perf(ci): checkfast profile for the test lanes — 4.9x faster compiles

The workspace ships with fat LTO + codegen-units=1, and under --release
every test binary pays that link. checkfast (inherits release, lto off,
cgu 16) compiles the workspace test targets in 32s where release takes
157s (measured locally; 773/773 tests pass identically, run time
unchanged at ~11s — verdicts are semantics, not codegen). nextest and
lambda-tests build against a second cached deps closure under the new
profile; clippy, abgen-all and every shipped artifact keep full-fat
release. The cache composite gains a flavor input (nix-v4 keys) so the
two closures never race for one entry.

Author: eordano

.github/actions/nix-store-cache/action.yml

@@ -10,6 +10,15 @@ description: >
   save action. runner.arch is part of the key so x86_64 and aarch64 stores
   never cross.
 
+inputs:
+  flavor:
+    description: >
+      Which deps closure this job consumes (build = release-profile deps;
+      checkfast = the test-lane profile). Separate entries per flavor keep
+      the first-save-wins race from caching one flavor's closure under the
+      other's key.
+    default: build
+
 runs:
   using: composite
   steps:
@@ -49,8 +58,8 @@ runs:
       id: restore
       with:
         path: /var/tmp/abgen-nixcache
-        key: nix-v3-${{ runner.os }}-${{ runner.arch }}-${{ hashFiles('flake.lock', 'Cargo.lock', '**/Cargo.toml', 'flake.nix', 'rust-toolchain.toml', 'nix/*.nix', 'crate/abgen-wasm/toolchain/flake.lock', 'crate/abgen-wasm/toolchain/flake.nix') }}
-        restore-keys: nix-v3-${{ runner.os }}-${{ runner.arch }}-
+        key: nix-v4-${{ runner.os }}-${{ runner.arch }}-${{ inputs.flavor }}-${{ hashFiles('flake.lock', 'Cargo.lock', '**/Cargo.toml', 'flake.nix', 'rust-toolchain.toml', 'nix/*.nix', 'crate/abgen-wasm/toolchain/flake.lock', 'crate/abgen-wasm/toolchain/flake.nix') }}
+        restore-keys: nix-v4-${{ runner.os }}-${{ runner.arch }}-${{ inputs.flavor }}-
 
     - name: expose cache facts to the save action
       shell: bash

.github/workflows/ci.yml

@@ -208,10 +208,11 @@ jobs:
           - lane: x86 nextest
             runner: ubuntu-24.04
             system: x86_64-linux
-            deps: .#checks.x86_64-linux.deps
+            deps: .#checks.x86_64-linux.deps-checkfast
             attrs: nextest
             gate: run_crate
             marker: x86-nextest
+            flavor: checkfast
           - lane: x86 build + smoke
             runner: ubuntu-24.04
             system: x86_64-linux
@@ -222,17 +223,19 @@ jobs:
           - lane: arm nextest
             runner: ubuntu-24.04-arm
             system: aarch64-linux
-            deps: .#checks.aarch64-linux.deps
+            deps: .#checks.aarch64-linux.deps-checkfast
             attrs: nextest
             gate: run_crate
             marker: arm-nextest
+            flavor: checkfast
           - lane: arm lambda + smoke
             runner: ubuntu-24.04-arm
             system: aarch64-linux
-            deps: .#checks.aarch64-linux.deps
+            deps: .#checks.aarch64-linux.deps-checkfast
             attrs: lambda-tests native-smoke
             gate: run_lambda
             marker: arm-lambda
+            flavor: checkfast
           - lane: arm lints
             runner: ubuntu-24.04-arm
             system: aarch64-linux
@@ -253,6 +256,8 @@ jobs:
       # silently a no-op.
       - uses: ./.github/actions/nix-store-cache
         if: needs.nix-gate.outputs[matrix.gate] == 'true' && needs.nix-gate.outputs[format('skip_{0}', matrix.marker)] != 'true'
+        with:
+          flavor: ${{ matrix.flavor || 'build' }}
       - name: deps (dependency closures)
         if: needs.nix-gate.outputs[matrix.gate] == 'true' && needs.nix-gate.outputs[format('skip_{0}', matrix.marker)] != 'true'
         run: nix build --no-link ${{ matrix.deps }}

Cargo.toml

@@ -21,3 +21,12 @@ version = "0.17.1"
 [profile.release]
 lto = "fat"
 codegen-units = 1
+
+# Test-lane profile: identical semantics, none of the fat-LTO link cost.
+# Shipped artifacts keep [profile.release]; only the CI test checks and
+# local test runs use this. Parity is semantics, not codegen: encoder
+# outputs are optimization-independent.
+[profile.checkfast]
+inherits = "release"
+lto = "off"
+codegen-units = 16

flake.nix

@@ -176,7 +176,8 @@
 
           checks = import ./nix/checks.nix {
             inherit lib system pkgs craneLib wasmCheck;
-            inherit (build) commonArgs cargoArtifacts abgenConsumersPkg;
+            inherit (build) commonArgs cargoArtifacts cargoArtifactsCheckfast
+              abgenConsumersPkg;
           };
 
         });

nix/build.nix

@@ -51,6 +51,18 @@ let
     doCheck = true;
   });
 
+  # Deps for the checkfast profile (lto off, codegen-units 16): the test
+  # checks compile 4.9x faster against these — measured 157s -> 32s per
+  # workspace test compile — with byte-identical test verdicts. Shipped
+  # artifacts and clippy keep the release-profile deps above.
+  cargoArtifactsCheckfast = craneLib.buildDepsOnly (commonArgs // {
+    inherit dummySrc;
+    pname = "abgen-checkfast";
+    version = "0";
+    doCheck = true;
+    CARGO_PROFILE = "checkfast";
+  });
+
   abgenAll = craneLib.buildPackage (commonArgs // {
     inherit cargoArtifacts;
     pname = "abgen-all";
@@ -77,6 +89,6 @@ let
   });
 in
 {
-  inherit commonArgs cargoArtifacts abgenAll abgenPkg abgenConsumersPkg
-    abgenCorpusPkg;
+  inherit commonArgs cargoArtifacts cargoArtifactsCheckfast abgenAll abgenPkg
+    abgenConsumersPkg abgenCorpusPkg;
 }

nix/checks.nix

@@ -1,9 +1,16 @@
-{ lib, system, pkgs, craneLib, commonArgs, cargoArtifacts, abgenConsumersPkg
-, wasmCheck }:
+{ lib, system, pkgs, craneLib, commonArgs, cargoArtifacts
+, cargoArtifactsCheckfast, abgenConsumersPkg, wasmCheck }:
 
 let
   src = commonArgs.src;
   withArtifacts = commonArgs // { inherit cargoArtifacts; };
+  # Test checks compile under the checkfast profile (lto off, cgu 16):
+  # 4.9x faster workspace test compiles, byte-identical verdicts. Clippy
+  # and the shipped packages stay on release-profile artifacts.
+  withCheckfast = commonArgs // {
+    cargoArtifacts = cargoArtifactsCheckfast;
+    CARGO_PROFILE = "checkfast";
+  };
   abgenRoot = ''export ABGEN_ROOT="$PWD"'';
 
   archIndependent = {
@@ -33,8 +40,9 @@ let
     # deps stage builds exactly this, then publishes the binary cache, so a
     # failure in any later stage never costs the next run its warm deps.
     deps = cargoArtifacts;
+    deps-checkfast = cargoArtifactsCheckfast;
 
-    nextest = craneLib.cargoNextest (withArtifacts // {
+    nextest = craneLib.cargoNextest (withCheckfast // {
       doCheck = true;
       __darwinAllowLocalNetworking = true;
       cargoExtraArgs = "--locked";
@@ -66,7 +74,7 @@ let
   # is rejected: feature unification differs, a merged compile would not
   # test the no-server config the mac/windows legs ship.
   lambdaTests = {
-    lambda-tests = craneLib.cargoTest (withArtifacts // {
+    lambda-tests = craneLib.cargoTest (withCheckfast // {
       doCheck = true;
       __darwinAllowLocalNetworking = true;
       pname = "abgen-lambda-tests";