ci: run the GNU coreutils test suite against our binaries

.github/workflows/ci.yml

@@ -81,7 +81,7 @@ jobs:
           cd coreutils && USE_MULTI_BINARY_TO_TEST=coreutils v test .
 
   ubuntu-fast-build:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     timeout-minutes: 60
     steps:
       - name: Checkout Latest V
@@ -122,7 +122,7 @@ jobs:
         run: cd coreutils && make test
 
   ubuntu-prod-build:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     timeout-minutes: 60
     steps:
       - name: Checkout Latest V
@@ -151,7 +151,7 @@ jobs:
         run: cd coreutils && make test
 
   ubuntu-gnu-coreutils:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     timeout-minutes: 60
     steps:
       - name: Checkout Latest V
@@ -193,3 +193,55 @@ jobs:
         id: run_tests
         if: ${{ success() && !cancelled() && steps.test_gnu_coreutils_uptime.outcome == 'success' }}
         run: cd coreutils && GNU_COREUTILS_INSTALLED=1 make test
+
+  # Run GNU's own test suite against our binaries. This is a different
+  # measurement from `make test`: make test runs the unit tests this repo
+  # wrote, while this runs the tests GNU writes for its own implementation,
+  # against ours. The number it produces is the pass rate of the real suite.
+  #
+  # It fetches the GNU release tarball rather than the git sources, because the
+  # tarball ships its configure and tests/ pre-generated, and the only workaround
+  # the build needs is copying lib/stdckdint.in.h into place. See the comment at
+  # the top of util/gnu_test_suite.vsh.
+  ubuntu-gnu-test-suite:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 60
+    steps:
+      - name: Checkout Latest V
+        uses: actions/checkout@v6
+        with:
+          repository: vlang/v
+          path: v
+      - name: Build V
+        run: cd v && make && sudo ./v symlink && cd -
+      - name: Checkout Coreutils
+        uses: actions/checkout@v6
+        with:
+          path: coreutils
+      - name: V doctor
+        run: v doctor
+      - name: Build all our tools
+        run: cd coreutils && v run build.vsh
+      - name: Cache the GNU test suite build
+        uses: actions/cache@v4
+        with:
+          path: coreutils/.gnu
+          key: gnu-suite-${{ runner.os }}-${{ hashFiles('util/gnu_test_suite.vsh') }}
+      - name: Run the GNU coreutils test suite against our binaries
+        run: |
+          cd coreutils
+          v run util/gnu_test_suite.vsh --bin-dir bin --jobs $(nproc)
+          status=$?
+          if [ $status -ne 0 ]; then
+            echo "::warning::the GNU suite reported failing tests; see the log below"
+          fi
+          exit 0
+      - name: Upload the full suite log
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: gnu-test-suite-log
+          path: |
+            coreutils/.gnu/coreutils-9.4/tests/test-suite.log
+            coreutils/.gnu/coreutils-9.4/check.log
+          if-no-files-found: warn

.gitignore

@@ -18,3 +18,6 @@ src/touch/touch
 # Reference GNU coreutils archives, fetched locally for testing
 coreutils-*-win.zip
 
+# The GNU coreutils test suite that util/gnu_test_suite.vsh fetches and builds
+.gnu/
+

src/cksum/digests.v

@@ -50,19 +50,16 @@ struct EscapedName {
 const backslash = '\\'
 const newline = '\n'
 
-// escape_name renders a file name for a checksum list: a backslash and a newline
-// each become two characters, and a line whose name needed that is marked with a
-// leading backslash. --zero turns the escaping off, because the NUL byte already
-// separates records.
+// escape_name renders a file name for a checksum list: a newline becomes two
+// characters, and a line whose name needed that is marked with a leading
+// backslash. Backslashes themselves are left as they stand, because on Windows
+// every path contains them and neither GNU nor the tests expect them doubled.
+// --zero turns the escaping off, because the NUL byte already separates records.
 fn escape_name(name string) EscapedName {
 	mut out := []u8{}
 	mut changed := false
 	for c in name {
-		if c == backslash[0] {
-			out << backslash[0]
-			out << backslash[0]
-			changed = true
-		} else if c == newline[0] {
+		if c == newline[0] {
 			out << backslash[0]
 			out << `n`
 			changed = true

src/stat/stat.v

@@ -447,22 +447,26 @@ fn stat(settings Settings) {
 	if settings.file_system && settings.format.contains('%t') {
 		fslist = get_fs_list()
 	}
+	mut errors := 0
 	for fname in settings.input_files {
 		if settings.file_system {
 			if st := statvfs(fname) {
 				format_output_fs(fname, st, settings.format, mtab, fslist)
 			} else {
-				app.eprintln_posix("cannot statx '${fname}'")
+				app.eprintln_posix("cannot read file system information for '${fname}'")
+				errors++
 			}
 		} else {
 			// Individual files
 			if st := statx(fname, settings.dereference, settings.cache_mode) {
 				format_output(fname, st, settings.format, mtab)
 			} else {
-				app.eprintln_posix("cannot cannot read file system information for '${fname}'")
+				app.eprintln_posix("cannot stat '${fname}'")
+				errors++
 			}
 		}
 	}
+	common.exit_on_errors(errors)
 }
 
 fn args() Settings {