x/tools/gopls/internal/test/integration/workspace: TestModuleWithExclude/default failures

[gopherbot]

#!watchflakes
default <- pkg == "golang.org/x/tools/gopls/internal/test/integration/workspace" && test == "TestModuleWithExclude/default"

Issue created automatically to collect these failures.

Example (log):

=== RUN   TestModuleWithExclude/default
panic: test timed out after 20m0s
	running tests:
		TestModuleWithExclude (19m24s)
		TestModuleWithExclude/default (19m24s)

goroutine 53618 [running]:
testing.(*M).startAlarm.func1()
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2802 +0x494
created by time.goFunc
...
golang.org/x/tools/gopls/internal/test/integration/fake.(*Sandbox).Close(0xc00a4ac600)
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/fake/sandbox.go:278 +0x110
golang.org/x/tools/gopls/internal/test/integration.(*Runner).Run.func1.1()
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/runner.go:196 +0x5c
golang.org/x/tools/gopls/internal/test/integration.(*Runner).Run.func1(0xc00a4b0008)
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/runner.go:250 +0xd5c
testing.tRunner(0xc00a4b0008, 0xc00b60dea0)
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2036 +0x168
created by testing.(*T).Run in goroutine 53387
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2101 +0x7c0

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "golang.org/x/tools/gopls/internal/test/integration/workspace" && test == "TestModuleWithExclude/default"

2026-04-09 13:34 x_tools-go1.26-darwin-arm64-race tools@0ae2de02 release-branch.go1.26@9c8bf0e7 x/tools/gopls/internal/test/integration/workspace.TestModuleWithExclude/default [SKIP] (log)

=== RUN   TestModuleWithExclude/default
panic: test timed out after 20m0s
	running tests:
		TestModuleWithExclude (19m24s)
		TestModuleWithExclude/default (19m24s)

goroutine 53618 [running]:
testing.(*M).startAlarm.func1()
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2802 +0x494
created by time.goFunc
...
golang.org/x/tools/gopls/internal/test/integration/fake.(*Sandbox).Close(0xc00a4ac600)
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/fake/sandbox.go:278 +0x110
golang.org/x/tools/gopls/internal/test/integration.(*Runner).Run.func1.1()
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/runner.go:196 +0x5c
golang.org/x/tools/gopls/internal/test/integration.(*Runner).Run.func1(0xc00a4b0008)
	/Users/swarming/.swarming/w/ir/x/w/targetrepo3096366441/gopls/internal/test/integration/runner.go:250 +0xd5c
testing.tRunner(0xc00a4b0008, 0xc00b60dea0)
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2036 +0x168
created by testing.(*T).Run in goroutine 53387
	/Users/swarming/.swarming/w/ir/x/w/goroot/src/testing/testing.go:2101 +0x7c0

— watchflakes

[spqr79]

The stack shows Sandbox.Close at sandbox.go:278 is where the goroutine dump was captured, but that's the watchdog firing after 20 minutes — the actual block happened earlier. Without a goroutine dump from inside the hang, we're guessing at the root cause.

Before any fix, we need visibility. Adding a runtime/pprof goroutine dump to the timeout handler in runner.go around line 196–250 would tell us exactly what's blocked and on what. Something like:

// in the cleanup/timeout path
pprof.Lookup("goroutine").WriteTo(os.Stderr, 1)

That's the one thing that makes everything else non-speculative.

The darwin-arm64-race combination is suspicious — FSEvents callbacks on Darwin can be delayed enough under -race scheduling to starve whatever gopls is waiting on after processing the exclude directive. If TestModuleWithExclude is doing an AwaitDiagnostics (or similar) after applying the exclude config, that await may not have its own deadline separate from the 20m test timeout, which would explain why it only surfaces as a flake under race on arm64.

Two concrete things worth checking in workspace/workspace_test.go:

1. Does the test have an explicit timeout on the diagnostic/file-watcher await after the exclude is applied, or does it rely purely on the 20m ceiling?
2. Does runner.go scale the timeout under -race? If the 20m is a hardcoded constant without a race-aware multiplier, that's low-hanging fruit while the real fix is being diagnosed.

The Sandbox.Close signature change (adding context propagation down from Runner.Run) makes sense as a follow-up, but it won't fix the hang — it'll just make the timeout cleaner. The goroutine dump needs to land first, on tools@0ae2de02 or a patch on top, so the next time this reproduces on x_tools-go1.26-darwin-arm64-race we get actionable output.

[gopherbot]

Change https://go.dev/cl/765306 mentions this issue: gopls/internal/test/marker: send SIGABRT to stuck go commands