.github: always run tests, and other helpful fixes (#9348)

During work on our new registry client, I ran into frustrations with CI
where a misspelling in a comment caused the linter to fail, which caused
the tests to not run, which caused the build to not be cached, which
caused the next run to be slow, which caused me to be sad.

This commit address these issues, and pulls in some helpful changes
we've had in CI on ollama.com for some time now.

They are:

* Always run tests, even if the other checks fail.

Tests are the most important part of CI, and should always run. Failures
in tests can be correlated with failures in other checks, and can help
surface the root cause of the failure sooner. This is especially
important when the failure is platform specific, and the tests are not
platform independent.

* Check that `go generate` is clean.

This prevents 'go generate' abuse regressions. This codebase used to use
it to generate platform specific binary build artifacts. Let's make sure
that does not happen again and this powerful tool is used correctly, and
the generated code is checked in.

Also, while adding `go generate` the check, it was revealed that the
generated metal code was putting dates in the comments, resulting in
non-deterministic builds. This is a bad practice, and this commit fixes
that. Git tells us the most important date: the commit date along with
other associated changes.

* Check that `go mod tidy` is clean.

A new job to check that `go mod tidy` is clean was added, to prevent
easily preventable merge conflicts or go.mod changes being deferred to a
future PR that is unrelated to the change that caused the go.mod to
change.

* More robust caching.

We now cache the go build cache, and the go mod download cache
independently. This is because the download cache contains zips that can
be unpacked in parallel faster than they can be fetched and extracted by
tar. This speeds up the build significantly.

The linter is hostile enough. It does not need to also punish us with
longer build times due to small failures like misspellings.

.github/workflows/test.yaml

@@ -140,6 +140,13 @@ jobs:
         env:
           CMAKE_GENERATOR: Ninja
 
+  go_mod_tidy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: check that 'go mod tidy' is clean
+        run: go mod tidy --diff || (echo "Please run 'go mod tidy'." && exit 1)
+
   test:
     strategy:
       matrix:
@@ -149,14 +156,81 @@ jobs:
       CGO_ENABLED: '1'
       GOEXPERIMENT: 'synctest'
     steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-go@v5
+      - name: checkout
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # 4.2.2
+
+      - name: cache restore
+        uses: actions/cache/restore@1bd1e32a3bdc45362d1e726936510720a7c30a57 # v4.2.0
+        with:
+          # Note: unlike the other setups, this is only grabbing the mod download
+          # cache, rather than the whole mod directory, as the download cache
+          # contains zips that can be unpacked in parallel faster than they can be
+          # fetched and extracted by tar
+          path: |
+            ~/.cache/go-build
+            ~/go/pkg/mod/cache
+            ~\AppData\Local\go-build
+          # NOTE: The -3- here should be incremented when the scheme of data to be
+          # cached changes (e.g. path above changes).
+          key: ${{ github.job }}-${{ runner.os }}-${{ matrix.goarch }}-${{ matrix.buildflags }}-go-3-${{ hashFiles('**/go.sum') }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ github.job }}-${{ runner.os }}-${{ matrix.goarch }}-${{ matrix.buildflags }}-go-3-${{ hashFiles('**/go.sum') }}
+            ${{ github.job }}-${{ runner.os }}-${{ matrix.goarch }}-${{ matrix.buildflags }}-go-3-
+
+      - name: Setup Go
+        uses: actions/setup-go@v5
         with:
+          # The caching strategy of setup-go is less than ideal, and wastes
+          # time by not saving artifacts due to small failures like the linter
+          # complaining, etc. This means subsequent have to rebuild their world
+          # again until all checks pass. For instance, if you mispell a word,
+          # you're punished until you fix it. This is more hostile than
+          # helpful.
+          cache: false
+
           go-version-file: go.mod
+
+      # TODO(bmizerany): replace this heavy tool with just the
+      # tools/checks/binaries we want and then make them all run in parallel
+      # across jobs, not on a single tiny vm on Github Actions.
       - uses: golangci/golangci-lint-action@v6
         with:
           args: --timeout 10m0s -v
-      - run: go test ./...
+
+      - name: go test
+        # Do not skip tests in the face of linter errors, or 'go mod tidy'
+        # checks, which are secondary to the tests. Tests trump linters.
+        if: always()
+        run: go test -count=1 -benchtime=1x ./...
+
+      # It is tempting to run this in a platform independent way, but the past
+      # shows this codebase will see introductions of platform specific code
+      # generation, and so we need to check this per platform to ensure we
+      # don't abuse go generate on specific platforms.
+      - name: check that 'go generate' is clean
+        run: |
+          go generate ./...
+          git diff --name-only --exit-code || (echo "Please run 'go generate ./...'." && exit 1)
+
+      - name: cache save
+        # Always save the cache, even if the job fails. The artifacts produced
+        # during the building of test binaries are not all for naught. They can
+        # be used to speed up subsequent runs.
+        if: always()
+
+        uses: actions/cache/save@1bd1e32a3bdc45362d1e726936510720a7c30a57 # v4.2.0
+        with:
+          # Note: unlike the other setups, this is only grabbing the mod download
+          # cache, rather than the whole mod directory, as the download cache
+          # contains zips that can be unpacked in parallel faster than they can be
+          # fetched and extracted by tar
+          path: |
+            ~/.cache/go-build
+            ~/go/pkg/mod/cache
+            ~\AppData\Local\go-build
+          # NOTE: The -3- here should be incremented when the scheme of data to be
+          # cached changes (e.g. path above changes).
+          key: ${{ github.job }}-${{ runner.os }}-${{ matrix.goarch }}-${{ matrix.buildflags }}-go-3-${{ hashFiles('**/go.sum') }}-${{ github.run_id }}
 
   patches:
     runs-on: ubuntu-latest

ml/backend/ggml/ggml/src/ggml-metal/ggml-metal-embed.metal

@@ -1,4 +1,4 @@
-// Code generated Fri Jan 10 13:05:45 PST 2025. DO NOT EDIT.
+// Code generated by go generate. DO NOT EDIT.
 #define GGML_COMMON_DECL_METAL
 #define GGML_COMMON_IMPL_METAL
 #if defined(GGML_METAL_EMBED_LIBRARY)

ml/backend/ggml/ggml/src/ggml-metal/metal.go

@@ -2,7 +2,7 @@
 
 package metal
 
-//go:generate sh -c "{ echo // Code generated $(date). DO NOT EDIT.; sed -e '/__embed_ggml-common.h__/r ../ggml-common.h' -e '/__embed_ggml-common.h__/d' -e '/#include \"ggml-metal-impl.h\"/r ggml-metal-impl.h' -e '/#include \"ggml-metal-impl.h\"/d' ggml-metal.metal; } >ggml-metal-embed.metal"
+//go:generate sh -c "{ echo // Code generated by 'go generate'. DO NOT EDIT.; sed -e '/__embed_ggml-common.h__/r ../ggml-common.h' -e '/__embed_ggml-common.h__/d' -e '/#include \"ggml-metal-impl.h\"/r ggml-metal-impl.h' -e '/#include \"ggml-metal-impl.h\"/d' ggml-metal.metal; } >ggml-metal-embed.metal"
 
 // #cgo CPPFLAGS: -DGGML_METAL_EMBED_LIBRARY -I.. -I../../include
 // #cgo LDFLAGS: -framework Metal -framework MetalKit