cl,goflags: implement maymorestack debug hook

[cpunion]

Summary

• parse the supported unpatterned and all= forms of -gcflags=-d=maymorestack=<hook> in internal/goflags
• carry the hook through typed build configuration and the per-package frontend cl.Options, without process-wide compiler state
• instrument ordinary function entries while excluding synthetic and //go:nosplit functions
• preserve Go package scope: unpatterned flags affect command-line roots, while all= affects dependencies too
• include the effective per-package hook in archive cache fingerprints
• remove the six Darwin/Linux xfails for Go 1.24, 1.25, and 1.26

Root cause

The runner already forwarded -gcflags, but LLGo did not translate the maymorestack debug option into frontend behavior. The old local prototype predated internal/goflags and used process-wide state plus repeated command wiring. This version reuses the current Go-compatible flag path and the frontend option isolation added on current main.

The official exact-count test also guards package scope: applying an unpatterned flag to dependencies would produce more than the expected 128 calls. Unit coverage adds the separate -d spelling, comma-separated debug options, last-list clearing, all= scope, //go:nosplit, cache isolation, and consecutive compilation isolation.

Validation

• go test ./internal/goflags -count=1
• go test -race ./internal/goflags -run "MayMoreStack|FrontendGCFlag" -count=1
• focused cl and internal/build maymorestack tests
• go test ./cmd/internal/flags ./cmd/internal/build ./cmd/internal/run ./cmd/internal/test ./cmd/internal/install -count=1
• project integration TestMayMoreStackGCFlag
• official test/maymorestack.go with an empty xfail set on Go 1.24.11, 1.25.7, and 1.26.5
• git diff --check upstream/main...HEAD

No CI timing/configuration, wasm, DWARF, PCLN, pthread, TLS, or unrelated runtime behavior is changed.

[fennoai]

Review: maymorestack debug hook

Solid, well-scoped implementation. The -d=maymorestack= flag parsing, all= vs. command-line package scoping, fingerprint/manifest integration, and eligibility logic (skip synthetic / //go:nosplit / self-recursion) all track Go's real maymorestack semantics, and test coverage is thorough (cl internal, build-scope, fingerprint, goflags matrix, and an end-to-end goroot-style test). Removing the maymorestack.go xfail entries is well-supported by the new tests. Findings below are minor; the two worth addressing before merge are the nosplit-match brittleness and the doc-comment attachment.

Findings not tied to a diff line:

• doc/design/pclntab-packaging.md:92 — the design doc states the -gcflags frontend subset is "(-lang, -N, and -l)". This PR adds -d=maymorestack= as a fourth frontend-interpreted gcflag mapped to typed Config fields, so that enumeration is now out of date and should include it. (File is outside the PR diff, so noted here.)

Inline comments cover the remaining findings.

[cpunion]

Addressed all current review findings in 4ed2c1fa3:

• fast-return before syntax scanning when maymorestack is disabled
• directive-parser-based //go:nosplit detection, with a trailing-space counterexample
• corrected MayMoreStackAll field documentation
• updated the frontend gcflags design-doc list
• removed the local variable shadowing

Focused validation passes:

• go test ./cl -run ^TestMayMoreStack -count=1
• go test ./internal/directive ./internal/goflags -count=1
• go test ./internal/build -run ^(TestMayMoreStackHookPackageScope|TestManifestBuilder_MayMoreStack)$ -count=1

A separate full ./internal/build attempt reached its existing 10-minute package timeout in TestStandardDWARF/Oz; the focused changed-path tests above completed successfully.

[codecov]

Codecov Report

❌ Patch coverage is 94.44444% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
cl/compile.go	85.71%	2 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

LLGo baseline benchmarks

4ed2c1fa3723 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs main	Build	vs main	Run	vs main
Linux	cprintf	18544 B	new	353.314 ms	new	1.367 ms	new
Linux	fmtprintf	2217608 B	new	3.323 s	new	2.710 ms	new
Linux	println	71504 B	new	347.879 ms	new	1.715 ms	new
macOS	cprintf	84672 B	new	398.096 ms	new	2.955 ms	new
macOS	fmtprintf	2361520 B	new	2.937 s	new	20.373 ms	new
macOS	println	125712 B	new	347.819 ms	new	3.811 ms	new

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	12.280 ns/op	new
Linux	BenchmarkMergeCompilerFlags	144.600 ns/op	new
Linux	BenchmarkMergeLinkerFlags	93.760 ns/op	new
Linux	BenchmarkChannelBuffered	36.310 ns/op	new
Linux	BenchmarkChannelHandoff	23949 ns/op	new
Linux	BenchmarkDefer	42.720 ns/op	new
Linux	BenchmarkDirectCall	1.757 ns/op	new
Linux	BenchmarkGlobalRead	1.761 ns/op	new
Linux	BenchmarkGlobalWrite	2.806 ns/op	new
Linux	BenchmarkGoroutine	30788 ns/op	new
Linux	BenchmarkInterfaceCall	8.801 ns/op	new
Linux	BenchmarkRuntimeGetG	1.759 ns/op	new
macOS	BenchmarkLookupPCRandom	11.490 ns/op	new
macOS	BenchmarkMergeCompilerFlags	105.600 ns/op	new
macOS	BenchmarkMergeLinkerFlags	78.450 ns/op	new
macOS	BenchmarkChannelBuffered	22.330 ns/op	new
macOS	BenchmarkChannelHandoff	6943 ns/op	new
macOS	BenchmarkDefer	27.770 ns/op	new
macOS	BenchmarkDirectCall	1.017 ns/op	new
macOS	BenchmarkGlobalRead	1.028 ns/op	new
macOS	BenchmarkGlobalWrite	1.017 ns/op	new
macOS	BenchmarkGoroutine	23946 ns/op	new
macOS	BenchmarkInterfaceCall	4.406 ns/op	new
macOS	BenchmarkRuntimeGetG	2.038 ns/op	new

No main baseline exists yet; all metrics are marked new.

Warning

• The rendered benchmark data could not be pushed.