr.cc

@@ -14,7 +14,7 @@
 
 namespace nixC {
 
-TEST_F(nix_api_store_test, nix_eval_state_lookup_path)
+TEST_F(nix_api_expr_test, nix_eval_state_lookup_path)
 {
     auto tmpDir = nix::createTempDir();
     auto delTmpDir = std::make_unique<nix::AutoDelete>(tmpDir, true);
@@ -42,12 +42,16 @@ TEST_F(nix_api_store_test, nix_eval_state_lookup_path)
     nix_expr_eval_from_string(ctx, state, "builtins.seq <nixos-config> <nixpkgs>", ".", value);
     assert_ctx_ok();
 
+    nix_state_free(state);
+
     ASSERT_EQ(nix_get_type(ctx, value), NIX_TYPE_PATH);
     assert_ctx_ok();
 
     auto pathStr = nix_get_path_string(ctx, value);
     assert_ctx_ok();
     ASSERT_EQ(0, strcmp(pathStr, nixpkgs.c_str()));
+
+    nix_gc_decref(nullptr, value);
 }
 
 TEST_F(nix_api_expr_test, nix_expr_eval_from_string)

src/libexpr-tests/primops.cc

@@ -661,8 +661,14 @@ INSTANTIATE_TEST_SUITE_P(
         CASE(R"(null)", ""),
         CASE(R"({ v = "bar"; __toString = self: self.v; })", "bar"),
         CASE(R"({ v = "bar"; __toString = self: self.v; outPath = "foo"; })", "bar"),
-        CASE(R"({ outPath = "foo"; })", "foo"),
-        CASE(R"(./test)", "/test")));
+        CASE(R"({ outPath = "foo"; })", "foo")
+// this is broken on cygwin because canonPath("//./test", false) returns //./test
+// FIXME: don't use canonPath
+#ifndef __CYGWIN__
+            ,
+        CASE(R"(./test)", "/test")
+#endif
+            ));
 #undef CASE
 
 TEST_F(PrimOpTest, substring)

src/libexpr-tests/value/print.cc

@@ -1,4 +1,5 @@
 #include "nix/expr/tests/libexpr.hh"
+#include "nix/expr/static-string-data.hh"
 
 #include "nix/expr/value.hh"
 #include "nix/expr/print.hh"
@@ -35,14 +36,14 @@ TEST_F(ValuePrintingTests, tBool)
 TEST_F(ValuePrintingTests, tString)
 {
     Value vString;
-    vString.mkStringNoCopy("some-string");
+    vString.mkStringNoCopy("some-string"_sds);
     test(vString, "\"some-string\"");
 }
 
 TEST_F(ValuePrintingTests, tPath)
 {
     Value vPath;
-    vPath.mkStringNoCopy("/foo");
+    vPath.mkStringNoCopy("/foo"_sds);
     test(vPath, "\"/foo\"");
 }
 
@@ -289,10 +290,10 @@ TEST_F(StringPrintingTests, maxLengthTruncation)
 TEST_F(ValuePrintingTests, attrsTypeFirst)
 {
     Value vType;
-    vType.mkStringNoCopy("puppy");
+    vType.mkStringNoCopy("puppy"_sds);
 
     Value vApple;
-    vApple.mkStringNoCopy("apple");
+    vApple.mkStringNoCopy("apple"_sds);
 
     BindingsBuilder builder = state.buildBindings(10);
     builder.insert(state.symbols.create("type"), &vType);
@@ -333,7 +334,7 @@ TEST_F(ValuePrintingTests, ansiColorsBool)
 TEST_F(ValuePrintingTests, ansiColorsString)
 {
     Value v;
-    v.mkStringNoCopy("puppy");
+    v.mkStringNoCopy("puppy"_sds);
 
     test(v, ANSI_MAGENTA "\"puppy\"" ANSI_NORMAL, PrintOptions{.ansiColors = true});
 }
@@ -341,7 +342,7 @@ TEST_F(ValuePrintingTests, ansiColorsString)
 TEST_F(ValuePrintingTests, ansiColorsStringElided)
 {
     Value v;
-    v.mkStringNoCopy("puppy");
+    v.mkStringNoCopy("puppy"_sds);
 
     test(
         v,
@@ -389,7 +390,7 @@ TEST_F(ValuePrintingTests, ansiColorsAttrs)
 TEST_F(ValuePrintingTests, ansiColorsDerivation)
 {
     Value vDerivation;
-    vDerivation.mkStringNoCopy("derivation");
+    vDerivation.mkStringNoCopy("derivation"_sds);
 
     BindingsBuilder builder = state.buildBindings(10);
     builder.insert(state.s.type, &vDerivation);
@@ -412,7 +413,7 @@ TEST_F(ValuePrintingTests, ansiColorsError)
 {
     Value throw_ = state.getBuiltin("throw");
     Value message;
-    message.mkStringNoCopy("uh oh!");
+    message.mkStringNoCopy("uh oh!"_sds);
     Value vError;
     vError.mkApp(&throw_, &message);
 
@@ -429,12 +430,12 @@ TEST_F(ValuePrintingTests, ansiColorsDerivationError)
 {
     Value throw_ = state.getBuiltin("throw");
     Value message;
-    message.mkStringNoCopy("uh oh!");
+    message.mkStringNoCopy("uh oh!"_sds);
     Value vError;
     vError.mkApp(&throw_, &message);
 
     Value vDerivation;
-    vDerivation.mkStringNoCopy("derivation");
+    vDerivation.mkStringNoCopy("derivation"_sds);
 
     BindingsBuilder builder = state.buildBindings(10);
     builder.insert(state.s.type, &vDerivation);

src/libexpr-tests/value/value.cc

@@ -1,4 +1,5 @@
 #include "nix/expr/value.hh"
+#include "nix/expr/static-string-data.hh"
 
 #include "nix/store/tests/libstore.hh"
 #include <gtest/gtest.h>
@@ -27,17 +28,17 @@ TEST_F(ValueTest, staticString)
 {
     Value vStr1;
     Value vStr2;
-    vStr1.mkStringNoCopy("foo");
-    vStr2.mkStringNoCopy("foo");
+    vStr1.mkStringNoCopy("foo"_sds);
+    vStr2.mkStringNoCopy("foo"_sds);
 
-    auto sd1 = vStr1.string_view();
-    auto sd2 = vStr2.string_view();
+    auto & sd1 = vStr1.string_data();
+    auto & sd2 = vStr2.string_data();
 
     // The strings should be the same
-    ASSERT_EQ(sd1, sd2);
+    ASSERT_EQ(sd1.view(), sd2.view());
 
     // The strings should also be backed by the same (static) allocation
-    ASSERT_EQ(sd1.data(), sd2.data());
+    ASSERT_EQ(&sd1, &sd2);
 }
 
 } // namespace nix

src/libexpr/eval-cache.cc

@@ -147,7 +147,7 @@ struct AttrDb
                 for (auto * elem : *context) {
                     if (!first)
                         ctx.push_back(' ');
-                    ctx.append(elem);
+                    ctx.append(elem->view());
                     first = false;
                 }
                 state->insertAttributeWithContext.use()(key.first)(symbols[key.second])(AttrType::String) (s) (ctx)

src/libexpr/eval.cc

@@ -3,6 +3,7 @@
 #include "nix/expr/primops.hh"
 #include "nix/expr/print-options.hh"
 #include "nix/expr/symbol-table.hh"
+#include "nix/expr/value.hh"
 #include "nix/util/exit.hh"
 #include "nix/util/types.hh"
 #include "nix/util/util.hh"
@@ -28,6 +29,8 @@
 #include "parser-tab.hh"
 
 #include <algorithm>
+#include <cstddef>
+#include <cstdlib>
 #include <iostream>
 #include <sstream>
 #include <cstring>
@@ -48,6 +51,9 @@ using json = nlohmann::json;
 
 namespace nix {
 
+/**
+ * Just for doc strings. Not for regular string values.
+ */
 static char * allocString(size_t size)
 {
     char * t;
@@ -61,6 +67,9 @@ static char * allocString(size_t size)
 // string allocations.
 // This function handles makeImmutableString(std::string_view()) by returning
 // the empty string.
+/**
+ * Just for doc strings. Not for regular string values.
+ */
 static const char * makeImmutableString(std::string_view s)
 {
     const size_t size = s.size();
@@ -72,6 +81,25 @@ static const char * makeImmutableString(std::string_view s)
     return t;
 }
 
+StringData & StringData::alloc(size_t size)
+{
+    void * t = GC_MALLOC_ATOMIC(sizeof(StringData) + size + 1);
+    if (!t)
+        throw std::bad_alloc();
+    auto res = new (t) StringData(size);
+    return *res;
+}
+
+const StringData & StringData::make(std::string_view s)
+{
+    if (s.empty())
+        return ""_sds;
+    auto & res = alloc(s.size());
+    std::memcpy(&res.data_, s.data(), s.size());
+    res.data_[s.size()] = '\0';
+    return res;
+}
+
 RootValue allocRootValue(Value * v)
 {
     return std::allocate_shared<Value *>(traceable_allocator<Value *>(), v);
@@ -489,15 +517,16 @@ Value * EvalState::addPrimOp(PrimOp && primOp)
     if (primOp.arity == 0) {
         primOp.arity = 1;
         auto vPrimOp = allocValue();
-        vPrimOp->mkPrimOp(new PrimOp(primOp));
+        vPrimOp->mkPrimOp(new PrimOp(std::move(primOp)));
         Value v;
         v.mkApp(vPrimOp, vPrimOp);
+        auto & primOp1 = *vPrimOp->primOp();
         return addConstant(
-            primOp.name,
+            primOp1.name,
             v,
             {
                 .type = nThunk, // FIXME
-                .doc = primOp.doc,
+                .doc = primOp1.doc ? primOp1.doc->c_str() : nullptr,
             });
     }
 
@@ -537,13 +566,14 @@ std::optional<EvalState::Doc> EvalState::getDoc(Value & v)
 {
     if (v.isPrimOp()) {
         auto v2 = &v;
-        if (auto * doc = v2->primOp()->doc)
+        auto & primOp = *v2->primOp();
+        if (primOp.doc)
             return Doc{
                 .pos = {},
-                .name = v2->primOp()->name,
-                .arity = v2->primOp()->arity,
-                .args = v2->primOp()->args,
-                .doc = doc,
+                .name = primOp.name,
+                .arity = primOp.arity,
+                .args = primOp.args,
+                .doc = primOp.doc->c_str(),
             };
     }
     if (v.isLambda()) {
@@ -585,7 +615,9 @@ std::optional<EvalState::Doc> EvalState::getDoc(Value & v)
             .name = name,
             .arity = 0, // FIXME: figure out how deep by syntax only? It's not semantically useful though...
             .args = {},
-            .doc = makeImmutableString(s.view()), // NOTE: memory leak when compiled without GC
+            /* N.B. Can't use StringData here, because that would lead to an interior pointer.
+               NOTE: memory leak when compiled without GC. */
+            .doc = makeImmutableString(s.view()),
         };
     }
     if (isFunctor(v)) {
@@ -819,33 +851,34 @@ DebugTraceStacker::DebugTraceStacker(EvalState & evalState, DebugTrace t)
 
 void Value::mkString(std::string_view s)
 {
-    mkStringNoCopy(makeImmutableString(s));
+    mkStringNoCopy(StringData::make(s));
 }
 
-Value::StringWithContext::Context * Value::StringWithContext::Context::fromBuilder(const NixStringContext & context)
+Value::StringWithContext::Context *
+Value::StringWithContext::Context::fromBuilder(const NixStringContext & context, EvalMemory & mem)
 {
     if (context.empty())
         return nullptr;
 
-    auto ctx = new (allocBytes(sizeof(Context) + context.size() * sizeof(value_type))) Context(context.size());
+    auto ctx = new (mem.allocBytes(sizeof(Context) + context.size() * sizeof(value_type))) Context(context.size());
     std::ranges::transform(
-        context, ctx->elems, [](const NixStringContextElem & elt) { return makeImmutableString(elt.to_string()); });
+        context, ctx->elems, [](const NixStringContextElem & elt) { return &StringData::make(elt.to_string()); });
     return ctx;
 }
 
-void Value::mkString(std::string_view s, const NixStringContext & context)
+void Value::mkString(std::string_view s, const NixStringContext & context, EvalMemory & mem)
 {
-    mkStringNoCopy(makeImmutableString(s), Value::StringWithContext::Context::fromBuilder(context));
+    mkStringNoCopy(StringData::make(s), Value::StringWithContext::Context::fromBuilder(context, mem));
 }
 
-void Value::mkStringMove(const char * s, const NixStringContext & context)
+void Value::mkStringMove(const StringData & s, const NixStringContext & context, EvalMemory & mem)
 {
-    mkStringNoCopy(s, Value::StringWithContext::Context::fromBuilder(context));
+    mkStringNoCopy(s, Value::StringWithContext::Context::fromBuilder(context, mem));
 }
 
 void Value::mkPath(const SourcePath & path)
 {
-    mkPath(&*path.accessor, makeImmutableString(path.path.abs()));
+    mkPath(&*path.accessor, StringData::make(path.path.abs()));
 }
 
 inline Value * EvalState::lookupVar(Env * env, const ExprVar & var, bool noEval)
@@ -881,9 +914,9 @@ inline Value * EvalState::lookupVar(Env * env, const ExprVar & var, bool noEval)
     }
 }
 
-ListBuilder::ListBuilder(size_t size)
+ListBuilder::ListBuilder(EvalMemory & mem, size_t size)
     : size(size)
-    , elems(size <= 2 ? inlineElems : (Value **) allocBytes(size * sizeof(Value *)))
+    , elems(size <= 2 ? inlineElems : (Value **) mem.allocBytes(size * sizeof(Value *)))
 {
 }
 
@@ -923,7 +956,8 @@ void EvalState::mkStorePathString(const StorePath & p, Value & v)
         store->printStorePath(p),
         NixStringContext{
             NixStringContextElem::Opaque{.path = p},
-        });
+        },
+        mem);
 }
 
 std::string EvalState::mkOutputStringRaw(
@@ -945,7 +979,7 @@ void EvalState::mkOutputString(
     std::optional<StorePath> optStaticOutputPath,
     const ExperimentalFeatureSettings & xpSettings)
 {
-    value.mkString(mkOutputStringRaw(b, optStaticOutputPath, xpSettings), NixStringContext{b});
+    value.mkString(mkOutputStringRaw(b, optStaticOutputPath, xpSettings), NixStringContext{b}, mem);
 }
 
 std::string EvalState::mkSingleDerivedPathStringRaw(const SingleDerivedPath & p)
@@ -980,7 +1014,8 @@ void EvalState::mkSingleDerivedPathString(const SingleDerivedPath & p, Value & v
         mkSingleDerivedPathStringRaw(p),
         NixStringContext{
             std::visit([](auto && v) -> NixStringContextElem { return v; }, p),
-        });
+        },
+        mem);
 }
 
 Value * Expr::maybeThunk(EvalState & state, Env & env)
@@ -2099,21 +2134,21 @@ void ExprConcatStrings::eval(EvalState & state, Env & env, Value & v)
                 .atPos(pos)
                 .withFrame(env, *this)
                 .debugThrow();
-        std::string result_str;
-        result_str.reserve(sSize);
+        std::string resultStr;
+        resultStr.reserve(sSize);
         for (const auto & part : strings) {
-            result_str += *part;
+            resultStr += *part;
         }
-        v.mkPath(state.rootPath(CanonPath(result_str)));
+        v.mkPath(state.rootPath(CanonPath(resultStr)));
     } else {
-        char * result_str = allocString(sSize + 1);
-        char * tmp = result_str;
+        auto & resultStr = StringData::alloc(sSize);
+        auto * tmp = resultStr.data();
         for (const auto & part : strings) {
-            memcpy(tmp, part->data(), part->size());
+            std::memcpy(tmp, part->data(), part->size());
             tmp += part->size();
         }
-        *tmp = 0;
-        v.mkStringMove(result_str, context);
+        *tmp = '\0';
+        v.mkStringMove(resultStr, context, state.mem);
     }
 }
 
@@ -2288,7 +2323,7 @@ void copyContext(const Value & v, NixStringContext & context, const Experimental
 {
     if (auto * ctx = v.context())
         for (auto * elem : *ctx)
-            context.insert(NixStringContextElem::parse(elem, xpSettings));
+            context.insert(NixStringContextElem::parse(elem->view(), xpSettings));
 }
 
 std::string_view EvalState::forceString(
@@ -2310,7 +2345,7 @@ std::string_view EvalState::forceStringNoCtx(Value & v, const PosIdx pos, std::s
         error<EvalError>(
             "the string '%1%' is not allowed to refer to a store path (such as '%2%')",
             v.string_view(),
-            *v.context()->begin())
+            (*v.context()->begin())->view())
             .withTrace(pos, errorCtx)
             .debugThrow();
     }

src/libexpr/include/nix/expr/eval-inline.hh

@@ -12,7 +12,7 @@ namespace nix {
  * Note: Various places expect the allocated memory to be zeroed.
  */
 [[gnu::always_inline]]
-inline void * allocBytes(size_t n)
+inline void * EvalMemory::allocBytes(size_t n)
 {
     void * p;
 #if NIX_USE_BOEHMGC

src/libexpr/include/nix/expr/eval.hh

@@ -108,7 +108,7 @@ struct PrimOp
     /**
      * Optional free-form documentation about the primop.
      */
-    const char * doc = nullptr;
+    std::optional<std::string> doc;
 
     /**
      * Add a trace item, while calling the `<name>` builtin.
@@ -335,6 +335,7 @@ public:
     EvalMemory & operator=(const EvalMemory &) = delete;
     EvalMemory & operator=(EvalMemory &&) = delete;
 
+    inline void * allocBytes(size_t n);
     inline Value * allocValue();
     inline Env & allocEnv(size_t size);
 
@@ -348,7 +349,7 @@ public:
     ListBuilder buildList(size_t size)
     {
         stats.nrListElems += size;
-        return ListBuilder(size);
+        return ListBuilder(*this, size);
     }
 
     const Statistics & getStats() const &

src/libexpr/include/nix/expr/meson.build

@@ -31,6 +31,7 @@ headers = [ config_pub_h ] + files(
   'print.hh',
   'repl-exit-status.hh',
   'search-path.hh',
+  'static-string-data.hh',
   'symbol-table.hh',
   'value-to-json.hh',
   'value-to-xml.hh',

src/libexpr/include/nix/expr/nixexpr.hh

@@ -3,6 +3,7 @@
 
 #include <map>
 #include <span>
+#include <memory>
 #include <vector>
 #include <memory_resource>
 #include <algorithm>
@@ -11,6 +12,7 @@
 #include "nix/expr/value.hh"
 #include "nix/expr/symbol-table.hh"
 #include "nix/expr/eval-error.hh"
+#include "nix/expr/static-string-data.hh"
 #include "nix/util/pos-idx.hh"
 #include "nix/expr/counter.hh"
 #include "nix/util/pos-table.hh"
@@ -186,22 +188,18 @@ struct ExprString : Expr
      * This is only for strings already allocated in our polymorphic allocator,
      * or that live at least that long (e.g. c++ string literals)
      */
-    ExprString(const char * s)
+    ExprString(const StringData & s)
     {
         v.mkStringNoCopy(s);
     };
 
     ExprString(std::pmr::polymorphic_allocator<char> & alloc, std::string_view sv)
     {
-        auto len = sv.length();
-        if (len == 0) {
-            v.mkStringNoCopy("");
+        if (sv.size() == 0) {
+            v.mkStringNoCopy(""_sds);
             return;
         }
-        char * s = alloc.allocate(len + 1);
-        sv.copy(s, len);
-        s[len] = '\0';
-        v.mkStringNoCopy(s);
+        v.mkStringNoCopy(StringData::make(*alloc.resource(), sv));
     };
 
     Value * maybeThunk(EvalState & state, Env & env) override;
@@ -216,11 +214,7 @@ struct ExprPath : Expr
     ExprPath(std::pmr::polymorphic_allocator<char> & alloc, ref<SourceAccessor> accessor, std::string_view sv)
         : accessor(accessor)
     {
-        auto len = sv.length();
-        char * s = alloc.allocate(len + 1);
-        sv.copy(s, len);
-        s[len] = '\0';
-        v.mkPath(&*accessor, s);
+        v.mkPath(&*accessor, StringData::make(*alloc.resource(), sv));
     }
 
     Value * maybeThunk(EvalState & state, Env & env) override;
@@ -345,7 +339,7 @@ struct ExprOpHasAttr : Expr
     Expr * e;
     std::span<AttrName> attrPath;
 
-    ExprOpHasAttr(std::pmr::polymorphic_allocator<char> & alloc, Expr * e, std::vector<AttrName> attrPath)
+    ExprOpHasAttr(std::pmr::polymorphic_allocator<char> & alloc, Expr * e, std::span<AttrName> attrPath)
         : e(e)
         , attrPath({alloc.allocate_object<AttrName>(attrPath.size()), attrPath.size()})
     {
@@ -439,7 +433,7 @@ struct ExprList : Expr
 {
     std::span<Expr *> elems;
 
-    ExprList(std::pmr::polymorphic_allocator<char> & alloc, std::vector<Expr *> exprs)
+    ExprList(std::pmr::polymorphic_allocator<char> & alloc, std::span<Expr *> exprs)
         : elems({alloc.allocate_object<Expr *>(exprs.size()), exprs.size()})
     {
         std::ranges::copy(exprs, elems.begin());
@@ -568,7 +562,7 @@ public:
         const PosTable & positions,
         std::pmr::polymorphic_allocator<char> & alloc,
         PosIdx pos,
-        FormalsBuilder formals,
+        const FormalsBuilder & formals,
         Expr * body)
         : ExprLambda(positions, alloc, pos, Symbol(), formals, body) {};
 
@@ -759,7 +753,19 @@ struct ExprConcatStrings : Expr
         std::pmr::polymorphic_allocator<char> & alloc,
         const PosIdx & pos,
         bool forceString,
-        const std::vector<std::pair<PosIdx, Expr *>> & es)
+        std::span<std::pair<PosIdx, Expr *>> es)
+        : pos(pos)
+        , forceString(forceString)
+        , es({alloc.allocate_object<std::pair<PosIdx, Expr *>>(es.size()), es.size()})
+    {
+        std::ranges::copy(es, this->es.begin());
+    };
+
+    ExprConcatStrings(
+        std::pmr::polymorphic_allocator<char> & alloc,
+        const PosIdx & pos,
+        bool forceString,
+        std::initializer_list<std::pair<PosIdx, Expr *>> es)
         : pos(pos)
         , forceString(forceString)
         , es({alloc.allocate_object<std::pair<PosIdx, Expr *>>(es.size()), es.size()})
@@ -839,7 +845,19 @@ public:
     add(std::pmr::polymorphic_allocator<char> & alloc,
         const PosIdx & pos,
         bool forceString,
-        const std::vector<std::pair<PosIdx, Expr *>> & es)
+        std::span<std::pair<PosIdx, Expr *>> es)
+        requires(std::same_as<C, ExprConcatStrings>)
+    {
+        return alloc.new_object<C>(alloc, pos, forceString, es);
+    }
+
+    template<class C>
+    [[gnu::always_inline]]
+    C *
+    add(std::pmr::polymorphic_allocator<char> & alloc,
+        const PosIdx & pos,
+        bool forceString,
+        std::initializer_list<std::pair<PosIdx, Expr *>> es)
         requires(std::same_as<C, ExprConcatStrings>)
     {
         return alloc.new_object<C>(alloc, pos, forceString, es);

src/libexpr/include/nix/expr/parser-state.hh

@@ -4,6 +4,8 @@
 #include <limits>
 
 #include "nix/expr/eval.hh"
+#include "nix/expr/value.hh"
+#include "nix/expr/static-string-data.hh"
 
 namespace nix {
 
@@ -94,7 +96,7 @@ struct ParserState
         ExprAttrs * attrs, AttrPath && attrPath, const ParserLocation & loc, Expr * e, const ParserLocation & exprLoc);
     void addAttr(ExprAttrs * attrs, AttrPath & attrPath, const Symbol & symbol, ExprAttrs::AttrDef && def);
     void validateFormals(FormalsBuilder & formals, PosIdx pos = noPos, Symbol arg = {});
-    Expr * stripIndentation(const PosIdx pos, std::vector<std::pair<PosIdx, std::variant<Expr *, StringToken>>> && es);
+    Expr * stripIndentation(const PosIdx pos, std::span<std::pair<PosIdx, std::variant<Expr *, StringToken>>> es);
     PosIdx at(const ParserLocation & loc);
 };
 
@@ -237,10 +239,10 @@ inline void ParserState::validateFormals(FormalsBuilder & formals, PosIdx pos, S
 }
 
 inline Expr *
-ParserState::stripIndentation(const PosIdx pos, std::vector<std::pair<PosIdx, std::variant<Expr *, StringToken>>> && es)
+ParserState::stripIndentation(const PosIdx pos, std::span<std::pair<PosIdx, std::variant<Expr *, StringToken>>> es)
 {
     if (es.empty())
-        return exprs.add<ExprString>("");
+        return exprs.add<ExprString>(""_sds);
 
     /* Figure out the minimum indentation.  Note that by design
        whitespace-only final lines are not taken into account.  (So
@@ -332,7 +334,7 @@ ParserState::stripIndentation(const PosIdx pos, std::vector<std::pair<PosIdx, st
     // If there is nothing at all, return the empty string directly.
     // This also ensures that equivalent empty strings result in the same ast, which is helpful when testing formatters.
     if (es2.size() == 0) {
-        auto * const result = exprs.add<ExprString>("");
+        auto * const result = exprs.add<ExprString>(""_sds);
         return result;
     }
 
@@ -341,7 +343,7 @@ ParserState::stripIndentation(const PosIdx pos, std::vector<std::pair<PosIdx, st
         auto * const result = (es2)[0].second;
         return result;
     }
-    return exprs.add<ExprConcatStrings>(exprs.alloc, pos, true, std::move(es2));
+    return exprs.add<ExprConcatStrings>(exprs.alloc, pos, true, es2);
 }
 
 inline PosIdx LexerState::at(const ParserLocation & loc)

src/libexpr/include/nix/expr/primops.hh

@@ -12,11 +12,7 @@ struct RegisterPrimOp
 {
     typedef std::vector<PrimOp> PrimOps;
 
-    static PrimOps & primOps()
-    {
-        static PrimOps primOps;
-        return primOps;
-    }
+    static PrimOps & primOps();
 
     /**
      * You can register a constant by passing an arity of 0. fun

src/libexpr/include/nix/expr/print-options.hh

@@ -110,7 +110,7 @@ struct PrintOptions
  * `PrintOptions` for unknown and therefore potentially large values in error messages,
  * to avoid printing "too much" output.
  */
-static PrintOptions errorPrintOptions = PrintOptions{
+static constexpr PrintOptions errorPrintOptions = PrintOptions{
     .ansiColors = true,
     .maxDepth = 10,
     .maxAttrs = 10,

src/libexpr/include/nix/expr/static-string-data.hh

@@ -0,0 +1,44 @@
+#pragma once
+///@file
+
+#include "nix/expr/value.hh"
+
+namespace nix {
+
+template<size_t N>
+struct StringData::Static
+{
+    /**
+     * @note Must be first to make layout compatible with StringData.
+     */
+    const size_t size = N - 1;
+    char data[N];
+
+    consteval Static(const char (&str)[N])
+    {
+        static_assert(N > 0);
+        if (str[size] != '\0')
+            throw;
+        std::copy_n(str, N, data);
+    }
+
+    operator const StringData &() const &
+    {
+        static_assert(sizeof(decltype(*this)) >= sizeof(StringData));
+        static_assert(alignof(decltype(*this)) == alignof(StringData));
+        /* NOTE: This cast is somewhat on the fence of what's legal in C++.
+           The question boils down to whether flexible array members are
+           layout compatible with fixed-size arrays. This is a gray area, since
+           FAMs are not standard anyway.
+           */
+        return *reinterpret_cast<const StringData *>(this);
+    }
+};
+
+template<StringData::Static S>
+const StringData & operator""_sds()
+{
+    return S;
+}
+
+} // namespace nix

src/libexpr/include/nix/expr/symbol-table.hh

@@ -3,6 +3,7 @@
 
 #include <memory_resource>
 #include "nix/expr/value.hh"
+#include "nix/expr/static-string-data.hh"
 #include "nix/util/chunked-vector.hh"
 #include "nix/util/error.hh"
 
@@ -16,7 +17,6 @@ class SymbolValue : protected Value
     friend class SymbolStr;
     friend class SymbolTable;
 
-    uint32_t size_;
     uint32_t idx;
 
     SymbolValue() = default;
@@ -24,7 +24,7 @@ class SymbolValue : protected Value
 public:
     operator std::string_view() const noexcept
     {
-        return {c_str(), size_};
+        return string_view();
     }
 };
 
@@ -96,13 +96,13 @@ class SymbolStr
         SymbolValueStore & store;
         std::string_view s;
         std::size_t hash;
-        std::pmr::polymorphic_allocator<char> & alloc;
+        std::pmr::memory_resource & resource;
 
-        Key(SymbolValueStore & store, std::string_view s, std::pmr::polymorphic_allocator<char> & stringAlloc)
+        Key(SymbolValueStore & store, std::string_view s, std::pmr::memory_resource & stringMemory)
             : store(store)
             , s(s)
             , hash(HashType{}(s))
-            , alloc(stringAlloc)
+            , resource(stringMemory)
         {
         }
     };
@@ -122,14 +122,10 @@ public:
         // for multi-threaded implementations: lock store and allocator here
         const auto & [v, idx] = key.store.add(SymbolValue{});
         if (size == 0) {
-            v.mkStringNoCopy("", nullptr);
+            v.mkStringNoCopy(""_sds, nullptr);
         } else {
-            auto s = key.alloc.allocate(size + 1);
-            memcpy(s, key.s.data(), size);
-            s[size] = '\0';
-            v.mkStringNoCopy(s, nullptr);
+            v.mkStringNoCopy(StringData::make(key.resource, key.s));
         }
-        v.size_ = size;
         v.idx = idx;
         this->s = &v;
     }
@@ -139,6 +135,12 @@ public:
         return *s == s2;
     }
 
+    [[gnu::always_inline]]
+    const StringData & string_data() const noexcept
+    {
+        return s->string_data();
+    }
+
     [[gnu::always_inline]]
     const char * c_str() const noexcept
     {
@@ -155,13 +157,17 @@ public:
     [[gnu::always_inline]]
     bool empty() const noexcept
     {
-        return s->size_ == 0;
+        auto * p = &s->string_data();
+        // Save a dereference in the sentinel value case
+        if (p == &""_sds)
+            return true;
+        return p->size() == 0;
     }
 
     [[gnu::always_inline]]
     size_t size() const noexcept
     {
-        return s->size_;
+        return s->string_data().size();
     }
 
     [[gnu::always_inline]]
@@ -259,7 +265,6 @@ private:
      * During its lifetime the monotonic buffer holds all strings and nodes, if the symbol set is node based.
      */
     std::pmr::monotonic_buffer_resource buffer;
-    std::pmr::polymorphic_allocator<char> stringAlloc{&buffer};
     SymbolStr::SymbolValueStore store{16};
 
     /**
@@ -282,7 +287,7 @@ public:
         // Most symbols are looked up more than once, so we trade off insertion performance
         // for lookup performance.
         // FIXME: make this thread-safe.
-        return Symbol(*symbols.insert(SymbolStr::Key{store, s, stringAlloc}).first);
+        return Symbol(*symbols.insert(SymbolStr::Key{store, s, buffer}).first);
     }
 
     std::vector<SymbolStr> resolve(const std::vector<Symbol> & symbols) const

src/libexpr/include/nix/expr/value.hh

@@ -1,8 +1,14 @@
 #pragma once
 ///@file
 
+#include <bit>
 #include <cassert>
+#include <cstddef>
+#include <cstring>
+#include <memory>
+#include <memory_resource>
 #include <span>
+#include <string_view>
 #include <type_traits>
 #include <concepts>
 
@@ -82,6 +88,7 @@ class PosIdx;
 struct Pos;
 class StorePath;
 class EvalState;
+class EvalMemory;
 class XMLWriter;
 class Printer;
 
@@ -155,7 +162,7 @@ class ListBuilder
     Value * inlineElems[2] = {nullptr, nullptr};
 public:
     Value ** elems;
-    ListBuilder(size_t size);
+    ListBuilder(EvalMemory & mem, size_t size);
 
     // NOTE: Can be noexcept because we are just copying integral values and
     // raw pointers.
@@ -186,6 +193,91 @@ public:
     friend struct Value;
 };
 
+class StringData
+{
+public:
+    using size_type = std::size_t;
+
+    size_type size_;
+    char data_[];
+
+    /*
+     * This in particular ensures that we cannot have a `StringData`
+     * that we use by value, which is just what we want!
+     *
+     * Dynamically sized types aren't a thing in C++ and even flexible array
+     * members are a language extension and beyond the realm of standard C++.
+     * Technically, sizeof data_ member is 0 and the intended way to use flexible
+     * array members is to allocate sizeof(StrindData) + count * sizeof(char) bytes
+     * and the compiler will consider alignment restrictions for the FAM.
+     *
+     */
+
+    StringData(StringData &&) = delete;
+    StringData & operator=(StringData &&) = delete;
+    StringData(const StringData &) = delete;
+    StringData & operator=(const StringData &) = delete;
+    ~StringData() = default;
+
+private:
+    StringData() = delete;
+
+    explicit StringData(size_type size)
+        : size_(size)
+    {
+    }
+
+public:
+    /**
+     * Allocate StringData on the (possibly) GC-managed heap and copy
+     * the contents of s to it.
+     */
+    static const StringData & make(std::string_view s);
+
+    /**
+     * Allocate StringData on the (possibly) GC-managed heap.
+     * @param size Length of the string (without the NUL terminator).
+     */
+    static StringData & alloc(size_t size);
+
+    size_t size() const
+    {
+        return size_;
+    }
+
+    char * data() noexcept
+    {
+        return data_;
+    }
+
+    const char * data() const noexcept
+    {
+        return data_;
+    }
+
+    const char * c_str() const noexcept
+    {
+        return data_;
+    }
+
+    constexpr std::string_view view() const noexcept
+    {
+        return std::string_view(data_, size_);
+    }
+
+    template<size_t N>
+    struct Static;
+
+    static StringData & make(std::pmr::memory_resource & resource, std::string_view s)
+    {
+        auto & res =
+            *new (resource.allocate(sizeof(StringData) + s.size() + 1, alignof(StringData))) StringData(s.size());
+        std::memcpy(res.data_, s.data(), s.size());
+        res.data_[s.size()] = '\0';
+        return res;
+    }
+};
+
 namespace detail {
 
 /**
@@ -219,7 +311,7 @@ struct ValueBase
      */
     struct StringWithContext
     {
-        const char * c_str;
+        const StringData * str;
 
         /**
          * The type of the context itself.
@@ -234,7 +326,7 @@ struct ValueBase
          */
         struct Context
         {
-            using value_type = const char *;
+            using value_type = const StringData *;
             using size_type = std::size_t;
             using iterator = const value_type *;
 
@@ -273,7 +365,7 @@ struct ValueBase
             /**
              * @return null pointer when context.empty()
              */
-            static Context * fromBuilder(const NixStringContext & context);
+            static Context * fromBuilder(const NixStringContext & context, EvalMemory & mem);
         };
 
         /**
@@ -285,7 +377,7 @@ struct ValueBase
     struct Path
     {
         SourceAccessor * accessor;
-        const char * path;
+        const StringData * path;
     };
 
     struct Null
@@ -646,13 +738,13 @@ protected:
     void getStorage(StringWithContext & string) const noexcept
     {
         string.context = untagPointer<decltype(string.context)>(payload[0]);
-        string.c_str = std::bit_cast<const char *>(payload[1]);
+        string.str = std::bit_cast<const StringData *>(payload[1]);
     }
 
     void getStorage(Path & path) const noexcept
     {
         path.accessor = untagPointer<decltype(path.accessor)>(payload[0]);
-        path.path = std::bit_cast<const char *>(payload[1]);
+        path.path = std::bit_cast<const StringData *>(payload[1]);
     }
 
     void setStorage(NixInt integer) noexcept
@@ -697,7 +789,7 @@ protected:
 
     void setStorage(StringWithContext string) noexcept
     {
-        setUntaggablePayload<pdString>(string.context, string.c_str);
+        setUntaggablePayload<pdString>(string.context, string.str);
     }
 
     void setStorage(Path path) noexcept
@@ -1050,22 +1142,22 @@ public:
         setStorage(b);
     }
 
-    void mkStringNoCopy(const char * s, const Value::StringWithContext::Context * context = nullptr) noexcept
+    void mkStringNoCopy(const StringData & s, const Value::StringWithContext::Context * context = nullptr) noexcept
     {
-        setStorage(StringWithContext{.c_str = s, .context = context});
+        setStorage(StringWithContext{.str = &s, .context = context});
     }
 
     void mkString(std::string_view s);
 
-    void mkString(std::string_view s, const NixStringContext & context);
+    void mkString(std::string_view s, const NixStringContext & context, EvalMemory & mem);
 
-    void mkStringMove(const char * s, const NixStringContext & context);
+    void mkStringMove(const StringData & s, const NixStringContext & context, EvalMemory & mem);
 
     void mkPath(const SourcePath & path);
 
-    inline void mkPath(SourceAccessor * accessor, const char * path) noexcept
+    inline void mkPath(SourceAccessor * accessor, const StringData & path) noexcept
     {
-        setStorage(Path{.accessor = accessor, .path = path});
+        setStorage(Path{.accessor = accessor, .path = &path});
     }
 
     inline void mkNull() noexcept
@@ -1163,17 +1255,23 @@ public:
 
     SourcePath path() const
     {
-        return SourcePath(ref(pathAccessor()->shared_from_this()), CanonPath(CanonPath::unchecked_t(), pathStr()));
+        return SourcePath(
+            ref(pathAccessor()->shared_from_this()), CanonPath(CanonPath::unchecked_t(), std::string(pathStrView())));
     }
 
-    std::string_view string_view() const noexcept
+    const StringData & string_data() const noexcept
     {
-        return std::string_view{getStorage<StringWithContext>().c_str};
+        return *getStorage<StringWithContext>().str;
     }
 
     const char * c_str() const noexcept
     {
-        return getStorage<StringWithContext>().c_str;
+        return getStorage<StringWithContext>().str->data();
+    }
+
+    std::string_view string_view() const noexcept
+    {
+        return string_data().view();
     }
 
     const Value::StringWithContext::Context * context() const noexcept
@@ -1233,12 +1331,12 @@ public:
 
     const char * pathStr() const noexcept
     {
-        return getStorage<Path>().path;
+        return getStorage<Path>().path->c_str();
     }
 
     std::string_view pathStrView() const noexcept
     {
-        return std::string_view{getStorage<Path>().path};
+        return getStorage<Path>().path->view();
     }
 
     SourceAccessor * pathAccessor() const noexcept

src/libexpr/package.nix

@@ -70,11 +70,6 @@ mkMesonLibrary (finalAttrs: {
     nix-util
     nix-store
     nix-fetchers
-  ]
-  ++ finalAttrs.passthru.externalPropagatedBuildInputs;
-
-  # Hack for sake of the dev shell
-  passthru.externalPropagatedBuildInputs = [
     boost
     nlohmann_json
   ]

src/libexpr/parser.y

@@ -126,26 +126,26 @@ static Expr * makeCall(Exprs & exprs, PosIdx pos, Expr * fn, Expr * arg) {
 
 %define api.value.type variant
 
-%type <nix::Expr *> start expr expr_function expr_if expr_op
-%type <nix::Expr *> expr_select expr_simple expr_app
-%type <nix::Expr *> expr_pipe_from expr_pipe_into
+%type <Expr *> start expr expr_function expr_if expr_op
+%type <Expr *> expr_select expr_simple expr_app
+%type <Expr *> expr_pipe_from expr_pipe_into
 %type <std::vector<Expr *>> list
-%type <nix::ExprAttrs *> binds binds1
-%type <nix::FormalsBuilder> formals formal_set
-%type <nix::Formal> formal
-%type <std::vector<nix::AttrName>> attrpath
-%type <std::vector<std::pair<nix::AttrName, nix::PosIdx>>> attrs
-%type <std::vector<std::pair<nix::PosIdx, nix::Expr *>>> string_parts_interpolated
-%type <std::vector<std::pair<nix::PosIdx, std::variant<nix::Expr *, nix::StringToken>>>> ind_string_parts
-%type <nix::Expr *> path_start
-%type <std::variant<nix::Expr *, std::string_view>> string_parts string_attr
-%type <nix::StringToken> attr
-%token <nix::StringToken> ID
-%token <nix::StringToken> STR IND_STR
-%token <nix::NixInt> INT_LIT
-%token <nix::NixFloat> FLOAT_LIT
-%token <nix::StringToken> PATH HPATH SPATH PATH_END
-%token <nix::StringToken> URI
+%type <ExprAttrs *> binds binds1
+%type <FormalsBuilder> formals formal_set
+%type <Formal> formal
+%type <std::vector<AttrName>> attrpath
+%type <std::vector<std::pair<AttrName, PosIdx>>> attrs
+%type <std::vector<std::pair<PosIdx, Expr *>>> string_parts_interpolated
+%type <std::vector<std::pair<PosIdx, std::variant<Expr *, StringToken>>>> ind_string_parts
+%type <Expr *> path_start
+%type <std::variant<Expr *, std::string_view>> string_parts string_attr
+%type <StringToken> attr
+%token <StringToken> ID
+%token <StringToken> STR IND_STR
+%token <NixInt> INT_LIT
+%token <NixFloat> FLOAT_LIT
+%token <StringToken> PATH HPATH SPATH PATH_END
+%token <StringToken> URI
 %token IF THEN ELSE ASSERT WITH LET IN_KW REC INHERIT EQ NEQ AND OR IMPL OR_KW
 %token PIPE_FROM PIPE_INTO /* <| and |> */
 %token DOLLAR_CURLY /* == ${ */
@@ -186,7 +186,7 @@ expr_function
   | formal_set ':' expr_function[body]
     {
       state->validateFormals($formal_set);
-      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, std::move($formal_set), $body);
+      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, $formal_set, $body);
       $$ = me;
       SET_DOC_POS(me, @1);
     }
@@ -194,7 +194,7 @@ expr_function
     {
       auto arg = state->symbols.create($ID);
       state->validateFormals($formal_set, CUR_POS, arg);
-      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, arg, std::move($formal_set), $body);
+      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, arg, $formal_set, $body);
       $$ = me;
       SET_DOC_POS(me, @1);
     }
@@ -202,7 +202,7 @@ expr_function
     {
       auto arg = state->symbols.create($ID);
       state->validateFormals($formal_set, CUR_POS, arg);
-      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, arg, std::move($formal_set), $body);
+      auto me = state->exprs.add<ExprLambda>(state->positions, state->exprs.alloc, CUR_POS, arg, $formal_set, $body);
       $$ = me;
       SET_DOC_POS(me, @1);
     }
@@ -251,7 +251,7 @@ expr_op
   | expr_op OR expr_op { $$ = state->exprs.add<ExprOpOr>(state->at(@2), $1, $3); }
   | expr_op IMPL expr_op { $$ = state->exprs.add<ExprOpImpl>(state->at(@2), $1, $3); }
   | expr_op UPDATE expr_op { $$ = state->exprs.add<ExprOpUpdate>(state->at(@2), $1, $3); }
-  | expr_op '?' attrpath { $$ = state->exprs.add<ExprOpHasAttr>(state->exprs.alloc, $1, std::move($3)); }
+  | expr_op '?' attrpath { $$ = state->exprs.add<ExprOpHasAttr>(state->exprs.alloc, $1, $3); }
   | expr_op '+' expr_op
     { $$ = state->exprs.add<ExprConcatStrings>(state->exprs.alloc, state->at(@2), false, {{state->at(@1), $1}, {state->at(@3), $3}}); }
   | expr_op '-' expr_op { $$ = state->exprs.add<ExprCall>(state->at(@2), state->exprs.add<ExprVar>(state->s.sub), {$1, $3}); }
@@ -272,9 +272,9 @@ expr_app
 
 expr_select
   : expr_simple '.' attrpath
-    { $$ = state->exprs.add<ExprSelect>(state->exprs.alloc, CUR_POS, $1, std::move($3), nullptr); }
+    { $$ = state->exprs.add<ExprSelect>(state->exprs.alloc, CUR_POS, $1, $3, nullptr); }
   | expr_simple '.' attrpath OR_KW expr_select
-    { $$ = state->exprs.add<ExprSelect>(state->exprs.alloc, CUR_POS, $1, std::move($3), $5); $5->warnIfCursedOr(state->symbols, state->positions); }
+    { $$ = state->exprs.add<ExprSelect>(state->exprs.alloc, CUR_POS, $1, $3, $5); $5->warnIfCursedOr(state->symbols, state->positions); }
   | /* Backwards compatibility: because Nixpkgs has a function named ‘or’,
        allow stuff like ‘map or [...]’. This production is problematic (see
        https://github.com/NixOS/nix/issues/11118) and will be refactored in the
@@ -304,12 +304,12 @@ expr_simple
       $2);
   }
   | IND_STRING_OPEN ind_string_parts IND_STRING_CLOSE {
-      $$ = state->stripIndentation(CUR_POS, std::move($2));
+      $$ = state->stripIndentation(CUR_POS, $2);
   }
   | path_start PATH_END
   | path_start string_parts_interpolated PATH_END {
       $2.insert($2.begin(), {state->at(@1), $1});
-      $$ = state->exprs.add<ExprConcatStrings>(state->exprs.alloc, CUR_POS, false, std::move($2));
+      $$ = state->exprs.add<ExprConcatStrings>(state->exprs.alloc, CUR_POS, false, $2);
   }
   | SPATH {
       std::string_view path($1.p + 1, $1.l - 2);
@@ -338,12 +338,12 @@ expr_simple
     { $2->pos = CUR_POS; $$ = $2; }
   | '{' '}'
     { $$ = state->exprs.add<ExprAttrs>(CUR_POS); }
-  | '[' list ']' { $$ = state->exprs.add<ExprList>(state->exprs.alloc, std::move($2)); }
+  | '[' list ']' { $$ = state->exprs.add<ExprList>(state->exprs.alloc, $2); }
   ;
 
 string_parts
   : STR { $$ = $1; }
-  | string_parts_interpolated { $$ = state->exprs.add<ExprConcatStrings>(state->exprs.alloc, CUR_POS, true, std::move($1)); }
+  | string_parts_interpolated { $$ = state->exprs.add<ExprConcatStrings>(state->exprs.alloc, CUR_POS, true, $1); }
   | { $$ = std::string_view(); }
   ;
 
@@ -425,7 +425,7 @@ binds1
       if (!$accum->inheritFromExprs)
           $accum->inheritFromExprs = std::make_unique<std::vector<Expr *>>();
       $accum->inheritFromExprs->push_back($expr);
-      auto from = new nix::ExprInheritFrom(state->at(@expr), $accum->inheritFromExprs->size() - 1);
+      auto from = state->exprs.add<ExprInheritFrom>(state->at(@expr), $accum->inheritFromExprs->size() - 1);
       for (auto & [i, iPos] : $attrs) {
           if ($accum->attrs.find(i.symbol) != $accum->attrs.end())
               state->dupAttr(i.symbol, iPos, $accum->attrs[i.symbol].pos);

src/libexpr/primops.cc

@@ -5,6 +5,7 @@
 #include "nix/expr/eval-settings.hh"
 #include "nix/expr/gc-small-vector.hh"
 #include "nix/expr/json-to-value.hh"
+#include "nix/expr/static-string-data.hh"
 #include "nix/store/globals.hh"
 #include "nix/store/names.hh"
 #include "nix/store/path-references.hh"
@@ -39,6 +40,12 @@
 
 namespace nix {
 
+RegisterPrimOp::PrimOps & RegisterPrimOp::primOps()
+{
+    static RegisterPrimOp::PrimOps primOps;
+    return primOps;
+}
+
 /*************************************************************
  * Miscellaneous
  *************************************************************/
@@ -221,7 +228,8 @@ void derivationToValue(
             path2,
             {
                 NixStringContextElem::DrvDeep{.drvPath = storePath},
-            });
+            },
+            state.mem);
     attrs.alloc(state.s.name).mkString(drv.env["name"]);
 
     auto list = state.buildList(drv.outputs.size());
@@ -487,34 +495,34 @@ static void prim_typeOf(EvalState & state, const PosIdx pos, Value ** args, Valu
     state.forceValue(*args[0], pos);
     switch (args[0]->type()) {
     case nInt:
-        v.mkStringNoCopy("int");
+        v.mkStringNoCopy("int"_sds);
         break;
     case nBool:
-        v.mkStringNoCopy("bool");
+        v.mkStringNoCopy("bool"_sds);
         break;
     case nString:
-        v.mkStringNoCopy("string");
+        v.mkStringNoCopy("string"_sds);
         break;
     case nPath:
-        v.mkStringNoCopy("path");
+        v.mkStringNoCopy("path"_sds);
         break;
     case nNull:
-        v.mkStringNoCopy("null");
+        v.mkStringNoCopy("null"_sds);
         break;
     case nAttrs:
-        v.mkStringNoCopy("set");
+        v.mkStringNoCopy("set"_sds);
         break;
     case nList:
-        v.mkStringNoCopy("list");
+        v.mkStringNoCopy("list"_sds);
         break;
     case nFunction:
-        v.mkStringNoCopy("lambda");
+        v.mkStringNoCopy("lambda"_sds);
         break;
     case nExternal:
         v.mkString(args[0]->external()->typeOf());
         break;
     case nFloat:
-        v.mkStringNoCopy("float");
+        v.mkStringNoCopy("float"_sds);
         break;
     case nThunk:
         unreachable();
@@ -1810,7 +1818,8 @@ static void derivationStrictInternal(EvalState & state, std::string_view drvName
             drvPathS,
             {
                 NixStringContextElem::DrvDeep{.drvPath = drvPath},
-            });
+            },
+            state.mem);
     for (auto & i : drv.outputs)
         mkOutputString(state, result, drvPath, i);
 
@@ -1863,7 +1872,7 @@ static void prim_toPath(EvalState & state, const PosIdx pos, Value ** args, Valu
     NixStringContext context;
     auto path =
         state.coerceToPath(pos, *args[0], context, "while evaluating the first argument passed to builtins.toPath");
-    v.mkString(path.path.abs(), context);
+    v.mkString(path.path.abs(), context, state.mem);
 }
 
 static RegisterPrimOp primop_toPath({
@@ -1906,7 +1915,7 @@ static void prim_storePath(EvalState & state, const PosIdx pos, Value ** args, V
     if (!settings.readOnlyMode)
         state.store->ensurePath(path2);
     context.insert(NixStringContextElem::Opaque{.path = path2});
-    v.mkString(path.abs(), context);
+    v.mkString(path.abs(), context, state.mem);
 }
 
 static RegisterPrimOp primop_storePath({
@@ -1988,7 +1997,8 @@ static void prim_baseNameOf(EvalState & state, const PosIdx pos, Value ** args,
     v.mkString(
         legacyBaseNameOf(*state.coerceToString(
             pos, *args[0], context, "while evaluating the first argument passed to builtins.baseNameOf", false, false)),
-        context);
+        context,
+        state.mem);
 }
 
 static RegisterPrimOp primop_baseNameOf({
@@ -2024,11 +2034,11 @@ static void prim_dirOf(EvalState & state, const PosIdx pos, Value ** args, Value
             pos, *args[0], context, "while evaluating the first argument passed to 'builtins.dirOf'", false, false);
         auto pos = path->rfind('/');
         if (pos == path->npos)
-            v.mkStringMove(".", context);
+            v.mkStringMove("."_sds, context, state.mem);
         else if (pos == 0)
-            v.mkStringMove("/", context);
+            v.mkStringMove("/"_sds, context, state.mem);
         else
-            v.mkString(path->substr(0, pos), context);
+            v.mkString(path->substr(0, pos), context, state.mem);
     }
 }
 
@@ -2070,7 +2080,7 @@ static void prim_readFile(EvalState & state, const PosIdx pos, Value ** args, Va
                 .path = std::move((StorePath &&) p),
             });
     }
-    v.mkString(s, context);
+    v.mkString(s, context, state.mem);
 }
 
 static RegisterPrimOp primop_readFile({
@@ -2309,10 +2319,10 @@ static const Value & fileTypeToString(EvalState & state, SourceAccessor::Type ty
 
     static const Constants stringValues = []() {
         Constants res;
-        res.regular.mkStringNoCopy("regular");
-        res.directory.mkStringNoCopy("directory");
-        res.symlink.mkStringNoCopy("symlink");
-        res.unknown.mkStringNoCopy("unknown");
+        res.regular.mkStringNoCopy("regular"_sds);
+        res.directory.mkStringNoCopy("directory"_sds);
+        res.symlink.mkStringNoCopy("symlink"_sds);
+        res.unknown.mkStringNoCopy("unknown"_sds);
         return res;
     }();
 
@@ -2466,7 +2476,7 @@ static void prim_toXML(EvalState & state, const PosIdx pos, Value ** args, Value
     std::ostringstream out;
     NixStringContext context;
     printValueAsXML(state, true, false, *args[0], out, context, pos);
-    v.mkString(out.view(), context);
+    v.mkString(out.view(), context, state.mem);
 }
 
 static RegisterPrimOp primop_toXML({
@@ -2574,7 +2584,7 @@ static void prim_toJSON(EvalState & state, const PosIdx pos, Value ** args, Valu
     std::ostringstream out;
     NixStringContext context;
     printValueAsJSON(state, true, *args[0], pos, out, context);
-    v.mkString(out.view(), context);
+    v.mkString(out.view(), context, state.mem);
 }
 
 static RegisterPrimOp primop_toJSON({
@@ -4403,7 +4413,7 @@ static void prim_toString(EvalState & state, const PosIdx pos, Value ** args, Va
     NixStringContext context;
     auto s = state.coerceToString(
         pos, *args[0], context, "while evaluating the first argument passed to builtins.toString", true, false);
-    v.mkString(*s, context);
+    v.mkString(*s, context, state.mem);
 }
 
 static RegisterPrimOp primop_toString({
@@ -4463,7 +4473,7 @@ static void prim_substring(EvalState & state, const PosIdx pos, Value ** args, V
     if (len == 0) {
         state.forceValue(*args[2], pos);
         if (args[2]->type() == nString) {
-            v.mkStringNoCopy("", args[2]->context());
+            v.mkStringNoCopy(""_sds, args[2]->context());
             return;
         }
     }
@@ -4476,7 +4486,7 @@ static void prim_substring(EvalState & state, const PosIdx pos, Value ** args, V
     auto s = state.coerceToString(
         pos, *args[2], context, "while evaluating the third argument (the string) passed to builtins.substring");
 
-    v.mkString(NixUInt(start) >= s->size() ? "" : s->substr(start, _len), context);
+    v.mkString(NixUInt(start) >= s->size() ? "" : s->substr(start, _len), context, state.mem);
 }
 
 static RegisterPrimOp primop_substring({
@@ -4874,7 +4884,7 @@ static void prim_concatStringsSep(EvalState & state, const PosIdx pos, Value **
             "while evaluating one element of the list of strings to concat passed to builtins.concatStringsSep");
     }
 
-    v.mkString(res, context);
+    v.mkString(res, context, state.mem);
 }
 
 static RegisterPrimOp primop_concatStringsSep({
@@ -4949,7 +4959,7 @@ static void prim_replaceStrings(EvalState & state, const PosIdx pos, Value ** ar
         }
     }
 
-    v.mkString(res, context);
+    v.mkString(res, context, state.mem);
 }
 
 static RegisterPrimOp primop_replaceStrings({

src/libexpr/primops/context.cc

@@ -69,7 +69,7 @@ static void prim_unsafeDiscardOutputDependency(EvalState & state, const PosIdx p
         }
     }
 
-    v.mkString(*s, context2);
+    v.mkString(*s, context2, state.mem);
 }
 
 static RegisterPrimOp primop_unsafeDiscardOutputDependency(
@@ -137,7 +137,7 @@ static void prim_addDrvOutputDependencies(EvalState & state, const PosIdx pos, V
             context.begin()->raw)}),
     };
 
-    v.mkString(*s, context2);
+    v.mkString(*s, context2, state.mem);
 }
 
 static RegisterPrimOp primop_addDrvOutputDependencies(
@@ -321,7 +321,7 @@ static void prim_appendContext(EvalState & state, const PosIdx pos, Value ** arg
         }
     }
 
-    v.mkString(orig, context);
+    v.mkString(orig, context, state.mem);
 }
 
 static RegisterPrimOp primop_appendContext({.name = "__appendContext", .arity = 2, .fun = prim_appendContext});

src/libexpr/primops/fetchMercurial.cc

@@ -81,7 +81,7 @@ static void prim_fetchMercurial(EvalState & state, const PosIdx pos, Value ** ar
         attrs.insert_or_assign("rev", rev->gitRev());
     auto input = fetchers::Input::fromAttrs(state.fetchSettings, std::move(attrs));
 
-    auto [storePath, input2] = input.fetchToStore(state.store);
+    auto [storePath, input2] = input.fetchToStore(state.fetchSettings, state.store);
 
     auto attrs2 = state.buildBindings(8);
     state.mkStorePathString(storePath, attrs2.alloc(state.s.outPath));

src/libexpr/primops/fetchTree.cc

@@ -82,7 +82,7 @@ struct FetchTreeParams
 static void fetchTree(
     EvalState & state, const PosIdx pos, Value ** args, Value & v, const FetchTreeParams & params = FetchTreeParams{})
 {
-    fetchers::Input input{state.fetchSettings};
+    fetchers::Input input{};
     NixStringContext context;
     std::optional<std::string> type;
     auto fetcher = params.isFetchGit ? "fetchGit" : "fetchTree";
@@ -194,9 +194,9 @@ static void fetchTree(
     }
 
     if (!state.settings.pureEval && !input.isDirect() && experimentalFeatureSettings.isEnabled(Xp::Flakes))
-        input = lookupInRegistries(state.store, input, fetchers::UseRegistries::Limited).first;
+        input = lookupInRegistries(state.fetchSettings, state.store, input, fetchers::UseRegistries::Limited).first;
 
-    if (state.settings.pureEval && !input.isLocked()) {
+    if (state.settings.pureEval && !input.isLocked(state.fetchSettings)) {
         if (input.getNarHash())
             warn(
                 "Input '%s' is unlocked (e.g. lacks a Git revision) but is checked by NAR hash. "
@@ -219,7 +219,8 @@ static void fetchTree(
             throw Error("input '%s' is not allowed to use the '__final' attribute", input.to_string());
     }
 
-    auto cachedInput = state.inputCache->getAccessor(state.store, input, fetchers::UseRegistries::No);
+    auto cachedInput =
+        state.inputCache->getAccessor(state.fetchSettings, state.store, input, fetchers::UseRegistries::No);
 
     auto storePath = state.mountInput(cachedInput.lockedInput, input, cachedInput.accessor);
 
@@ -234,229 +235,127 @@ static void prim_fetchTree(EvalState & state, const PosIdx pos, Value ** args, V
 static RegisterPrimOp primop_fetchTree({
     .name = "fetchTree",
     .args = {"input"},
-    .doc = R"(
-      Fetch a file system tree or a plain file using one of the supported backends and return an attribute set with:
+    .doc = []() -> std::string {
+        std::string doc = stripIndentation(R"(
+          Fetch a file system tree or a plain file using one of the supported backends and return an attribute set with:
 
-      - the resulting fixed-output [store path](@docroot@/store/store-path.md)
-      - the corresponding [NAR](@docroot@/store/file-system-object/content-address.md#serial-nix-archive) hash
-      - backend-specific metadata (currently not documented). <!-- TODO: document output attributes -->
+          - the resulting fixed-output [store path](@docroot@/store/store-path.md)
+          - the corresponding [NAR](@docroot@/store/file-system-object/content-address.md#serial-nix-archive) hash
+          - backend-specific metadata (currently not documented). <!-- TODO: document output attributes -->
 
-      *input* must be an attribute set with the following attributes:
+          *input* must be an attribute set with the following attributes:
 
-      - `type` (String, required)
+          - `type` (String, required)
 
-        One of the [supported source types](#source-types).
-        This determines other required and allowed input attributes.
+            One of the [supported source types](#source-types).
+            This determines other required and allowed input attributes.
 
-      - `narHash` (String, optional)
+          - `narHash` (String, optional)
 
-        The `narHash` parameter can be used to substitute the source of the tree.
-        It also allows for verification of tree contents that may not be provided by the underlying transfer mechanism.
-        If `narHash` is set, the source is first looked up is the Nix store and [substituters](@docroot@/command-ref/conf-file.md#conf-substituters), and only fetched if not available.
+            The `narHash` parameter can be used to substitute the source of the tree.
+            It also allows for verification of tree contents that may not be provided by the underlying transfer mechanism.
+            If `narHash` is set, the source is first looked up is the Nix store and [substituters](@docroot@/command-ref/conf-file.md#conf-substituters), and only fetched if not available.
 
-      A subset of the output attributes of `fetchTree` can be re-used for subsequent calls to `fetchTree` to produce the same result again.
-      That is, `fetchTree` is idempotent.
+          A subset of the output attributes of `fetchTree` can be re-used for subsequent calls to `fetchTree` to produce the same result again.
+          That is, `fetchTree` is idempotent.
 
-      Downloads are cached in `$XDG_CACHE_HOME/nix`.
-      The remote source is fetched from the network if both are true:
-      - A NAR hash is supplied and the corresponding store path is not [valid](@docroot@/glossary.md#gloss-validity), that is, not available in the store
+          Downloads are cached in `$XDG_CACHE_HOME/nix`.
+          The remote source is fetched from the network if both are true:
+          - A NAR hash is supplied and the corresponding store path is not [valid](@docroot@/glossary.md#gloss-validity), that is, not available in the store
 
-        > **Note**
-        >
-        > [Substituters](@docroot@/command-ref/conf-file.md#conf-substituters) are not used in fetching.
-
-      - There is no cache entry or the cache entry is older than [`tarball-ttl`](@docroot@/command-ref/conf-file.md#conf-tarball-ttl)
-
-      ## Source types
-
-      The following source types and associated input attributes are supported.
-
-      <!-- TODO: It would be soooo much more predictable to work with (and
-      document) if `fetchTree` was a curried call with the first parameter for
-      `type` or an attribute like `builtins.fetchTree.git`! -->
-
-      - `"file"`
-
-        Place a plain file into the Nix store.
-        This is similar to [`builtins.fetchurl`](@docroot@/language/builtins.md#builtins-fetchurl)
-
-        - `url` (String, required)
-
-          Supported protocols:
-
-          - `https`
-
-            > **Example**
+            > **Note**
             >
-            > ```nix
-            > fetchTree {
-            >   type = "file";
-            >   url = "https://example.com/index.html";
-            > }
-            > ```
+            > [Substituters](@docroot@/command-ref/conf-file.md#conf-substituters) are not used in fetching.
 
-          - `http`
+          - There is no cache entry or the cache entry is older than [`tarball-ttl`](@docroot@/command-ref/conf-file.md#conf-tarball-ttl)
 
-            Insecure HTTP transfer for legacy sources.
+          ## Source types
 
-            > **Warning**
-            >
-            > HTTP performs no encryption or authentication.
-            > Use a `narHash` known in advance to ensure the output has expected contents.
+          The following source types and associated input attributes are supported.
 
-          - `file`
+          <!-- TODO: It would be soooo much more predictable to work with (and
+          document) if `fetchTree` was a curried call with the first parameter for
+          `type` or an attribute like `builtins.fetchTree.git`! -->
+        )");
 
-            A file on the local file system.
+        auto indentString = [](std::string const & str, std::string const & indent) {
+            std::string result;
+            std::istringstream stream(str);
+            std::string line;
+            bool first = true;
+            while (std::getline(stream, line)) {
+                if (!first)
+                    result += "\n";
+                result += indent + line;
+                first = false;
+            }
+            return result;
+        };
 
-            > **Example**
-            >
-            > ```nix
-            > fetchTree {
-            >   type = "file";
-            >   url = "file:///home/eelco/nix/README.md";
-            > }
-            > ```
+        for (const auto & [schemeName, scheme] : fetchers::getAllInputSchemes()) {
+            doc += "\n- `" + quoteString(schemeName, '"') + "`\n\n";
+            doc += indentString(scheme->schemeDescription(), "  ");
+            if (!doc.empty() && doc.back() != '\n')
+                doc += "\n";
 
-      - `"tarball"`
+            for (const auto & [attrName, attribute] : scheme->allowedAttrs()) {
+                doc += "\n  - `" + attrName + "` (" + attribute.type + ", "
+                       + (attribute.required ? "required" : "optional") + ")\n\n";
+                doc += indentString(stripIndentation(attribute.doc), "    ");
+                if (!doc.empty() && doc.back() != '\n')
+                    doc += "\n";
+            }
+        }
 
-        Download a tar archive and extract it into the Nix store.
-        This has the same underlying implementation as [`builtins.fetchTarball`](@docroot@/language/builtins.md#builtins-fetchTarball)
+        doc += "\n" + stripIndentation(R"(
+          The following input types are still subject to change:
 
-        - `url` (String, required)
+          - `"path"`
+          - `"github"`
+          - `"gitlab"`
+          - `"sourcehut"`
+          - `"mercurial"`
 
-           > **Example**
-           >
-           > ```nix
-           > fetchTree {
-           >   type = "tarball";
-           >   url = "https://github.com/NixOS/nixpkgs/tarball/nixpkgs-23.11";
-           > }
-           > ```
-
-      - `"git"`
-
-        Fetch a Git tree and copy it to the Nix store.
-        This is similar to [`builtins.fetchGit`](@docroot@/language/builtins.md#builtins-fetchGit).
-
-        - `url` (String, required)
-
-          The URL formats supported are the same as for Git itself.
+         *input* can also be a [URL-like reference](@docroot@/command-ref/new-cli/nix3-flake.md#flake-references).
+         The additional input types and the URL-like syntax requires the [`flakes` experimental feature](@docroot@/development/experimental-features.md#xp-feature-flakes) to be enabled.
 
           > **Example**
           >
+          > Fetch a GitHub repository using the attribute set representation:
+          >
           > ```nix
-          > fetchTree {
-          >   type = "git";
-          >   url = "git@github.com:NixOS/nixpkgs.git";
+          > builtins.fetchTree {
+          >   type = "github";
+          >   owner = "NixOS";
+          >   repo = "nixpkgs";
+          >   rev = "ae2e6b3958682513d28f7d633734571fb18285dd";
+          > }
+          > ```
+          >
+          > This evaluates to the following attribute set:
+          >
+          > ```nix
+          > {
+          >   lastModified = 1686503798;
+          >   lastModifiedDate = "20230611171638";
+          >   narHash = "sha256-rA9RqKP9OlBrgGCPvfd5HVAXDOy8k2SmPtB/ijShNXc=";
+          >   outPath = "/nix/store/l5m6qlvfs9sdw14ja3qbzpglcjlb6j1x-source";
+          >   rev = "ae2e6b3958682513d28f7d633734571fb18285dd";
+          >   shortRev = "ae2e6b3";
           > }
           > ```
 
-          > **Note**
+          > **Example**
           >
-          > If the URL points to a local directory, and no `ref` or `rev` is given, Nix only considers files added to the Git index, as listed by `git ls-files` but use the *current file contents* of the Git working directory.
+          > Fetch the same GitHub repository using the URL-like syntax:
+          >
+          >   ```nix
+          >   builtins.fetchTree "github:NixOS/nixpkgs/ae2e6b3958682513d28f7d633734571fb18285dd"
+          >   ```
+        )");
 
-        - `ref` (String, optional)
-
-          By default, this has no effect. This becomes relevant only once `shallow` cloning is disabled.
-
-          A [Git reference](https://git-scm.com/book/en/v2/Git-Internals-Git-References), such as a branch or tag name.
-
-          Default: `"HEAD"`
-
-        - `rev` (String, optional)
-
-          A Git revision; a commit hash.
-
-          Default: the tip of `ref`
-
-        - `shallow` (Bool, optional)
-
-          Make a shallow clone when fetching the Git tree.
-          When this is enabled, the options `ref` and `allRefs` have no effect anymore.
-
-          Default: `true`
-
-        - `submodules` (Bool, optional)
-
-          Also fetch submodules if available.
-
-          Default: `false`
-
-        - `lfs` (Bool, optional)
-
-          Fetch any [Git LFS](https://git-lfs.com/) files.
-
-          Default: `false`
-
-        - `allRefs` (Bool, optional)
-
-          By default, this has no effect. This becomes relevant only once `shallow` cloning is disabled.
-
-          Whether to fetch all references (eg. branches and tags) of the repository.
-          With this argument being true, it's possible to load a `rev` from *any* `ref`.
-          (Without setting this option, only `rev`s from the specified `ref` are supported).
-
-          Default: `false`
-
-        - `lastModified` (Integer, optional)
-
-          Unix timestamp of the fetched commit.
-
-          If set, pass through the value to the output attribute set.
-          Otherwise, generated from the fetched Git tree.
-
-        - `revCount` (Integer, optional)
-
-          Number of revisions in the history of the Git repository before the fetched commit.
-
-          If set, pass through the value to the output attribute set.
-          Otherwise, generated from the fetched Git tree.
-
-      The following input types are still subject to change:
-
-      - `"path"`
-      - `"github"`
-      - `"gitlab"`
-      - `"sourcehut"`
-      - `"mercurial"`
-
-     *input* can also be a [URL-like reference](@docroot@/command-ref/new-cli/nix3-flake.md#flake-references).
-     The additional input types and the URL-like syntax requires the [`flakes` experimental feature](@docroot@/development/experimental-features.md#xp-feature-flakes) to be enabled.
-
-      > **Example**
-      >
-      > Fetch a GitHub repository using the attribute set representation:
-      >
-      > ```nix
-      > builtins.fetchTree {
-      >   type = "github";
-      >   owner = "NixOS";
-      >   repo = "nixpkgs";
-      >   rev = "ae2e6b3958682513d28f7d633734571fb18285dd";
-      > }
-      > ```
-      >
-      > This evaluates to the following attribute set:
-      >
-      > ```nix
-      > {
-      >   lastModified = 1686503798;
-      >   lastModifiedDate = "20230611171638";
-      >   narHash = "sha256-rA9RqKP9OlBrgGCPvfd5HVAXDOy8k2SmPtB/ijShNXc=";
-      >   outPath = "/nix/store/l5m6qlvfs9sdw14ja3qbzpglcjlb6j1x-source";
-      >   rev = "ae2e6b3958682513d28f7d633734571fb18285dd";
-      >   shortRev = "ae2e6b3";
-      > }
-      > ```
-
-      > **Example**
-      >
-      > Fetch the same GitHub repository using the URL-like syntax:
-      >
-      >   ```nix
-      >   builtins.fetchTree "github:NixOS/nixpkgs/ae2e6b3958682513d28f7d633734571fb18285dd"
-      >   ```
-    )",
+        return doc;
+    }(),
     .fun = prim_fetchTree,
     .experimentalFeature = Xp::FetchTree,
 });

src/libexpr/primops/fromTOML.cc

@@ -1,5 +1,6 @@
 #include "nix/expr/primops.hh"
 #include "nix/expr/eval-inline.hh"
+#include "nix/expr/static-string-data.hh"
 
 #include "expr-config-private.hh"
 
@@ -136,7 +137,7 @@ static void prim_fromTOML(EvalState & state, const PosIdx pos, Value ** args, Va
                 normalizeDatetimeFormat(t);
 #endif
                 auto attrs = state.buildBindings(2);
-                attrs.alloc("_type").mkStringNoCopy("timestamp");
+                attrs.alloc("_type").mkStringNoCopy("timestamp"_sds);
                 std::ostringstream s;
                 s << t;
                 auto str = s.view();

src/libfetchers-tests/git.cc

@@ -196,7 +196,7 @@ TEST_F(GitTest, submodulePeriodSupport)
             {"ref", "main"},
         });
 
-    auto [accessor, i] = input.getAccessor(store);
+    auto [accessor, i] = input.getAccessor(settings, store);
 
     ASSERT_EQ(accessor->readFile(CanonPath("deps/sub/lib.txt")), "hello from submodule\n");
 }

src/libfetchers-tests/input.cc

@@ -0,0 +1,61 @@
+#include "nix/fetchers/fetch-settings.hh"
+#include "nix/fetchers/attrs.hh"
+#include "nix/fetchers/fetchers.hh"
+
+#include <gtest/gtest.h>
+
+#include <string>
+
+namespace nix {
+
+using fetchers::Attr;
+
+struct InputFromAttrsTestCase
+{
+    fetchers::Attrs attrs;
+    std::string expectedUrl;
+    std::string description;
+    fetchers::Attrs expectedAttrs = attrs;
+};
+
+class InputFromAttrsTest : public ::testing::WithParamInterface<InputFromAttrsTestCase>, public ::testing::Test
+{};
+
+TEST_P(InputFromAttrsTest, attrsAreCorrectAndRoundTrips)
+{
+    fetchers::Settings fetchSettings;
+
+    const auto & testCase = GetParam();
+
+    auto input = fetchers::Input::fromAttrs(fetchSettings, fetchers::Attrs(testCase.attrs));
+
+    EXPECT_EQ(input.toAttrs(), testCase.expectedAttrs);
+    EXPECT_EQ(input.toURLString(), testCase.expectedUrl);
+
+    auto input2 = fetchers::Input::fromAttrs(fetchSettings, input.toAttrs());
+    EXPECT_EQ(input, input2);
+    EXPECT_EQ(input.toAttrs(), input2.toAttrs());
+}
+
+INSTANTIATE_TEST_SUITE_P(
+    InputFromAttrs,
+    InputFromAttrsTest,
+    ::testing::Values(
+        // Test for issue #14429.
+        InputFromAttrsTestCase{
+            .attrs =
+                {
+                    {"url", Attr("git+ssh://git@github.com/NixOS/nixpkgs")},
+                    {"type", Attr("git")},
+                },
+            .expectedUrl = "git+ssh://git@github.com/NixOS/nixpkgs",
+            .description = "strips_git_plus_prefix",
+            .expectedAttrs =
+                {
+                    {"url", Attr("ssh://git@github.com/NixOS/nixpkgs")},
+                    {"type", Attr("git")},
+                },
+        }),
+    [](const ::testing::TestParamInfo<InputFromAttrsTestCase> & info) { return info.param.description; });
+
+} // namespace nix

src/libfetchers-tests/meson.build

@@ -42,6 +42,7 @@ sources = files(
   'access-tokens.cc',
   'git-utils.cc',
   'git.cc',
+  'input.cc',
   'nix_api_fetchers.cc',
   'public-key.cc',
 )

src/libfetchers/fetchers.cc

@@ -26,18 +26,9 @@ void registerInputScheme(std::shared_ptr<InputScheme> && inputScheme)
         throw Error("Input scheme with name %s already registered", schemeName);
 }
 
-nlohmann::json dumpRegisterInputSchemeInfo()
+const InputSchemeMap & getAllInputSchemes()
 {
-    using nlohmann::json;
-
-    auto res = json::object();
-
-    for (auto & [name, scheme] : inputSchemes()) {
-        auto & r = res[name] = json::object();
-        r["allowedAttrs"] = scheme->allowedAttrs();
-    }
-
-    return res;
+    return inputSchemes();
 }
 
 Input Input::fromURL(const Settings & settings, const std::string & url, bool requireTree)
@@ -89,7 +80,7 @@ Input Input::fromAttrs(const Settings & settings, Attrs && attrs)
         // but not all of them. Doing this is to support those other
         // operations which are supposed to be robust on
         // unknown/uninterpretable inputs.
-        Input input{settings};
+        Input input;
         input.attrs = attrs;
         fixupInput(input);
         return input;
@@ -159,9 +150,9 @@ bool Input::isDirect() const
     return !scheme || scheme->isDirect(*this);
 }
 
-bool Input::isLocked() const
+bool Input::isLocked(const Settings & settings) const
 {
-    return scheme && scheme->isLocked(*this);
+    return scheme && scheme->isLocked(settings, *this);
 }
 
 bool Input::isFinal() const
@@ -198,17 +189,17 @@ bool Input::contains(const Input & other) const
 }
 
 // FIXME: remove
-std::pair<StorePath, Input> Input::fetchToStore(ref<Store> store) const
+std::pair<StorePath, Input> Input::fetchToStore(const Settings & settings, ref<Store> store) const
 {
     if (!scheme)
         throw Error("cannot fetch unsupported input '%s'", attrsToJSON(toAttrs()));
 
     auto [storePath, input] = [&]() -> std::pair<StorePath, Input> {
         try {
-            auto [accessor, result] = getAccessorUnchecked(store);
+            auto [accessor, result] = getAccessorUnchecked(settings, store);
 
             auto storePath =
-                nix::fetchToStore(*settings, *store, SourcePath(accessor), FetchMode::Copy, result.getName());
+                nix::fetchToStore(settings, *store, SourcePath(accessor), FetchMode::Copy, result.getName());
 
             auto narHash = store->queryPathInfo(storePath)->narHash;
             result.attrs.insert_or_assign("narHash", narHash.to_string(HashFormat::SRI, true));
@@ -297,10 +288,10 @@ void Input::checkLocks(Input specified, Input & result)
     }
 }
 
-std::pair<ref<SourceAccessor>, Input> Input::getAccessor(ref<Store> store) const
+std::pair<ref<SourceAccessor>, Input> Input::getAccessor(const Settings & settings, ref<Store> store) const
 {
     try {
-        auto [accessor, result] = getAccessorUnchecked(store);
+        auto [accessor, result] = getAccessorUnchecked(settings, store);
 
         result.attrs.insert_or_assign("__final", Explicit<bool>(true));
 
@@ -313,7 +304,7 @@ std::pair<ref<SourceAccessor>, Input> Input::getAccessor(ref<Store> store) const
     }
 }
 
-std::pair<ref<SourceAccessor>, Input> Input::getAccessorUnchecked(ref<Store> store) const
+std::pair<ref<SourceAccessor>, Input> Input::getAccessorUnchecked(const Settings & settings, ref<Store> store) const
 {
     // FIXME: cache the accessor
 
@@ -349,7 +340,7 @@ std::pair<ref<SourceAccessor>, Input> Input::getAccessorUnchecked(ref<Store> sto
             if (accessor->fingerprint) {
                 ContentAddressMethod method = ContentAddressMethod::Raw::NixArchive;
                 auto cacheKey = makeFetchToStoreCacheKey(getName(), *accessor->fingerprint, method, "/");
-                settings->getCache()->upsert(cacheKey, *store, {}, storePath);
+                settings.getCache()->upsert(cacheKey, *store, {}, storePath);
             }
 
             accessor->setPathDisplay("«" + to_string() + "»");
@@ -360,7 +351,7 @@ std::pair<ref<SourceAccessor>, Input> Input::getAccessorUnchecked(ref<Store> sto
         }
     }
 
-    auto [accessor, result] = scheme->getAccessor(store, *this);
+    auto [accessor, result] = scheme->getAccessor(settings, store, *this);
 
     if (!accessor->fingerprint)
         accessor->fingerprint = result.getFingerprint(store);
@@ -377,10 +368,10 @@ Input Input::applyOverrides(std::optional<std::string> ref, std::optional<Hash>
     return scheme->applyOverrides(*this, ref, rev);
 }
 
-void Input::clone(const Path & destDir) const
+void Input::clone(const Settings & settings, const Path & destDir) const
 {
     assert(scheme);
-    scheme->clone(*this, destDir);
+    scheme->clone(settings, *this, destDir);
 }
 
 std::optional<std::filesystem::path> Input::getSourcePath() const
@@ -493,7 +484,7 @@ void InputScheme::putFile(
     throw Error("input '%s' does not support modifying file '%s'", input.to_string(), path);
 }
 
-void InputScheme::clone(const Input & input, const Path & destDir) const
+void InputScheme::clone(const Settings & settings, const Input & input, const Path & destDir) const
 {
     throw Error("do not know how to clone input '%s'", input.to_string());
 }

src/libfetchers/git-utils.cc

@@ -1328,13 +1328,18 @@ std::vector<std::tuple<GitRepoImpl::Submodule, Hash>> GitRepoImpl::getSubmodules
     return result;
 }
 
-ref<GitRepo> getTarballCache()
-{
-    static auto repoDir = std::filesystem::path(getCacheDir()) / "tarball-cache";
+namespace fetchers {
 
-    return GitRepo::openRepo(repoDir, true, true);
+ref<GitRepo> Settings::getTarballCache() const
+{
+    auto tarballCache(_tarballCache.lock());
+    if (!*tarballCache)
+        *tarballCache = GitRepo::openRepo(std::filesystem::path(getCacheDir()) / "tarball-cache", true, true);
+    return ref<GitRepo>(*tarballCache);
 }
 
+} // namespace fetchers
+
 GitRepo::WorkdirInfo GitRepo::getCachedWorkdirInfo(const std::filesystem::path & path)
 {
     static Sync<std::map<std::filesystem::path, WorkdirInfo>> _cache;

src/libfetchers/git.cc

@@ -168,8 +168,6 @@ struct GitInputScheme : InputScheme
             return {};
 
         auto url2(url);
-        if (hasPrefix(url2.scheme, "git+"))
-            url2.scheme = std::string(url2.scheme, 4);
         url2.query.clear();
 
         Attrs attrs;
@@ -196,28 +194,183 @@ struct GitInputScheme : InputScheme
         return "git";
     }
 
-    StringSet allowedAttrs() const override
+    std::string schemeDescription() const override
     {
-        return {
-            "url",
-            "ref",
-            "rev",
-            "shallow",
-            "submodules",
-            "lfs",
-            "exportIgnore",
-            "lastModified",
-            "revCount",
-            "narHash",
-            "allRefs",
-            "name",
-            "dirtyRev",
-            "dirtyShortRev",
-            "verifyCommit",
-            "keytype",
-            "publicKey",
-            "publicKeys",
+        return stripIndentation(R"(
+          Fetch a Git tree and copy it to the Nix store.
+          This is similar to [`builtins.fetchGit`](@docroot@/language/builtins.md#builtins-fetchGit).
+        )");
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "url",
+                {
+                    .type = "String",
+                    .required = true,
+                    .doc = R"(
+                      The URL formats supported are the same as for Git itself.
+
+                      > **Example**
+                      >
+                      > ```nix
+                      > fetchTree {
+                      >   type = "git";
+                      >   url = "git@github.com:NixOS/nixpkgs.git";
+                      > }
+                      > ```
+
+                      > **Note**
+                      >
+                      > If the URL points to a local directory, and no `ref` or `rev` is given, Nix only considers files added to the Git index, as listed by `git ls-files` but uses the *current file contents* of the Git working directory.
+                    )",
+                },
+            },
+            {
+                "ref",
+                {
+                    .type = "String",
+                    .required = false,
+                    .doc = R"(
+                      By default, this has no effect. This becomes relevant only once `shallow` cloning is disabled.
+
+                      A [Git reference](https://git-scm.com/book/en/v2/Git-Internals-Git-References), such as a branch or tag name.
+
+                      Default: `"HEAD"`
+                    )",
+                },
+            },
+            {
+                "rev",
+                {
+                    .type = "String",
+                    .required = false,
+                    .doc = R"(
+                      A Git revision; a commit hash.
+
+                      Default: the tip of `ref`
+                    )",
+                },
+            },
+            {
+                "shallow",
+                {
+                    .type = "Bool",
+                    .required = false,
+                    .doc = R"(
+                      Make a shallow clone when fetching the Git tree.
+                      When this is enabled, the options `ref` and `allRefs` have no effect anymore.
+
+                      Default: `true`
+                    )",
+                },
+            },
+            {
+                "submodules",
+                {
+                    .type = "Bool",
+                    .required = false,
+                    .doc = R"(
+                      Also fetch submodules if available.
+
+                      Default: `false`
+                    )",
+                },
+            },
+            {
+                "lfs",
+                {
+                    .type = "Bool",
+                    .required = false,
+                    .doc = R"(
+                      Fetch any [Git LFS](https://git-lfs.com/) files.
+
+                      Default: `false`
+                    )",
+                },
+            },
+            {
+                "exportIgnore",
+                {},
+            },
+            {
+                "lastModified",
+                {
+                    .type = "Integer",
+                    .required = false,
+                    .doc = R"(
+                      Unix timestamp of the fetched commit.
+
+                      If set, pass through the value to the output attribute set.
+                      Otherwise, generated from the fetched Git tree.
+                    )",
+                },
+            },
+            {
+                "revCount",
+                {
+                    .type = "Integer",
+                    .required = false,
+                    .doc = R"(
+                      Number of revisions in the history of the Git repository before the fetched commit.
+
+                      If set, pass through the value to the output attribute set.
+                      Otherwise, generated from the fetched Git tree.
+                    )",
+                },
+            },
+            {
+                "narHash",
+                {},
+            },
+            {
+                "allRefs",
+                {
+                    .type = "Bool",
+                    .required = false,
+                    .doc = R"(
+                      By default, this has no effect. This becomes relevant only once `shallow` cloning is disabled.
+
+                      Whether to fetch all references (eg. branches and tags) of the repository.
+                      With this argument being true, it's possible to load a `rev` from *any* `ref`.
+                      (Without setting this option, only `rev`s from the specified `ref` are supported).
+
+                      Default: `false`
+                    )",
+                },
+            },
+            {
+                "name",
+                {},
+            },
+            {
+                "dirtyRev",
+                {},
+            },
+            {
+                "dirtyShortRev",
+                {},
+            },
+            {
+                "verifyCommit",
+                {},
+            },
+            {
+                "keytype",
+                {},
+            },
+            {
+                "publicKey",
+                {},
+            },
+            {
+                "publicKeys",
+                {},
+            },
         };
+        return attrs;
     }
 
     std::optional<Input> inputFromAttrs(const Settings & settings, const Attrs & attrs) const override
@@ -231,7 +384,7 @@ struct GitInputScheme : InputScheme
         if (auto ref = maybeGetStrAttr(attrs, "ref"); ref && !isLegalRefName(*ref))
             throw BadURL("invalid Git branch/tag name '%s'", *ref);
 
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
         input.attrs["url"] = fixGitURL(getStrAttr(attrs, "url")).to_string();
         getShallowAttr(input);
@@ -280,7 +433,7 @@ struct GitInputScheme : InputScheme
         return res;
     }
 
-    void clone(const Input & input, const Path & destDir) const override
+    void clone(const Settings & settings, const Input & input, const Path & destDir) const override
     {
         auto repoInfo = getRepoInfo(input);
 
@@ -625,7 +778,7 @@ struct GitInputScheme : InputScheme
     }
 
     std::pair<ref<SourceAccessor>, Input>
-    getAccessorFromCommit(ref<Store> store, RepoInfo & repoInfo, Input && input) const
+    getAccessorFromCommit(const Settings & settings, ref<Store> store, RepoInfo & repoInfo, Input && input) const
     {
         assert(!repoInfo.workdirInfo.isDirty);
 
@@ -735,10 +888,10 @@ struct GitInputScheme : InputScheme
 
         auto rev = *input.getRev();
 
-        input.attrs.insert_or_assign("lastModified", getLastModified(*input.settings, repoInfo, repoDir, rev));
+        input.attrs.insert_or_assign("lastModified", getLastModified(settings, repoInfo, repoDir, rev));
 
         if (!getShallowAttr(input))
-            input.attrs.insert_or_assign("revCount", getRevCount(*input.settings, repoInfo, repoDir, rev));
+            input.attrs.insert_or_assign("revCount", getRevCount(settings, repoInfo, repoDir, rev));
 
         printTalkative("using revision %s of repo '%s'", rev.gitRev(), repoInfo.locationToArg());
 
@@ -781,8 +934,8 @@ struct GitInputScheme : InputScheme
                 attrs.insert_or_assign("submodules", Explicit<bool>{true});
                 attrs.insert_or_assign("lfs", Explicit<bool>{smudgeLfs});
                 attrs.insert_or_assign("allRefs", Explicit<bool>{true});
-                auto submoduleInput = fetchers::Input::fromAttrs(*input.settings, std::move(attrs));
-                auto [submoduleAccessor, submoduleInput2] = submoduleInput.getAccessor(store);
+                auto submoduleInput = fetchers::Input::fromAttrs(settings, std::move(attrs));
+                auto [submoduleAccessor, submoduleInput2] = submoduleInput.getAccessor(settings, store);
                 submoduleAccessor->setPathDisplay("«" + submoduleInput.to_string() + "»");
                 mounts.insert_or_assign(submodule.path, submoduleAccessor);
             }
@@ -799,7 +952,7 @@ struct GitInputScheme : InputScheme
     }
 
     std::pair<ref<SourceAccessor>, Input>
-    getAccessorFromWorkdir(ref<Store> store, RepoInfo & repoInfo, Input && input) const
+    getAccessorFromWorkdir(const Settings & settings, ref<Store> store, RepoInfo & repoInfo, Input && input) const
     {
         auto repoPath = repoInfo.getPath().value();
 
@@ -831,8 +984,8 @@ struct GitInputScheme : InputScheme
                 // TODO: fall back to getAccessorFromCommit-like fetch when submodules aren't checked out
                 // attrs.insert_or_assign("allRefs", Explicit<bool>{ true });
 
-                auto submoduleInput = fetchers::Input::fromAttrs(*input.settings, std::move(attrs));
-                auto [submoduleAccessor, submoduleInput2] = submoduleInput.getAccessor(store);
+                auto submoduleInput = fetchers::Input::fromAttrs(settings, std::move(attrs));
+                auto [submoduleAccessor, submoduleInput2] = submoduleInput.getAccessor(settings, store);
                 submoduleAccessor->setPathDisplay("«" + submoduleInput.to_string() + "»");
 
                 /* If the submodule is dirty, mark this repo dirty as
@@ -859,12 +1012,12 @@ struct GitInputScheme : InputScheme
             input.attrs.insert_or_assign("rev", rev.gitRev());
             if (!getShallowAttr(input)) {
                 input.attrs.insert_or_assign(
-                    "revCount", rev == nullRev ? 0 : getRevCount(*input.settings, repoInfo, repoPath, rev));
+                    "revCount", rev == nullRev ? 0 : getRevCount(settings, repoInfo, repoPath, rev));
             }
 
             verifyCommit(input, repo);
         } else {
-            repoInfo.warnDirty(*input.settings);
+            repoInfo.warnDirty(settings);
 
             if (repoInfo.workdirInfo.headRev) {
                 input.attrs.insert_or_assign("dirtyRev", repoInfo.workdirInfo.headRev->gitRev() + "-dirty");
@@ -876,14 +1029,14 @@ struct GitInputScheme : InputScheme
 
         input.attrs.insert_or_assign(
             "lastModified",
-            repoInfo.workdirInfo.headRev
-                ? getLastModified(*input.settings, repoInfo, repoPath, *repoInfo.workdirInfo.headRev)
-                : 0);
+            repoInfo.workdirInfo.headRev ? getLastModified(settings, repoInfo, repoPath, *repoInfo.workdirInfo.headRev)
+                                         : 0);
 
         return {accessor, std::move(input)};
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
         Input input(_input);
 
@@ -899,8 +1052,8 @@ struct GitInputScheme : InputScheme
         }
 
         auto [accessor, final] = input.getRef() || input.getRev() || !repoInfo.getPath()
-                                     ? getAccessorFromCommit(store, repoInfo, std::move(input))
-                                     : getAccessorFromWorkdir(store, repoInfo, std::move(input));
+                                     ? getAccessorFromCommit(settings, store, repoInfo, std::move(input))
+                                     : getAccessorFromWorkdir(settings, store, repoInfo, std::move(input));
 
         return {accessor, std::move(final)};
     }
@@ -936,7 +1089,7 @@ struct GitInputScheme : InputScheme
         }
     }
 
-    bool isLocked(const Input & input) const override
+    bool isLocked(const Settings & settings, const Input & input) const override
     {
         auto rev = input.getRev();
         return rev && rev != nullRev;

src/libfetchers/github.cc

@@ -92,7 +92,7 @@ struct GitArchiveInputScheme : InputScheme
         if (ref && rev)
             throw BadURL("URL '%s' contains both a commit hash and a branch/tag name %s %s", url, *ref, rev->gitRev());
 
-        Input input{settings};
+        Input input{};
         input.attrs.insert_or_assign("type", std::string{schemeName()});
         input.attrs.insert_or_assign("owner", path[0]);
         input.attrs.insert_or_assign("repo", path[1]);
@@ -110,18 +110,43 @@ struct GitArchiveInputScheme : InputScheme
         return input;
     }
 
-    StringSet allowedAttrs() const override
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
     {
-        return {
-            "owner",
-            "repo",
-            "ref",
-            "rev",
-            "narHash",
-            "lastModified",
-            "host",
-            "treeHash",
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "owner",
+                {},
+            },
+            {
+                "repo",
+                {},
+            },
+            {
+                "ref",
+                {},
+            },
+            {
+                "rev",
+                {},
+            },
+            {
+                "narHash",
+                {},
+            },
+            {
+                "lastModified",
+                {},
+            },
+            {
+                "host",
+                {},
+            },
+            {
+                "treeHash",
+                {},
+            },
         };
+        return attrs;
     }
 
     std::optional<Input> inputFromAttrs(const fetchers::Settings & settings, const Attrs & attrs) const override
@@ -129,7 +154,7 @@ struct GitArchiveInputScheme : InputScheme
         getStrAttr(attrs, "owner");
         getStrAttr(attrs, "repo");
 
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
         return input;
     }
@@ -233,9 +258,9 @@ struct GitArchiveInputScheme : InputScheme
         std::optional<Hash> treeHash;
     };
 
-    virtual RefInfo getRevFromRef(nix::ref<Store> store, const Input & input) const = 0;
+    virtual RefInfo getRevFromRef(const Settings & settings, nix::ref<Store> store, const Input & input) const = 0;
 
-    virtual DownloadUrl getDownloadUrl(const Input & input) const = 0;
+    virtual DownloadUrl getDownloadUrl(const Settings & settings, const Input & input) const = 0;
 
     struct TarballInfo
     {
@@ -243,7 +268,7 @@ struct GitArchiveInputScheme : InputScheme
         time_t lastModified;
     };
 
-    std::pair<Input, TarballInfo> downloadArchive(ref<Store> store, Input input) const
+    std::pair<Input, TarballInfo> downloadArchive(const Settings & settings, ref<Store> store, Input input) const
     {
         if (!maybeGetStrAttr(input.attrs, "ref"))
             input.attrs.insert_or_assign("ref", "HEAD");
@@ -252,7 +277,7 @@ struct GitArchiveInputScheme : InputScheme
 
         auto rev = input.getRev();
         if (!rev) {
-            auto refInfo = getRevFromRef(store, input);
+            auto refInfo = getRevFromRef(settings, store, input);
             rev = refInfo.rev;
             upstreamTreeHash = refInfo.treeHash;
             debug("HEAD revision for '%s' is %s", input.to_string(), refInfo.rev.gitRev());
@@ -261,7 +286,7 @@ struct GitArchiveInputScheme : InputScheme
         input.attrs.erase("ref");
         input.attrs.insert_or_assign("rev", rev->gitRev());
 
-        auto cache = input.settings->getCache();
+        auto cache = settings.getCache();
 
         Cache::Key treeHashKey{"gitRevToTreeHash", {{"rev", rev->gitRev()}}};
         Cache::Key lastModifiedKey{"gitRevToLastModified", {{"rev", rev->gitRev()}}};
@@ -270,7 +295,7 @@ struct GitArchiveInputScheme : InputScheme
             if (auto lastModifiedAttrs = cache->lookup(lastModifiedKey)) {
                 auto treeHash = getRevAttr(*treeHashAttrs, "treeHash");
                 auto lastModified = getIntAttr(*lastModifiedAttrs, "lastModified");
-                if (getTarballCache()->hasObject(treeHash))
+                if (settings.getTarballCache()->hasObject(treeHash))
                     return {std::move(input), TarballInfo{.treeHash = treeHash, .lastModified = (time_t) lastModified}};
                 else
                     debug("Git tree with hash '%s' has disappeared from the cache, refetching...", treeHash.gitRev());
@@ -278,7 +303,7 @@ struct GitArchiveInputScheme : InputScheme
         }
 
         /* Stream the tarball into the tarball cache. */
-        auto url = getDownloadUrl(input);
+        auto url = getDownloadUrl(settings, input);
 
         auto source = sinkToSource([&](Sink & sink) {
             FileTransferRequest req(url.url);
@@ -290,7 +315,7 @@ struct GitArchiveInputScheme : InputScheme
             *logger, lvlInfo, actUnknown, fmt("unpacking '%s' into the Git cache", input.to_string()));
 
         TarArchive archive{*source};
-        auto tarballCache = getTarballCache();
+        auto tarballCache = settings.getTarballCache();
         auto parseSink = tarballCache->getFileSystemObjectSink();
         auto lastModified = unpackTarfileToSink(archive, *parseSink);
         auto tree = parseSink->flush();
@@ -315,28 +340,29 @@ struct GitArchiveInputScheme : InputScheme
         return {std::move(input), tarballInfo};
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
-        auto [input, tarballInfo] = downloadArchive(store, _input);
+        auto [input, tarballInfo] = downloadArchive(settings, store, _input);
 
 #if 0
         input.attrs.insert_or_assign("treeHash", tarballInfo.treeHash.gitRev());
 #endif
         input.attrs.insert_or_assign("lastModified", uint64_t(tarballInfo.lastModified));
 
-        auto accessor = getTarballCache()->getAccessor(tarballInfo.treeHash, false, "«" + input.to_string() + "»");
+        auto accessor =
+            settings.getTarballCache()->getAccessor(tarballInfo.treeHash, false, "«" + input.to_string() + "»");
 
         return {accessor, input};
     }
 
-    bool isLocked(const Input & input) const override
+    bool isLocked(const Settings & settings, const Input & input) const override
     {
         /* Since we can't verify the integrity of the tarball from the
            Git revision alone, we also require a NAR hash for
            locking. FIXME: in the future, we may want to require a Git
            tree hash instead of a NAR hash. */
-        return input.getRev().has_value()
-               && (input.settings->trustTarballsFromGitForges || input.getNarHash().has_value());
+        return input.getRev().has_value() && (settings.trustTarballsFromGitForges || input.getNarHash().has_value());
     }
 
     std::optional<ExperimentalFeature> experimentalFeature() const override
@@ -360,6 +386,12 @@ struct GitHubInputScheme : GitArchiveInputScheme
         return "github";
     }
 
+    std::string schemeDescription() const override
+    {
+        // TODO
+        return "";
+    }
+
     std::optional<std::pair<std::string, std::string>> accessHeaderFromToken(const std::string & token) const override
     {
         // Github supports PAT/OAuth2 tokens and HTTP Basic
@@ -386,7 +418,7 @@ struct GitHubInputScheme : GitArchiveInputScheme
         return getStrAttr(input.attrs, "repo");
     }
 
-    RefInfo getRevFromRef(nix::ref<Store> store, const Input & input) const override
+    RefInfo getRevFromRef(const Settings & settings, nix::ref<Store> store, const Input & input) const override
     {
         auto host = getHost(input);
         auto url = fmt(
@@ -396,9 +428,9 @@ struct GitHubInputScheme : GitArchiveInputScheme
             getRepo(input),
             *input.getRef());
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
 
-        auto downloadResult = downloadFile(store, *input.settings, url, "source", headers);
+        auto downloadResult = downloadFile(store, settings, url, "source", headers);
         auto json = nlohmann::json::parse(
             store->requireStoreObjectAccessor(downloadResult.storePath)->readFile(CanonPath::root));
 
@@ -407,11 +439,11 @@ struct GitHubInputScheme : GitArchiveInputScheme
             .treeHash = Hash::parseAny(std::string{json["commit"]["tree"]["sha"]}, HashAlgorithm::SHA1)};
     }
 
-    DownloadUrl getDownloadUrl(const Input & input) const override
+    DownloadUrl getDownloadUrl(const Settings & settings, const Input & input) const override
     {
         auto host = getHost(input);
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
 
         // If we have no auth headers then we default to the public archive
         // urls so we do not run into rate limits.
@@ -425,12 +457,12 @@ struct GitHubInputScheme : GitArchiveInputScheme
         return DownloadUrl{parseURL(url), headers};
     }
 
-    void clone(const Input & input, const Path & destDir) const override
+    void clone(const Settings & settings, const Input & input, const Path & destDir) const override
     {
         auto host = getHost(input);
-        Input::fromURL(*input.settings, fmt("git+https://%s/%s/%s.git", host, getOwner(input), getRepo(input)))
+        Input::fromURL(settings, fmt("git+https://%s/%s/%s.git", host, getOwner(input), getRepo(input)))
             .applyOverrides(input.getRef(), input.getRev())
-            .clone(destDir);
+            .clone(settings, destDir);
     }
 };
 
@@ -441,6 +473,12 @@ struct GitLabInputScheme : GitArchiveInputScheme
         return "gitlab";
     }
 
+    std::string schemeDescription() const override
+    {
+        // TODO
+        return "";
+    }
+
     std::optional<std::pair<std::string, std::string>> accessHeaderFromToken(const std::string & token) const override
     {
         // Gitlab supports 4 kinds of authorization, two of which are
@@ -460,7 +498,7 @@ struct GitLabInputScheme : GitArchiveInputScheme
         return std::make_pair(token.substr(0, fldsplit), token.substr(fldsplit + 1));
     }
 
-    RefInfo getRevFromRef(nix::ref<Store> store, const Input & input) const override
+    RefInfo getRevFromRef(const Settings & settings, nix::ref<Store> store, const Input & input) const override
     {
         auto host = maybeGetStrAttr(input.attrs, "host").value_or("gitlab.com");
         // See rate limiting note below
@@ -471,9 +509,9 @@ struct GitLabInputScheme : GitArchiveInputScheme
                 getStrAttr(input.attrs, "repo"),
                 *input.getRef());
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
 
-        auto downloadResult = downloadFile(store, *input.settings, url, "source", headers);
+        auto downloadResult = downloadFile(store, settings, url, "source", headers);
         auto json = nlohmann::json::parse(
             store->requireStoreObjectAccessor(downloadResult.storePath)->readFile(CanonPath::root));
 
@@ -487,7 +525,7 @@ struct GitLabInputScheme : GitArchiveInputScheme
         }
     }
 
-    DownloadUrl getDownloadUrl(const Input & input) const override
+    DownloadUrl getDownloadUrl(const Settings & settings, const Input & input) const override
     {
         // This endpoint has a rate limit threshold that may be
         // server-specific and vary based whether the user is
@@ -502,19 +540,19 @@ struct GitLabInputScheme : GitArchiveInputScheme
                 getStrAttr(input.attrs, "repo"),
                 input.getRev()->to_string(HashFormat::Base16, false));
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
         return DownloadUrl{parseURL(url), headers};
     }
 
-    void clone(const Input & input, const Path & destDir) const override
+    void clone(const Settings & settings, const Input & input, const Path & destDir) const override
     {
         auto host = maybeGetStrAttr(input.attrs, "host").value_or("gitlab.com");
         // FIXME: get username somewhere
         Input::fromURL(
-            *input.settings,
+            settings,
             fmt("git+https://%s/%s/%s.git", host, getStrAttr(input.attrs, "owner"), getStrAttr(input.attrs, "repo")))
             .applyOverrides(input.getRef(), input.getRev())
-            .clone(destDir);
+            .clone(settings, destDir);
     }
 };
 
@@ -525,6 +563,12 @@ struct SourceHutInputScheme : GitArchiveInputScheme
         return "sourcehut";
     }
 
+    std::string schemeDescription() const override
+    {
+        // TODO
+        return "";
+    }
+
     std::optional<std::pair<std::string, std::string>> accessHeaderFromToken(const std::string & token) const override
     {
         // SourceHut supports both PAT and OAuth2. See
@@ -535,7 +579,7 @@ struct SourceHutInputScheme : GitArchiveInputScheme
         // Once it is implemented, however, should work as expected.
     }
 
-    RefInfo getRevFromRef(nix::ref<Store> store, const Input & input) const override
+    RefInfo getRevFromRef(const Settings & settings, nix::ref<Store> store, const Input & input) const override
     {
         // TODO: In the future, when the sourcehut graphql API is implemented for mercurial
         // and with anonymous access, this method should use it instead.
@@ -546,11 +590,11 @@ struct SourceHutInputScheme : GitArchiveInputScheme
         auto base_url =
             fmt("https://%s/%s/%s", host, getStrAttr(input.attrs, "owner"), getStrAttr(input.attrs, "repo"));
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
 
         std::string refUri;
         if (ref == "HEAD") {
-            auto downloadFileResult = downloadFile(store, *input.settings, fmt("%s/HEAD", base_url), "source", headers);
+            auto downloadFileResult = downloadFile(store, settings, fmt("%s/HEAD", base_url), "source", headers);
             auto contents = store->requireStoreObjectAccessor(downloadFileResult.storePath)->readFile(CanonPath::root);
 
             auto remoteLine = git::parseLsRemoteLine(getLine(contents).first);
@@ -563,8 +607,7 @@ struct SourceHutInputScheme : GitArchiveInputScheme
         }
         std::regex refRegex(refUri);
 
-        auto downloadFileResult =
-            downloadFile(store, *input.settings, fmt("%s/info/refs", base_url), "source", headers);
+        auto downloadFileResult = downloadFile(store, settings, fmt("%s/info/refs", base_url), "source", headers);
         auto contents = store->requireStoreObjectAccessor(downloadFileResult.storePath)->readFile(CanonPath::root);
         std::istringstream is(contents);
 
@@ -582,7 +625,7 @@ struct SourceHutInputScheme : GitArchiveInputScheme
         return RefInfo{.rev = Hash::parseAny(*id, HashAlgorithm::SHA1)};
     }
 
-    DownloadUrl getDownloadUrl(const Input & input) const override
+    DownloadUrl getDownloadUrl(const Settings & settings, const Input & input) const override
     {
         auto host = maybeGetStrAttr(input.attrs, "host").value_or("git.sr.ht");
         auto url =
@@ -592,18 +635,18 @@ struct SourceHutInputScheme : GitArchiveInputScheme
                 getStrAttr(input.attrs, "repo"),
                 input.getRev()->to_string(HashFormat::Base16, false));
 
-        Headers headers = makeHeadersWithAuthTokens(*input.settings, host, input);
+        Headers headers = makeHeadersWithAuthTokens(settings, host, input);
         return DownloadUrl{parseURL(url), headers};
     }
 
-    void clone(const Input & input, const Path & destDir) const override
+    void clone(const Settings & settings, const Input & input, const Path & destDir) const override
     {
         auto host = maybeGetStrAttr(input.attrs, "host").value_or("git.sr.ht");
         Input::fromURL(
-            *input.settings,
+            settings,
             fmt("git+https://%s/%s/%s", host, getStrAttr(input.attrs, "owner"), getStrAttr(input.attrs, "repo")))
             .applyOverrides(input.getRef(), input.getRev())
-            .clone(destDir);
+            .clone(settings, destDir);
     }
 };
 

src/libfetchers/include/nix/fetchers/fetch-settings.hh

@@ -11,6 +11,12 @@
 
 #include <sys/types.h>
 
+namespace nix {
+
+struct GitRepo;
+
+}
+
 namespace nix::fetchers {
 
 struct Cache;
@@ -125,8 +131,12 @@ struct Settings : public Config
 
     ref<Cache> getCache() const;
 
+    ref<GitRepo> getTarballCache() const;
+
 private:
     mutable Sync<std::shared_ptr<Cache>> _cache;
+
+    mutable Sync<std::shared_ptr<GitRepo>> _tarballCache;
 };
 
 } // namespace nix::fetchers

src/libfetchers/include/nix/fetchers/fetchers.hh

@@ -36,13 +36,6 @@ struct Input
 {
     friend struct InputScheme;
 
-    const Settings * settings;
-
-    Input(const Settings & settings)
-        : settings{&settings}
-    {
-    }
-
     std::shared_ptr<InputScheme> scheme; // note: can be null
     Attrs attrs;
 
@@ -87,7 +80,7 @@ public:
      * attributes like a Git revision or NAR hash that uniquely
      * identify its contents.
      */
-    bool isLocked() const;
+    bool isLocked(const Settings & settings) const;
 
     /**
      * Only for relative path flakes, i.e. 'path:./foo', returns the
@@ -120,7 +113,7 @@ public:
      * Fetch the entire input into the Nix store, returning the
      * location in the Nix store and the locked input.
      */
-    std::pair<StorePath, Input> fetchToStore(ref<Store> store) const;
+    std::pair<StorePath, Input> fetchToStore(const Settings & settings, ref<Store> store) const;
 
     /**
      * Check the locking attributes in `result` against
@@ -140,17 +133,17 @@ public:
      * input without copying it to the store. Also return a possibly
      * unlocked input.
      */
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store) const;
+    std::pair<ref<SourceAccessor>, Input> getAccessor(const Settings & settings, ref<Store> store) const;
 
 private:
 
-    std::pair<ref<SourceAccessor>, Input> getAccessorUnchecked(ref<Store> store) const;
+    std::pair<ref<SourceAccessor>, Input> getAccessorUnchecked(const Settings & settings, ref<Store> store) const;
 
 public:
 
     Input applyOverrides(std::optional<std::string> ref, std::optional<Hash> rev) const;
 
-    void clone(const Path & destDir) const;
+    void clone(const Settings & settings, const Path & destDir) const;
 
     std::optional<std::filesystem::path> getSourcePath() const;
 
@@ -210,20 +203,33 @@ struct InputScheme
      */
     virtual std::string_view schemeName() const = 0;
 
+    /**
+     * Longform description of this scheme, for documentation purposes.
+     */
+    virtual std::string schemeDescription() const = 0;
+
+    // TODO remove these defaults
+    struct AttributeInfo
+    {
+        const char * type = "String";
+        bool required = true;
+        const char * doc = "";
+    };
+
     /**
      * Allowed attributes in an attribute set that is converted to an
-     * input.
+     * input, and documentation for each attribute.
      *
-     * `type` is not included from this set, because the `type` field is
+     * `type` is not included from this map, because the `type` field is
       parsed first to choose which scheme; `type` is always required.
      */
-    virtual StringSet allowedAttrs() const = 0;
+    virtual const std::map<std::string, AttributeInfo> & allowedAttrs() const = 0;
 
     virtual ParsedURL toURL(const Input & input) const;
 
     virtual Input applyOverrides(const Input & input, std::optional<std::string> ref, std::optional<Hash> rev) const;
 
-    virtual void clone(const Input & input, const Path & destDir) const;
+    virtual void clone(const Settings & settings, const Input & input, const Path & destDir) const;
 
     virtual std::optional<std::filesystem::path> getSourcePath(const Input & input) const;
 
@@ -233,7 +239,8 @@ struct InputScheme
         std::string_view contents,
         std::optional<std::string> commitMsg) const;
 
-    virtual std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & input) const = 0;
+    virtual std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & input) const = 0;
 
     /**
      * Is this `InputScheme` part of an experimental feature?
@@ -250,7 +257,7 @@ struct InputScheme
         return std::nullopt;
     }
 
-    virtual bool isLocked(const Input & input) const
+    virtual bool isLocked(const Settings & settings, const Input & input) const
     {
         return false;
     }
@@ -269,7 +276,12 @@ struct InputScheme
 
 void registerInputScheme(std::shared_ptr<InputScheme> && fetcher);
 
-nlohmann::json dumpRegisterInputSchemeInfo();
+using InputSchemeMap = std::map<std::string_view, std::shared_ptr<InputScheme>>;
+
+/**
+ * Use this for docs, not for finding a specific scheme
+ */
+const InputSchemeMap & getAllInputSchemes();
 
 struct PublicKey
 {

src/libfetchers/include/nix/fetchers/git-utils.hh

@@ -120,8 +120,6 @@ struct GitRepo
     virtual Hash dereferenceSingletonDirectory(const Hash & oid) = 0;
 };
 
-ref<GitRepo> getTarballCache();
-
 // A helper to ensure that the `git_*_free` functions get called.
 template<auto del>
 struct Deleter

src/libfetchers/include/nix/fetchers/input-cache.hh

@@ -3,6 +3,7 @@
 namespace nix::fetchers {
 
 enum class UseRegistries : int;
+struct Settings;
 
 struct InputCache
 {
@@ -14,7 +15,8 @@ struct InputCache
         Attrs extraAttrs;
     };
 
-    CachedResult getAccessor(ref<Store> store, const Input & originalInput, UseRegistries useRegistries);
+    CachedResult
+    getAccessor(const Settings & settings, ref<Store> store, const Input & originalInput, UseRegistries useRegistries);
 
     struct CachedInput
     {

src/libfetchers/include/nix/fetchers/registry.hh

@@ -59,7 +59,7 @@ Path getUserRegistryPath();
 
 Registries getRegistries(const Settings & settings, ref<Store> store);
 
-void overrideRegistry(const Input & from, const Input & to, const Attrs & extraAttrs);
+void overrideRegistry(const Settings & settings, const Input & from, const Input & to, const Attrs & extraAttrs);
 
 enum class UseRegistries : int {
     No,
@@ -71,6 +71,7 @@ enum class UseRegistries : int {
  * Rewrite a flakeref using the registries. If `filter` is set, only
  * use the registries for which the filter function returns true.
  */
-std::pair<Input, Attrs> lookupInRegistries(ref<Store> store, const Input & input, UseRegistries useRegistries);
+std::pair<Input, Attrs>
+lookupInRegistries(const Settings & settings, ref<Store> store, const Input & input, UseRegistries useRegistries);
 
 } // namespace nix::fetchers

src/libfetchers/indirect.cc

@@ -44,7 +44,7 @@ struct IndirectInputScheme : InputScheme
 
         // FIXME: forbid query params?
 
-        Input input{settings};
+        Input input{};
         input.attrs.insert_or_assign("type", "indirect");
         input.attrs.insert_or_assign("id", id);
         if (rev)
@@ -60,14 +60,33 @@ struct IndirectInputScheme : InputScheme
         return "indirect";
     }
 
-    StringSet allowedAttrs() const override
+    std::string schemeDescription() const override
     {
-        return {
-            "id",
-            "ref",
-            "rev",
-            "narHash",
+        // TODO
+        return "";
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "id",
+                {},
+            },
+            {
+                "ref",
+                {},
+            },
+            {
+                "rev",
+                {},
+            },
+            {
+                "narHash",
+                {},
+            },
         };
+        return attrs;
     }
 
     std::optional<Input> inputFromAttrs(const Settings & settings, const Attrs & attrs) const override
@@ -76,7 +95,7 @@ struct IndirectInputScheme : InputScheme
         if (!std::regex_match(id, flakeRegex))
             throw BadURL("'%s' is not a valid flake ID", id);
 
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
         return input;
     }
@@ -106,7 +125,8 @@ struct IndirectInputScheme : InputScheme
         return input;
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & input) const override
     {
         throw Error("indirect input '%s' cannot be fetched directly", input.to_string());
     }

src/libfetchers/input-cache.cc

@@ -5,23 +5,23 @@
 
 namespace nix::fetchers {
 
-InputCache::CachedResult
-InputCache::getAccessor(ref<Store> store, const Input & originalInput, UseRegistries useRegistries)
+InputCache::CachedResult InputCache::getAccessor(
+    const Settings & settings, ref<Store> store, const Input & originalInput, UseRegistries useRegistries)
 {
     auto fetched = lookup(originalInput);
     Input resolvedInput = originalInput;
 
     if (!fetched) {
         if (originalInput.isDirect()) {
-            auto [accessor, lockedInput] = originalInput.getAccessor(store);
+            auto [accessor, lockedInput] = originalInput.getAccessor(settings, store);
             fetched.emplace(CachedInput{.lockedInput = lockedInput, .accessor = accessor});
         } else {
             if (useRegistries != UseRegistries::No) {
-                auto [res, extraAttrs] = lookupInRegistries(store, originalInput, useRegistries);
+                auto [res, extraAttrs] = lookupInRegistries(settings, store, originalInput, useRegistries);
                 resolvedInput = std::move(res);
                 fetched = lookup(resolvedInput);
                 if (!fetched) {
-                    auto [accessor, lockedInput] = resolvedInput.getAccessor(store);
+                    auto [accessor, lockedInput] = resolvedInput.getAccessor(settings, store);
                     fetched.emplace(
                         CachedInput{.lockedInput = lockedInput, .accessor = accessor, .extraAttrs = extraAttrs});
                 }

src/libfetchers/mercurial.cc

@@ -68,16 +68,41 @@ struct MercurialInputScheme : InputScheme
         return "hg";
     }
 
-    StringSet allowedAttrs() const override
+    std::string schemeDescription() const override
     {
-        return {
-            "url",
-            "ref",
-            "rev",
-            "revCount",
-            "narHash",
-            "name",
+        // TODO
+        return "";
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "url",
+                {},
+            },
+            {
+                "ref",
+                {},
+            },
+            {
+                "rev",
+                {},
+            },
+            {
+                "revCount",
+                {},
+            },
+            {
+                "narHash",
+                {},
+            },
+            {
+                "name",
+                {},
+            },
         };
+        return attrs;
     }
 
     std::optional<Input> inputFromAttrs(const Settings & settings, const Attrs & attrs) const override
@@ -89,7 +114,7 @@ struct MercurialInputScheme : InputScheme
                 throw BadURL("invalid Mercurial branch/tag name '%s'", *ref);
         }
 
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
         return input;
     }
@@ -154,7 +179,7 @@ struct MercurialInputScheme : InputScheme
         return {isLocal, isLocal ? renderUrlPathEnsureLegal(url.path) : url.to_string()};
     }
 
-    StorePath fetchToStore(ref<Store> store, Input & input) const
+    StorePath fetchToStore(const Settings & settings, ref<Store> store, Input & input) const
     {
         auto origRev = input.getRev();
 
@@ -176,10 +201,10 @@ struct MercurialInputScheme : InputScheme
                 /* This is an unclean working tree. So copy all tracked
                    files. */
 
-                if (!input.settings->allowDirty)
+                if (!settings.allowDirty)
                     throw Error("Mercurial tree '%s' is unclean", actualUrl);
 
-                if (input.settings->warnDirty)
+                if (settings.warnDirty)
                     warn("Mercurial tree '%s' is unclean", actualUrl);
 
                 input.attrs.insert_or_assign("ref", chomp(runHg({"branch", "-R", actualUrl})));
@@ -240,13 +265,13 @@ struct MercurialInputScheme : InputScheme
         Cache::Key refToRevKey{"hgRefToRev", {{"url", actualUrl}, {"ref", *input.getRef()}}};
 
         if (!input.getRev()) {
-            if (auto res = input.settings->getCache()->lookupWithTTL(refToRevKey))
+            if (auto res = settings.getCache()->lookupWithTTL(refToRevKey))
                 input.attrs.insert_or_assign("rev", getRevAttr(*res, "rev").gitRev());
         }
 
         /* If we have a rev, check if we have a cached store path. */
         if (auto rev = input.getRev()) {
-            if (auto res = input.settings->getCache()->lookupStorePath(revInfoKey(*rev), *store))
+            if (auto res = settings.getCache()->lookupStorePath(revInfoKey(*rev), *store))
                 return makeResult(res->value, res->storePath);
         }
 
@@ -300,7 +325,7 @@ struct MercurialInputScheme : InputScheme
 
         /* Now that we have the rev, check the cache again for a
            cached store path. */
-        if (auto res = input.settings->getCache()->lookupStorePath(revInfoKey(rev), *store))
+        if (auto res = settings.getCache()->lookupStorePath(revInfoKey(rev), *store))
             return makeResult(res->value, res->storePath);
 
         Path tmpDir = createTempDir();
@@ -317,18 +342,19 @@ struct MercurialInputScheme : InputScheme
         });
 
         if (!origRev)
-            input.settings->getCache()->upsert(refToRevKey, {{"rev", rev.gitRev()}});
+            settings.getCache()->upsert(refToRevKey, {{"rev", rev.gitRev()}});
 
-        input.settings->getCache()->upsert(revInfoKey(rev), *store, infoAttrs, storePath);
+        settings.getCache()->upsert(revInfoKey(rev), *store, infoAttrs, storePath);
 
         return makeResult(infoAttrs, std::move(storePath));
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
         Input input(_input);
 
-        auto storePath = fetchToStore(store, input);
+        auto storePath = fetchToStore(settings, store, input);
         auto accessor = store->requireStoreObjectAccessor(storePath);
 
         accessor->setPathDisplay("«" + input.to_string() + "»");
@@ -336,7 +362,7 @@ struct MercurialInputScheme : InputScheme
         return {accessor, input};
     }
 
-    bool isLocked(const Input & input) const override
+    bool isLocked(const Settings & settings, const Input & input) const override
     {
         return (bool) input.getRev();
     }

src/libfetchers/path.cc

@@ -17,7 +17,7 @@ struct PathInputScheme : InputScheme
         if (url.authority && url.authority->host.size())
             throw Error("path URL '%s' should not have an authority ('%s')", url, *url.authority);
 
-        Input input{settings};
+        Input input{};
         input.attrs.insert_or_assign("type", "path");
         input.attrs.insert_or_assign("path", renderUrlPathEnsureLegal(url.path));
 
@@ -40,27 +40,49 @@ struct PathInputScheme : InputScheme
         return "path";
     }
 
-    StringSet allowedAttrs() const override
+    std::string schemeDescription() const override
     {
-        return {
-            "path",
+        // TODO
+        return "";
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "path",
+                {},
+            },
             /* Allow the user to pass in "fake" tree info
                attributes. This is useful for making a pinned tree work
                the same as the repository from which is exported (e.g.
                path:/nix/store/...-source?lastModified=1585388205&rev=b0c285...).
              */
-            "rev",
-            "revCount",
-            "lastModified",
-            "narHash",
+            {
+                "rev",
+                {},
+            },
+            {
+                "revCount",
+                {},
+            },
+            {
+                "lastModified",
+                {},
+            },
+            {
+                "narHash",
+                {},
+            },
         };
+        return attrs;
     }
 
     std::optional<Input> inputFromAttrs(const Settings & settings, const Attrs & attrs) const override
     {
         getStrAttr(attrs, "path");
 
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
         return input;
     }
@@ -101,7 +123,7 @@ struct PathInputScheme : InputScheme
             return path;
     }
 
-    bool isLocked(const Input & input) const override
+    bool isLocked(const Settings & settings, const Input & input) const override
     {
         return (bool) input.getNarHash();
     }
@@ -116,7 +138,8 @@ struct PathInputScheme : InputScheme
         throw Error("cannot fetch input '%s' because it uses a relative path", input.to_string());
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
         Input input(_input);
         auto path = getStrAttr(input.attrs, "path");
@@ -145,7 +168,7 @@ struct PathInputScheme : InputScheme
         auto info = store->queryPathInfo(*storePath);
         accessor->fingerprint =
             fmt("path:%s", store->queryPathInfo(*storePath)->narHash.to_string(HashFormat::SRI, true));
-        input.settings->getCache()->upsert(
+        settings.getCache()->upsert(
             makeFetchToStoreCacheKey(
                 input.getName(), *accessor->fingerprint, ContentAddressMethod::Raw::NixArchive, "/"),
             *store,

src/libfetchers/registry.cc

@@ -131,9 +131,9 @@ std::shared_ptr<Registry> getFlagRegistry(const Settings & settings)
     return flagRegistry;
 }
 
-void overrideRegistry(const Input & from, const Input & to, const Attrs & extraAttrs)
+void overrideRegistry(const Settings & settings, const Input & from, const Input & to, const Attrs & extraAttrs)
 {
-    getFlagRegistry(*from.settings)->add(from, to, extraAttrs);
+    getFlagRegistry(settings)->add(from, to, extraAttrs);
 }
 
 static std::shared_ptr<Registry> getGlobalRegistry(const Settings & settings, ref<Store> store)
@@ -172,7 +172,8 @@ Registries getRegistries(const Settings & settings, ref<Store> store)
     return registries;
 }
 
-std::pair<Input, Attrs> lookupInRegistries(ref<Store> store, const Input & _input, UseRegistries useRegistries)
+std::pair<Input, Attrs>
+lookupInRegistries(const Settings & settings, ref<Store> store, const Input & _input, UseRegistries useRegistries)
 {
     Attrs extraAttrs;
     int n = 0;
@@ -187,7 +188,7 @@ restart:
     if (n > 100)
         throw Error("cycle detected in flake registry for '%s'", input.to_string());
 
-    for (auto & registry : getRegistries(*input.settings, store)) {
+    for (auto & registry : getRegistries(settings, store)) {
         if (useRegistries == UseRegistries::Limited
             && !(registry->type == fetchers::Registry::Flag || registry->type == fetchers::Registry::Global))
             continue;

src/libfetchers/tarball.cc

@@ -136,11 +136,11 @@ static DownloadTarballResult downloadTarball_(
             .treeHash = treeHash,
             .lastModified = (time_t) getIntAttr(infoAttrs, "lastModified"),
             .immutableUrl = maybeGetStrAttr(infoAttrs, "immutableUrl"),
-            .accessor = getTarballCache()->getAccessor(treeHash, false, displayPrefix),
+            .accessor = settings.getTarballCache()->getAccessor(treeHash, false, displayPrefix),
         };
     };
 
-    if (cached && !getTarballCache()->hasObject(getRevAttr(cached->value, "treeHash")))
+    if (cached && !settings.getTarballCache()->hasObject(getRevAttr(cached->value, "treeHash")))
         cached.reset();
 
     if (cached && !cached->expired)
@@ -179,7 +179,7 @@ static DownloadTarballResult downloadTarball_(
         TarArchive{path};
     })
                                                                                     : TarArchive{*source};
-    auto tarballCache = getTarballCache();
+    auto tarballCache = settings.getTarballCache();
     auto parseSink = tarballCache->getFileSystemObjectSink();
     auto lastModified = unpackTarfileToSink(archive, *parseSink);
     auto tree = parseSink->flush();
@@ -224,7 +224,7 @@ ref<SourceAccessor> downloadTarball(ref<Store> store, const Settings & settings,
 
     auto input = Input::fromAttrs(settings, std::move(attrs));
 
-    return input.getAccessor(store).first;
+    return input.getAccessor(settings, store).first;
 }
 
 // An input scheme corresponding to a curl-downloadable resource.
@@ -252,7 +252,7 @@ struct CurlInputScheme : InputScheme
         if (!isValidURL(_url, requireTree))
             return std::nullopt;
 
-        Input input{settings};
+        Input input{};
 
         auto url = _url;
 
@@ -278,7 +278,7 @@ struct CurlInputScheme : InputScheme
            HTTP request. Now that we've processed the Nix-specific
            attributes above, remove them so we don't also send them as
            part of the HTTP request. */
-        for (auto & param : allowedAttrs())
+        for (auto & [param, _] : allowedAttrs())
             url.query.erase(param);
 
         input.attrs.insert_or_assign("type", std::string{schemeName()});
@@ -286,23 +286,88 @@ struct CurlInputScheme : InputScheme
         return input;
     }
 
-    StringSet allowedAttrs() const override
+    static const std::map<std::string, AttributeInfo> & allowedAttrsImpl()
     {
-        return {
-            "type",
-            "url",
-            "narHash",
-            "name",
-            "unpack",
-            "rev",
-            "revCount",
-            "lastModified",
+        static const std::map<std::string, AttributeInfo> attrs = {
+            {
+                "url",
+                {
+                    .type = "String",
+                    .required = true,
+                    .doc = R"(
+                      Supported protocols:
+
+                      - `https`
+
+                        > **Example**
+                        >
+                        > ```nix
+                        > fetchTree {
+                        >   type = "file";
+                        >   url = "https://example.com/index.html";
+                        > }
+                        > ```
+
+                      - `http`
+
+                        Insecure HTTP transfer for legacy sources.
+
+                        > **Warning**
+                        >
+                        > HTTP performs no encryption or authentication.
+                        > Use a `narHash` known in advance to ensure the output has expected contents.
+
+                      - `file`
+
+                        A file on the local file system.
+
+                        > **Example**
+                        >
+                        > ```nix
+                        > fetchTree {
+                        >   type = "file";
+                        >   url = "file:///home/eelco/nix/README.md";
+                        > }
+                        > ```
+                    )",
+                },
+            },
+            {
+                "narHash",
+                {},
+            },
+            {
+                "name",
+                {},
+            },
+            {
+                "unpack",
+                {},
+            },
+            {
+                "rev",
+                {},
+            },
+            {
+                "revCount",
+                {},
+            },
+            {
+                "lastModified",
+                {},
+            },
         };
+        return attrs;
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        return allowedAttrsImpl();
     }
 
     std::optional<Input> inputFromAttrs(const Settings & settings, const Attrs & attrs) const override
     {
-        Input input{settings};
+        Input input{};
         input.attrs = attrs;
 
         // input.locked = (bool) maybeGetStrAttr(input.attrs, "hash");
@@ -319,7 +384,7 @@ struct CurlInputScheme : InputScheme
         return url;
     }
 
-    bool isLocked(const Input & input) const override
+    bool isLocked(const Settings & settings, const Input & input) const override
     {
         return (bool) input.getNarHash();
     }
@@ -332,6 +397,14 @@ struct FileInputScheme : CurlInputScheme
         return "file";
     }
 
+    std::string schemeDescription() const override
+    {
+        return stripIndentation(R"(
+          Place a plain file into the Nix store.
+          This is similar to [`builtins.fetchurl`](@docroot@/language/builtins.md#builtins-fetchurl)
+        )");
+    }
+
     bool isValidURL(const ParsedURL & url, bool requireTree) const override
     {
         auto parsedUrlScheme = parseUrlScheme(url.scheme);
@@ -340,7 +413,8 @@ struct FileInputScheme : CurlInputScheme
                                                : (!requireTree && !hasTarballExtension(url)));
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
         auto input(_input);
 
@@ -348,7 +422,7 @@ struct FileInputScheme : CurlInputScheme
            the Nix store directly, since there is little deduplication
            benefit in using the Git cache for single big files like
            tarballs. */
-        auto file = downloadFile(store, *input.settings, getStrAttr(input.attrs, "url"), input.getName());
+        auto file = downloadFile(store, settings, getStrAttr(input.attrs, "url"), input.getName());
 
         auto narHash = store->queryPathInfo(file.storePath)->narHash;
         input.attrs.insert_or_assign("narHash", narHash.to_string(HashFormat::SRI, true));
@@ -368,6 +442,34 @@ struct TarballInputScheme : CurlInputScheme
         return "tarball";
     }
 
+    std::string schemeDescription() const override
+    {
+        return stripIndentation(R"(
+          Download a tar archive and extract it into the Nix store.
+          This has the same underlying implementation as [`builtins.fetchTarball`](@docroot@/language/builtins.md#builtins-fetchTarball)
+        )");
+    }
+
+    const std::map<std::string, AttributeInfo> & allowedAttrs() const override
+    {
+        static const std::map<std::string, AttributeInfo> attrs = [] {
+            auto attrs = CurlInputScheme::allowedAttrsImpl();
+            // Override the "url" attribute to add tarball-specific example
+            attrs["url"].doc = R"(
+              > **Example**
+              >
+              > ```nix
+              > fetchTree {
+              >   type = "tarball";
+              >   url = "https://github.com/NixOS/nixpkgs/tarball/nixpkgs-23.11";
+              > }
+              > ```
+            )";
+            return attrs;
+        }();
+        return attrs;
+    }
+
     bool isValidURL(const ParsedURL & url, bool requireTree) const override
     {
         auto parsedUrlScheme = parseUrlScheme(url.scheme);
@@ -377,15 +479,15 @@ struct TarballInputScheme : CurlInputScheme
                                                : (requireTree || hasTarballExtension(url)));
     }
 
-    std::pair<ref<SourceAccessor>, Input> getAccessor(ref<Store> store, const Input & _input) const override
+    std::pair<ref<SourceAccessor>, Input>
+    getAccessor(const Settings & settings, ref<Store> store, const Input & _input) const override
     {
         auto input(_input);
 
-        auto result =
-            downloadTarball_(*input.settings, getStrAttr(input.attrs, "url"), {}, "«" + input.to_string() + "»");
+        auto result = downloadTarball_(settings, getStrAttr(input.attrs, "url"), {}, "«" + input.to_string() + "»");
 
         if (result.immutableUrl) {
-            auto immutableInput = Input::fromURL(*input.settings, *result.immutableUrl);
+            auto immutableInput = Input::fromURL(settings, *result.immutableUrl);
             // FIXME: would be nice to support arbitrary flakerefs
             // here, e.g. git flakes.
             if (immutableInput.getType() != "tarball")
@@ -398,7 +500,7 @@ struct TarballInputScheme : CurlInputScheme
 
         input.attrs.insert_or_assign(
             "narHash",
-            getTarballCache()->treeHashToNarHash(*input.settings, result.treeHash).to_string(HashFormat::SRI, true));
+            settings.getTarballCache()->treeHashToNarHash(settings, result.treeHash).to_string(HashFormat::SRI, true));
 
         return {result.accessor, input};
     }

src/libflake/flake-primops.cc

@@ -38,7 +38,7 @@ PrimOp getFlake(const Settings & settings)
         std::string flakeRefS(
             state.forceStringNoCtx(*args[0], pos, "while evaluating the argument passed to builtins.getFlake"));
         auto flakeRef = nix::parseFlakeRef(state.fetchSettings, flakeRefS, {}, true);
-        if (state.settings.pureEval && !flakeRef.input.isLocked())
+        if (state.settings.pureEval && !flakeRef.input.isLocked(state.fetchSettings))
             throw Error(
                 "cannot call 'getFlake' on unlocked flake reference '%s', at %s (use --impure to override)",
                 flakeRefS,

src/libflake/flake.cc

@@ -372,7 +372,8 @@ static Flake getFlake(
     const InputAttrPath & lockRootAttrPath)
 {
     // Fetch a lazy tree first.
-    auto cachedInput = state.inputCache->getAccessor(state.store, originalRef.input, useRegistries);
+    auto cachedInput =
+        state.inputCache->getAccessor(state.fetchSettings, state.store, originalRef.input, useRegistries);
 
     auto subdir = fetchers::maybeGetStrAttr(cachedInput.extraAttrs, "dir").value_or(originalRef.subdir);
     auto resolvedRef = FlakeRef(std::move(cachedInput.resolvedInput), subdir);
@@ -388,7 +389,8 @@ static Flake getFlake(
         debug("refetching input '%s' due to self attribute", newLockedRef);
         // FIXME: need to remove attrs that are invalidated by the changed input attrs, such as 'narHash'.
         newLockedRef.input.attrs.erase("narHash");
-        auto cachedInput2 = state.inputCache->getAccessor(state.store, newLockedRef.input, fetchers::UseRegistries::No);
+        auto cachedInput2 = state.inputCache->getAccessor(
+            state.fetchSettings, state.store, newLockedRef.input, fetchers::UseRegistries::No);
         cachedInput.accessor = cachedInput2.accessor;
         lockedRef = FlakeRef(std::move(cachedInput2.lockedInput), newLockedRef.subdir);
     }
@@ -704,7 +706,8 @@ lockFlake(const Settings & settings, EvalState & state, const FlakeRef & topRef,
                            this input. */
                         debug("creating new input '%s'", inputAttrPathS);
 
-                        if (!lockFlags.allowUnlocked && !input.ref->input.isLocked() && !input.ref->input.isRelative())
+                        if (!lockFlags.allowUnlocked && !input.ref->input.isLocked(state.fetchSettings)
+                            && !input.ref->input.isRelative())
                             throw Error("cannot update unlocked flake input '%s' in pure mode", inputAttrPathS);
 
                         /* Note: in case of an --override-input, we use
@@ -753,7 +756,7 @@ lockFlake(const Settings & settings, EvalState & state, const FlakeRef & topRef,
                                     return {*resolvedPath, *input.ref};
                                 } else {
                                     auto cachedInput = state.inputCache->getAccessor(
-                                        state.store, input.ref->input, useRegistriesInputs);
+                                        state.fetchSettings, state.store, input.ref->input, useRegistriesInputs);
 
                                     auto lockedRef = FlakeRef(std::move(cachedInput.lockedInput), input.ref->subdir);
 

src/libflake/flakeref.cc

@@ -64,9 +64,10 @@ std::ostream & operator<<(std::ostream & str, const FlakeRef & flakeRef)
     return str;
 }
 
-FlakeRef FlakeRef::resolve(ref<Store> store, fetchers::UseRegistries useRegistries) const
+FlakeRef FlakeRef::resolve(
+    const fetchers::Settings & fetchSettings, ref<Store> store, fetchers::UseRegistries useRegistries) const
 {
-    auto [input2, extraAttrs] = lookupInRegistries(store, input, useRegistries);
+    auto [input2, extraAttrs] = lookupInRegistries(fetchSettings, store, input, useRegistries);
     return FlakeRef(std::move(input2), fetchers::maybeGetStrAttr(extraAttrs, "dir").value_or(subdir));
 }
 
@@ -287,9 +288,10 @@ FlakeRef FlakeRef::fromAttrs(const fetchers::Settings & fetchSettings, const fet
         fetchers::maybeGetStrAttr(attrs, "dir").value_or(""));
 }
 
-std::pair<ref<SourceAccessor>, FlakeRef> FlakeRef::lazyFetch(ref<Store> store) const
+std::pair<ref<SourceAccessor>, FlakeRef>
+FlakeRef::lazyFetch(const fetchers::Settings & fetchSettings, ref<Store> store) const
 {
-    auto [accessor, lockedInput] = input.getAccessor(store);
+    auto [accessor, lockedInput] = input.getAccessor(fetchSettings, store);
     return {accessor, FlakeRef(std::move(lockedInput), subdir)};
 }
 

src/libflake/include/nix/flake/flakeref.hh

@@ -71,11 +71,15 @@ struct FlakeRef
 
     fetchers::Attrs toAttrs() const;
 
-    FlakeRef resolve(ref<Store> store, fetchers::UseRegistries useRegistries = fetchers::UseRegistries::All) const;
+    FlakeRef resolve(
+        const fetchers::Settings & fetchSettings,
+        ref<Store> store,
+        fetchers::UseRegistries useRegistries = fetchers::UseRegistries::All) const;
 
     static FlakeRef fromAttrs(const fetchers::Settings & fetchSettings, const fetchers::Attrs & attrs);
 
-    std::pair<ref<SourceAccessor>, FlakeRef> lazyFetch(ref<Store> store) const;
+    std::pair<ref<SourceAccessor>, FlakeRef>
+    lazyFetch(const fetchers::Settings & fetchSettings, ref<Store> store) const;
 
     /**
      * Canonicalize a flakeref for the purpose of comparing "old" and

src/libflake/lockfile.cc

@@ -74,7 +74,7 @@ LockedNode::LockedNode(const fetchers::Settings & fetchSettings, const nlohmann:
     , parentInputAttrPath(
           json.find("parent") != json.end() ? (std::optional<InputAttrPath>) json["parent"] : std::nullopt)
 {
-    if (!lockedRef.input.isLocked() && !lockedRef.input.isRelative()) {
+    if (!lockedRef.input.isLocked(fetchSettings) && !lockedRef.input.isRelative()) {
         if (lockedRef.input.getNarHash())
             warn(
                 "Lock file entry '%s' is unlocked (e.g. lacks a Git revision) but is checked by NAR hash. "
@@ -282,7 +282,7 @@ std::optional<FlakeRef> LockFile::isUnlocked(const fetchers::Settings & fetchSet
        latter case, we can verify the input but we may not be able to
        fetch it from anywhere. */
     auto isConsideredLocked = [&](const fetchers::Input & input) {
-        return input.isLocked() || (fetchSettings.allowDirtyLocks && input.getNarHash());
+        return input.isLocked(fetchSettings) || (fetchSettings.allowDirtyLocks && input.getNarHash());
     };
 
     for (auto & i : nodes) {

src/libmain/progress-bar.cc

@@ -467,8 +467,6 @@ public:
 
     std::string getStatus(State & state)
     {
-        auto MiB = 1024.0 * 1024.0;
-
         std::string res;
 
         auto renderActivity =
@@ -516,6 +514,65 @@ public:
                 return s;
             };
 
+        auto renderSizeActivity = [&](ActivityType type, const std::string & itemFmt = "%s") {
+            auto & act = state.activitiesByType[type];
+            uint64_t done = act.done, expected = act.done, running = 0, failed = act.failed;
+            for (auto & j : act.its) {
+                done += j.second->done;
+                expected += j.second->expected;
+                running += j.second->running;
+                failed += j.second->failed;
+            }
+
+            expected = std::max(expected, act.expected);
+
+            std::optional<SizeUnit> commonUnit;
+            std::string s;
+
+            if (running || done || expected || failed) {
+                if (running)
+                    if (expected != 0) {
+                        commonUnit = getCommonSizeUnit({(int64_t) running, (int64_t) done, (int64_t) expected});
+                        s =
+                            fmt(ANSI_BLUE "%s" ANSI_NORMAL "/" ANSI_GREEN "%s" ANSI_NORMAL "/%s",
+                                commonUnit ? renderSizeWithoutUnit(running, *commonUnit) : renderSize(running),
+                                commonUnit ? renderSizeWithoutUnit(done, *commonUnit) : renderSize(done),
+                                commonUnit ? renderSizeWithoutUnit(expected, *commonUnit) : renderSize(expected));
+                    } else {
+                        commonUnit = getCommonSizeUnit({(int64_t) running, (int64_t) done});
+                        s =
+                            fmt(ANSI_BLUE "%s" ANSI_NORMAL "/" ANSI_GREEN "%s" ANSI_NORMAL,
+                                commonUnit ? renderSizeWithoutUnit(running, *commonUnit) : renderSize(running),
+                                commonUnit ? renderSizeWithoutUnit(done, *commonUnit) : renderSize(done));
+                    }
+                else if (expected != done)
+                    if (expected != 0) {
+                        commonUnit = getCommonSizeUnit({(int64_t) done, (int64_t) expected});
+                        s =
+                            fmt(ANSI_GREEN "%s" ANSI_NORMAL "/%s",
+                                commonUnit ? renderSizeWithoutUnit(done, *commonUnit) : renderSize(done),
+                                commonUnit ? renderSizeWithoutUnit(expected, *commonUnit) : renderSize(expected));
+                    } else {
+                        commonUnit = getSizeUnit(done);
+                        s = fmt(ANSI_GREEN "%s" ANSI_NORMAL, renderSizeWithoutUnit(done, *commonUnit));
+                    }
+                else {
+                    commonUnit = getSizeUnit(done);
+                    s = fmt(done ? ANSI_GREEN "%s" ANSI_NORMAL : "%s", renderSizeWithoutUnit(done, *commonUnit));
+                }
+
+                if (commonUnit)
+                    s = fmt("%s %siB", s, getSizeUnitSuffix(*commonUnit));
+
+                s = fmt(itemFmt, s);
+
+                if (failed)
+                    s += fmt(" (" ANSI_RED "%s failed" ANSI_NORMAL ")", renderSize(failed));
+            }
+
+            return s;
+        };
+
         auto showActivity =
             [&](ActivityType type, const std::string & itemFmt, const std::string & numberFmt = "%d", double unit = 1) {
                 auto s = renderActivity(type, itemFmt, numberFmt, unit);
@@ -529,7 +586,7 @@ public:
         showActivity(actBuilds, "%s built");
 
         auto s1 = renderActivity(actCopyPaths, "%s copied");
-        auto s2 = renderActivity(actCopyPath, "%s MiB", "%.1f", MiB);
+        auto s2 = renderSizeActivity(actCopyPath);
 
         if (!s1.empty() || !s2.empty()) {
             if (!res.empty())
@@ -545,12 +602,12 @@ public:
             }
         }
 
-        showActivity(actFileTransfer, "%s MiB DL", "%.1f", MiB);
+        renderSizeActivity(actFileTransfer, "%s DL");
 
         {
             auto s = renderActivity(actOptimiseStore, "%s paths optimised");
             if (s != "") {
-                s += fmt(", %.1f MiB / %d inodes freed", state.bytesLinked / MiB, state.filesLinked);
+                s += fmt(", %s / %d inodes freed", renderSize(state.bytesLinked), state.filesLinked);
                 if (!res.empty())
                     res += ", ";
                 res += s;

src/libstore-test-support/include/nix/store/tests/protocol.hh

@@ -6,6 +6,7 @@
 
 #include "nix/store/tests/libstore.hh"
 #include "nix/util/tests/characterization.hh"
+#include "nix/util/tests/json-characterization.hh"
 
 namespace nix {
 
@@ -16,12 +17,30 @@ class ProtoTest : public CharacterizationTest
 
     std::filesystem::path goldenMaster(std::string_view testStem) const override
     {
-        return unitTestData / (std::string{testStem + ".bin"});
+        return unitTestData / testStem;
     }
 
 public:
     Path storeDir = "/nix/store";
     StoreDirConfig store{storeDir};
+
+    /**
+     * Golden test for `T` JSON reading
+     */
+    template<typename T>
+    void readJsonTest(PathView testStem, const T & expected)
+    {
+        nix::readJsonTest(*this, testStem, expected);
+    }
+
+    /**
+     * Golden test for `T` JSON write
+     */
+    template<typename T>
+    void writeJsonTest(PathView testStem, const T & decoded)
+    {
+        nix::writeJsonTest(*this, testStem, decoded);
+    }
 };
 
 template<class Proto, const char * protocolDir>
@@ -34,7 +53,7 @@ public:
     template<typename T>
     void readProtoTest(PathView testStem, typename Proto::Version version, T expected)
     {
-        CharacterizationTest::readTest(testStem, [&](const auto & encoded) {
+        CharacterizationTest::readTest(std::string{testStem + ".bin"}, [&](const auto & encoded) {
             T got = ({
                 StringSource from{encoded};
                 Proto::template Serialise<T>::read(
@@ -55,7 +74,7 @@ public:
     template<typename T>
     void writeProtoTest(PathView testStem, typename Proto::Version version, const T & decoded)
     {
-        CharacterizationTest::writeTest(testStem, [&]() {
+        CharacterizationTest::writeTest(std::string{testStem + ".bin"}, [&]() {
             StringSink to;
             Proto::template Serialise<T>::write(
                 this->store,
@@ -69,14 +88,25 @@ public:
     }
 };
 
-#define VERSIONED_CHARACTERIZATION_TEST(FIXTURE, NAME, STEM, VERSION, VALUE) \
-    TEST_F(FIXTURE, NAME##_read)                                             \
-    {                                                                        \
-        readProtoTest(STEM, VERSION, VALUE);                                 \
-    }                                                                        \
-    TEST_F(FIXTURE, NAME##_write)                                            \
-    {                                                                        \
-        writeProtoTest(STEM, VERSION, VALUE);                                \
+#define VERSIONED_CHARACTERIZATION_TEST_NO_JSON(FIXTURE, NAME, STEM, VERSION, VALUE) \
+    TEST_F(FIXTURE, NAME##_read)                                                     \
+    {                                                                                \
+        readProtoTest(STEM, VERSION, VALUE);                                         \
+    }                                                                                \
+    TEST_F(FIXTURE, NAME##_write)                                                    \
+    {                                                                                \
+        writeProtoTest(STEM, VERSION, VALUE);                                        \
+    }
+
+#define VERSIONED_CHARACTERIZATION_TEST(FIXTURE, NAME, STEM, VERSION, VALUE)     \
+    VERSIONED_CHARACTERIZATION_TEST_NO_JSON(FIXTURE, NAME, STEM, VERSION, VALUE) \
+    TEST_F(FIXTURE, NAME##_json_read)                                            \
+    {                                                                            \
+        readJsonTest(STEM, VALUE);                                               \
+    }                                                                            \
+    TEST_F(FIXTURE, NAME##_json_write)                                           \
+    {                                                                            \
+        writeJsonTest(STEM, VALUE);                                              \
     }
 
 } // namespace nix

src/libstore-tests/common-protocol.cc

@@ -3,6 +3,7 @@
 #include <nlohmann/json.hpp>
 #include <gtest/gtest.h>
 
+#include "nix/util/json-utils.hh"
 #include "nix/store/common-protocol.hh"
 #include "nix/store/common-protocol-impl.hh"
 #include "nix/store/build-result.hh"
@@ -22,7 +23,7 @@ public:
     template<typename T>
     void readProtoTest(PathView testStem, const T & expected)
     {
-        CharacterizationTest::readTest(testStem, [&](const auto & encoded) {
+        CharacterizationTest::readTest(std::string{testStem + ".bin"}, [&](const auto & encoded) {
             T got = ({
                 StringSource from{encoded};
                 CommonProto::Serialise<T>::read(store, CommonProto::ReadConn{.from = from});
@@ -38,7 +39,7 @@ public:
     template<typename T>
     void writeProtoTest(PathView testStem, const T & decoded)
     {
-        CharacterizationTest::writeTest(testStem, [&]() -> std::string {
+        CharacterizationTest::writeTest(std::string{testStem + ".bin"}, [&]() -> std::string {
             StringSink to;
             CommonProto::Serialise<T>::write(store, CommonProto::WriteConn{.to = to}, decoded);
             return to.s;
@@ -54,6 +55,14 @@ public:
     TEST_F(CommonProtoTest, NAME##_write)        \
     {                                            \
         writeProtoTest(STEM, VALUE);             \
+    }                                            \
+    TEST_F(CommonProtoTest, NAME##_json_read)    \
+    {                                            \
+        readJsonTest(STEM, VALUE);               \
+    }                                            \
+    TEST_F(CommonProtoTest, NAME##_json_write)   \
+    {                                            \
+        writeJsonTest(STEM, VALUE);              \
     }
 
 CHARACTERIZATION_TEST(

src/libstore-tests/data/common-protocol/content-address.json

@@ -0,0 +1,26 @@
+[
+  {
+    "hash": {
+      "algorithm": "sha256",
+      "format": "base64",
+      "hash": "+Xc9Ll6mcPltwaewrk/BAQ56Y3G5T//wzhKUc0zrYu0="
+    },
+    "method": "text"
+  },
+  {
+    "hash": {
+      "algorithm": "sha1",
+      "format": "base64",
+      "hash": "gGemBoenViNZM3hiwqns/Fgzqwo="
+    },
+    "method": "flat"
+  },
+  {
+    "hash": {
+      "algorithm": "sha256",
+      "format": "base64",
+      "hash": "EMIJ+giQ/gLIWoxmPKjno3zHZrxbGymgzGGyZvZBIdM="
+    },
+    "method": "nar"
+  }
+]

src/libstore-tests/data/common-protocol/drv-output.json

@@ -0,0 +1,4 @@
+[
+  "sha256:15e3c560894cbb27085cf65b5a2ecb18488c999497f4531b6907a7581ce6d527!baz",
+  "sha256:6f869f9ea2823bda165e06076fd0de4366dead2c0e8d2dbbad277d4f15c373f5!quux"
+]

src/libstore-tests/data/common-protocol/optional-content-address.json

@@ -0,0 +1,11 @@
+[
+  null,
+  {
+    "hash": {
+      "algorithm": "sha1",
+      "format": "base64",
+      "hash": "gGemBoenViNZM3hiwqns/Fgzqwo="
+    },
+    "method": "flat"
+  }
+]

src/libstore-tests/data/common-protocol/optional-store-path.json

@@ -0,0 +1,4 @@
+[
+  null,
+  "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo-bar"
+]

src/libstore-tests/data/common-protocol/realisation-with-deps.json

@@ -0,0 +1,13 @@
+[
+  {
+    "dependentRealisations": {
+      "sha256:6f869f9ea2823bda165e06076fd0de4366dead2c0e8d2dbbad277d4f15c373f5!quux": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo"
+    },
+    "id": "sha256:15e3c560894cbb27085cf65b5a2ecb18488c999497f4531b6907a7581ce6d527!baz",
+    "outPath": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo",
+    "signatures": [
+      "asdf",
+      "qwer"
+    ]
+  }
+]

src/libstore-tests/data/common-protocol/realisation.json

@@ -0,0 +1,17 @@
+[
+  {
+    "dependentRealisations": {},
+    "id": "sha256:15e3c560894cbb27085cf65b5a2ecb18488c999497f4531b6907a7581ce6d527!baz",
+    "outPath": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo",
+    "signatures": []
+  },
+  {
+    "dependentRealisations": {},
+    "id": "sha256:15e3c560894cbb27085cf65b5a2ecb18488c999497f4531b6907a7581ce6d527!baz",
+    "outPath": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo",
+    "signatures": [
+      "asdf",
+      "qwer"
+    ]
+  }
+]

src/libstore-tests/data/common-protocol/set.json

@@ -0,0 +1,22 @@
+[
+  [],
+  [
+    ""
+  ],
+  [
+    "",
+    "bar",
+    "foo"
+  ],
+  [
+    [],
+    [
+      ""
+    ],
+    [
+      "",
+      "1",
+      "2"
+    ]
+  ]
+]

src/libstore-tests/data/common-protocol/store-path.json

@@ -0,0 +1,4 @@
+[
+  "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo",
+  "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo-bar"
+]

src/libstore-tests/data/common-protocol/string.json

@@ -0,0 +1,7 @@
+[
+  "",
+  "hi",
+  "white rabbit",
+  "大白兔",
+  "oh no "
+]

src/libstore-tests/data/common-protocol/vector.json

@@ -0,0 +1,22 @@
+[
+  [],
+  [
+    ""
+  ],
+  [
+    "",
+    "foo",
+    "bar"
+  ],
+  [
+    [],
+    [
+      ""
+    ],
+    [
+      "",
+      "1",
+      "2"
+    ]
+  ]
+]

src/libstore-tests/data/derivation/ca/advanced-attributes-structured-attrs.json

@@ -69,7 +69,8 @@
     "outputChecks": {
       "bin": {
         "disallowedReferences": [
-          "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g"
+          "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g",
+          "dev"
         ],
         "disallowedRequisites": [
           "/07f301yqyz8c6wf6bbbavb2q39j4n8kmcly1s09xadyhgy6x2wr8"
@@ -84,7 +85,8 @@
           "/164j69y6zir9z0339n8pjigg3rckinlr77bxsavzizdaaljb7nh9"
         ],
         "allowedRequisites": [
-          "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z"
+          "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z",
+          "bin"
         ]
       }
     },

src/libstore-tests/data/derivation/ca/advanced-attributes.json

@@ -11,9 +11,9 @@
     "__sandboxProfile": "sandcastle",
     "allowSubstitutes": "",
     "allowedReferences": "/164j69y6zir9z0339n8pjigg3rckinlr77bxsavzizdaaljb7nh9",
-    "allowedRequisites": "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z",
+    "allowedRequisites": "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z bin",
     "builder": "/bin/bash",
-    "disallowedReferences": "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g",
+    "disallowedReferences": "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g dev",
     "disallowedRequisites": "/07f301yqyz8c6wf6bbbavb2q39j4n8kmcly1s09xadyhgy6x2wr8",
     "exportReferencesGraph": "refs1 /164j69y6zir9z0339n8pjigg3rckinlr77bxsavzizdaaljb7nh9 refs2 /nix/store/qnml92yh97a6fbrs2m5qg5cqlc8vni58-bar.drv",
     "impureEnvVars": "UNICORN",

src/libstore-tests/data/derivation/ca/all_set.json

@@ -23,10 +23,12 @@
         "/164j69y6zir9z0339n8pjigg3rckinlr77bxsavzizdaaljb7nh9"
       ],
       "allowedRequisites": [
-        "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z"
+        "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z",
+        "bin"
       ],
       "disallowedReferences": [
-        "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g"
+        "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g",
+        "dev"
       ],
       "disallowedRequisites": [
         "/07f301yqyz8c6wf6bbbavb2q39j4n8kmcly1s09xadyhgy6x2wr8"

src/libstore-tests/data/derivation/ca/structuredAttrs_all_set.json

@@ -23,7 +23,8 @@
         "allowedReferences": null,
         "allowedRequisites": null,
         "disallowedReferences": [
-          "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g"
+          "/0nyw57wm2iicnm9rglvjmbci3ikmcp823czdqdzdcgsnnwqps71g",
+          "dev"
         ],
         "disallowedRequisites": [
           "/07f301yqyz8c6wf6bbbavb2q39j4n8kmcly1s09xadyhgy6x2wr8"
@@ -46,7 +47,8 @@
           "/164j69y6zir9z0339n8pjigg3rckinlr77bxsavzizdaaljb7nh9"
         ],
         "allowedRequisites": [
-          "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z"
+          "/0nr45p69vn6izw9446wsh9bng9nndhvn19kpsm4n96a5mycw0s4z",
+          "bin"
         ],
         "disallowedReferences": [],
         "disallowedRequisites": [],

src/libstore-tests/data/derivation/ia/advanced-attributes-structured-attrs.json

@@ -5,9 +5,9 @@
   ],
   "builder": "/bin/bash",
   "env": {
-    "bin": "/nix/store/33qms3h55wlaspzba3brlzlrm8m2239g-advanced-attributes-structured-attrs-bin",
-    "dev": "/nix/store/wyfgwsdi8rs851wmy1xfzdxy7y5vrg5l-advanced-attributes-structured-attrs-dev",
-    "out": "/nix/store/7cxy4zx1vqc885r4jl2l64pymqbdmhii-advanced-attributes-structured-attrs"
+    "bin": "/nix/store/cnpasdljgkhnwaf78cf3qygcp4qbki1c-advanced-attributes-structured-attrs-bin",
+    "dev": "/nix/store/ijq6mwpa9jbnpnl33qldfqihrr38kprx-advanced-attributes-structured-attrs-dev",
+    "out": "/nix/store/h1vh648d3p088kdimy0r8ngpfx7c3nzw-advanced-attributes-structured-attrs"
   },
   "inputs": {
     "drvs": {
@@ -33,13 +33,13 @@
   "name": "advanced-attributes-structured-attrs",
   "outputs": {
     "bin": {
-      "path": "33qms3h55wlaspzba3brlzlrm8m2239g-advanced-attributes-structured-attrs-bin"
+      "path": "cnpasdljgkhnwaf78cf3qygcp4qbki1c-advanced-attributes-structured-attrs-bin"
     },
     "dev": {
-      "path": "wyfgwsdi8rs851wmy1xfzdxy7y5vrg5l-advanced-attributes-structured-attrs-dev"
+      "path": "ijq6mwpa9jbnpnl33qldfqihrr38kprx-advanced-attributes-structured-attrs-dev"
     },
     "out": {
-      "path": "7cxy4zx1vqc885r4jl2l64pymqbdmhii-advanced-attributes-structured-attrs"
+      "path": "h1vh648d3p088kdimy0r8ngpfx7c3nzw-advanced-attributes-structured-attrs"
     }
   },
   "structuredAttrs": {
@@ -66,7 +66,8 @@
     "outputChecks": {
       "bin": {
         "disallowedReferences": [
-          "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar"
+          "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar",
+          "dev"
         ],
         "disallowedRequisites": [
           "/nix/store/9b61w26b4avv870dw0ymb6rw4r1hzpws-bar-dev"
@@ -81,7 +82,8 @@
           "/nix/store/p0hax2lzvjpfc2gwkk62xdglz0fcqfzn-foo"
         ],
         "allowedRequisites": [
-          "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev"
+          "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev",
+          "bin"
         ]
       }
     },

src/libstore-tests/data/derivation/ia/advanced-attributes.json

@@ -11,14 +11,14 @@
     "__sandboxProfile": "sandcastle",
     "allowSubstitutes": "",
     "allowedReferences": "/nix/store/p0hax2lzvjpfc2gwkk62xdglz0fcqfzn-foo",
-    "allowedRequisites": "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev",
+    "allowedRequisites": "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev bin",
     "builder": "/bin/bash",
-    "disallowedReferences": "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar",
+    "disallowedReferences": "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar dev",
     "disallowedRequisites": "/nix/store/9b61w26b4avv870dw0ymb6rw4r1hzpws-bar-dev",
     "exportReferencesGraph": "refs1 /nix/store/p0hax2lzvjpfc2gwkk62xdglz0fcqfzn-foo refs2 /nix/store/vj2i49jm2868j2fmqvxm70vlzmzvgv14-bar.drv",
     "impureEnvVars": "UNICORN",
     "name": "advanced-attributes",
-    "out": "/nix/store/wyhpwd748pns4k7svh48wdrc8kvjk0ra-advanced-attributes",
+    "out": "/nix/store/ymqmybkq5j4nd1xplw6ccdpbjnfi017v-advanced-attributes",
     "preferLocalBuild": "1",
     "requiredSystemFeatures": "rainbow uid-range",
     "system": "my-system"
@@ -47,7 +47,7 @@
   "name": "advanced-attributes",
   "outputs": {
     "out": {
-      "path": "wyhpwd748pns4k7svh48wdrc8kvjk0ra-advanced-attributes"
+      "path": "ymqmybkq5j4nd1xplw6ccdpbjnfi017v-advanced-attributes"
     }
   },
   "system": "my-system",

src/libstore-tests/data/derivation/ia/all_set.json

@@ -23,10 +23,12 @@
         "/nix/store/p0hax2lzvjpfc2gwkk62xdglz0fcqfzn-foo"
       ],
       "allowedRequisites": [
-        "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev"
+        "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev",
+        "bin"
       ],
       "disallowedReferences": [
-        "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar"
+        "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar",
+        "dev"
       ],
       "disallowedRequisites": [
         "/nix/store/9b61w26b4avv870dw0ymb6rw4r1hzpws-bar-dev"

src/libstore-tests/data/derivation/ia/structuredAttrs_all_set.json

@@ -23,7 +23,8 @@
         "allowedReferences": null,
         "allowedRequisites": null,
         "disallowedReferences": [
-          "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar"
+          "/nix/store/r5cff30838majxk5mp3ip2diffi8vpaj-bar",
+          "dev"
         ],
         "disallowedRequisites": [
           "/nix/store/9b61w26b4avv870dw0ymb6rw4r1hzpws-bar-dev"
@@ -46,7 +47,8 @@
           "/nix/store/p0hax2lzvjpfc2gwkk62xdglz0fcqfzn-foo"
         ],
         "allowedRequisites": [
-          "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev"
+          "/nix/store/z0rjzy29v9k5qa4nqpykrbzirj7sd43v-foo-dev",
+          "bin"
         ],
         "disallowedReferences": [],
         "disallowedRequisites": [],

src/libstore-tests/data/nar-info/impure.json

@@ -9,9 +9,17 @@
   },
   "compression": "xz",
   "deriver": "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar.drv",
-  "downloadHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "downloadHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "downloadSize": 4029176,
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 34878,
   "references": [
     "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar",

src/libstore-tests/data/nar-info/pure.json

@@ -7,7 +7,11 @@
     },
     "method": "nar"
   },
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 34878,
   "references": [
     "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar",

src/libstore-tests/data/path-info/empty_impure.json

@@ -1,7 +1,11 @@
 {
   "ca": null,
   "deriver": null,
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 0,
   "references": [],
   "registrationTime": null,

src/libstore-tests/data/path-info/empty_pure.json

@@ -1,6 +1,10 @@
 {
   "ca": null,
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 0,
   "references": [],
   "version": 2

src/libstore-tests/data/path-info/impure.json

@@ -8,7 +8,11 @@
     "method": "nar"
   },
   "deriver": "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar.drv",
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 34878,
   "references": [
     "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar",

src/libstore-tests/data/path-info/pure.json

@@ -7,7 +7,11 @@
     },
     "method": "nar"
   },
-  "narHash": "sha256-FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc=",
+  "narHash": {
+    "algorithm": "sha256",
+    "format": "base64",
+    "hash": "FePFYIlMuycIXPZbWi7LGEiMmZSX9FMbaQenWBzm1Sc="
+  },
   "narSize": 34878,
   "references": [
     "/nix/store/g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar",

src/libstore-tests/data/serve-protocol/build-result-2.2.json

@@ -0,0 +1,28 @@
+[
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": false,
+    "startTime": 0,
+    "status": "OutputRejected",
+    "stopTime": 0,
+    "success": false,
+    "timesBuilt": 0
+  },
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": false,
+    "startTime": 0,
+    "status": "NotDeterministic",
+    "stopTime": 0,
+    "success": false,
+    "timesBuilt": 0
+  },
+  {
+    "builtOutputs": {},
+    "startTime": 0,
+    "status": "Built",
+    "stopTime": 0,
+    "success": true,
+    "timesBuilt": 0
+  }
+]

src/libstore-tests/data/serve-protocol/build-result-2.3.json

@@ -0,0 +1,28 @@
+[
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": false,
+    "startTime": 0,
+    "status": "OutputRejected",
+    "stopTime": 0,
+    "success": false,
+    "timesBuilt": 0
+  },
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": true,
+    "startTime": 30,
+    "status": "NotDeterministic",
+    "stopTime": 50,
+    "success": false,
+    "timesBuilt": 3
+  },
+  {
+    "builtOutputs": {},
+    "startTime": 30,
+    "status": "Built",
+    "stopTime": 50,
+    "success": true,
+    "timesBuilt": 0
+  }
+]

src/libstore-tests/data/serve-protocol/build-result-2.6.json

@@ -0,0 +1,41 @@
+[
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": false,
+    "startTime": 0,
+    "status": "OutputRejected",
+    "stopTime": 0,
+    "success": false,
+    "timesBuilt": 0
+  },
+  {
+    "errorMsg": "no idea why",
+    "isNonDeterministic": true,
+    "startTime": 30,
+    "status": "NotDeterministic",
+    "stopTime": 50,
+    "success": false,
+    "timesBuilt": 3
+  },
+  {
+    "builtOutputs": {
+      "bar": {
+        "dependentRealisations": {},
+        "id": "sha256:6f869f9ea2823bda165e06076fd0de4366dead2c0e8d2dbbad277d4f15c373f5!bar",
+        "outPath": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-bar",
+        "signatures": []
+      },
+      "foo": {
+        "dependentRealisations": {},
+        "id": "sha256:6f869f9ea2823bda165e06076fd0de4366dead2c0e8d2dbbad277d4f15c373f5!foo",
+        "outPath": "g1w7hy3qg1w7hy3qg1w7hy3qg1w7hy3q-foo",
+        "signatures": []
+      }
+    },
+    "startTime": 30,
+    "status": "Built",
+    "stopTime": 50,
+    "success": true,
+    "timesBuilt": 1
+  }
+]

src/libstore-tests/data/serve-protocol/content-address.json

@@ -0,0 +1,26 @@
+[
+  {
+    "hash": {
+      "algorithm": "sha256",
+      "format": "base64",
+      "hash": "+Xc9Ll6mcPltwaewrk/BAQ56Y3G5T//wzhKUc0zrYu0="
+    },
+    "method": "text"
+  },
+  {
+    "hash": {
+      "algorithm": "sha1",
+      "format": "base64",
+      "hash": "gGemBoenViNZM3hiwqns/Fgzqwo="
+    },
+    "method": "flat"
+  },
+  {
+    "hash": {
+      "algorithm": "sha256",
+      "format": "base64",
+      "hash": "EMIJ+giQ/gLIWoxmPKjno3zHZrxbGymgzGGyZvZBIdM="
+    },
+    "method": "nar"
+  }
+]

src/libstore-tests/data/serve-protocol/drv-output.json

@@ -0,0 +1,4 @@
+[
+  "sha256:15e3c560894cbb27085cf65b5a2ecb18488c999497f4531b6907a7581ce6d527!baz",
+  "sha256:6f869f9ea2823bda165e06076fd0de4366dead2c0e8d2dbbad277d4f15c373f5!quux"
+]