debug: consolidate native DWARF correctness and defaults

[cpunion]

Native builds omit DWARF by default, and explicit debug builds can lose current parameter/local values or reliable Darwin source locations. This PR preserves native DWARF by default and fixes those correctness issues while keeping optimized C ABI parameter-home reuse.

Based directly on main f06143ba2, with no unmerged prerequisites. Implements the native DWARF part of #2154 and #2164; native packaging follows the platform toolchain.

• Preserve DWARF for native executables and archives, with cmd/link's Darwin c-shared exception. Explicit -w/-w=false takes precedence; ordinary Wasm/embedded defaults remain unchanged.
• Keep stable O0 parameter homes and current values after assignments. Protect O0 aggregate declarations while retaining O2 parameter-home reuse. Addressable locals retain their real storage at every optimization level.
• Allocate aggregate debug snapshots once per function, avoiding per-iteration O0 stack growth. Keep map/channel debug locations as pointer values and size map parameter homes as Go values.
• Fix Darwin PC-line anchors and persistent cold/warm debug-map archive paths.
• Remove replaced Cgo function-address placeholders and their debug indexes, preventing undefined COFF DWARF relocations. Revisited or absent placeholders no longer create unused C declarations.
• Preserve current C ABI/module-flag behavior and finite reflection-method reachability when LLVM folds returned string components independently.
• Add source-debug acceptance under test/debug/native, including exact panic/divide/nil caller locations, O2 inline frames/step-over, and current aggregate local/parameter values after pointer mutation.

Consolidates #2235, #2148, #2157 and #2202 plus the relevant native acceptance from cpunion#140.

Fixes #2206.
Fixes #2115.
Fixes #2119.

Validation for 08d0aef2e:

• Full C ABI and SSA suites pass; focused CL debug/home and Cgo/GenExtern/Objective-C tests pass.
• New regressions reproduce the old O2 reuse, competing local snapshot, and missing Cgo-placeholder failures.
• Actual Apple LLDB native acceptance passes, including the new O2 aggregate assertions and existing panic/inline/step-over checks.
• The four-contribution combined tree merges without conflicts and passes eight debugger/ABI/browser/SSA Go packages and workflow syntax checks.
• Earlier platform CI qualified the Windows COFF/Cgo fixes and the native default/explicit-DWARF policy. Fresh PR CI is running for this revision.

The dedicated panic/inline suite is currently a Linux/macOS CI gate. Windows retains runtime LLDB acceptance and TestDWARFPCLNLineSites, but those do not establish the separate panic/inline guarantees; this limit is documented in the native suite README.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

❌ Patch coverage is 98.19820% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/build/cgo.go	80.00%	1 Missing ⚠️
ssa/decl.go	90.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[fennoai]

Review: match cmd/link DWARF defaults

This is a well-structured change. Making DWARF preserved by default is cleanly factored in internal/build/link_options.go, the -s implies -w precedence is order-independent and matches cmd/link, and the ripple effect of DWARF now being on by default (SSA DebugRefs present) is handled consistently through nonDebugReferrers and explicit *ssa.DebugRef handling. Test coverage of the new policy branches is thorough. Security is fine — the acceptance tests use t.TempDir() and constant source templates, and defaulting DWARF on matches Go's own cmd/link behavior.

A few findings below; the inline comment is the only code-level item, the rest are stale docs.

Findings

1. Stale design doc — DWARF default is now inverted
doc/design/pclntab-packaging.md:28-64 still documents the removed policy: "LLGo temporarily omits DWARF by default … LLGo currently behaves as if -w=true", ModeGen framing, and "Broad CI uses the DWARF-free default". After this PR, linked builds emit DWARF unless -w/-s is passed (only Darwin c-shared defaults to omit). This section should be rewritten to the Go-standard default; otherwise the primary design doc for this feature contradicts the shipped behavior.

2. Stale design doc — Darwin builds are no longer "site-free"
doc/design/pclntab-packaging.md:138-142 still says "Darwin embedded builds that emit DWARF keep the historical site-free path". This PR splits site emission: shouldEmitRuntimePCLineSites no longer consults DWARF (Darwin DWARF builds now do emit PC-line sites), and only shouldEmitRuntimeAddressSites suppresses entry/stub address sites on embedded Darwin DWARF. The in-code comments (build.go, funcinfo_table.go) are correct; only this doc is stale. Update it to say PC-line sites are retained and only address sites are suppressed.

See the inline comment for the one code-level note.

[cpunion]

Review follow-up after restacking on the updated #2157:

• doc/design/pclntab-packaging.md now describes DWARF-preserving linked defaults, -s/-w precedence, and Darwin's PC-line versus function-entry address-site policy;
• the moveInstrsAfter precondition guard and tests are inherited from final debug: keep optimization and SSA rewrites DWARF-safe #2143;
• defer debug locations were integrated with the existing native/Wasm selector dispatch instead of replacing it.

The focused tests and go test ./... pass.

[github-actions]

LLGo baseline benchmarks

08d0aef2e9e9 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs base	Text size	vs base	Build	vs base	Run	vs base
Linux	cprintf	8488 B	+1328 B / +18.5% (worse)	387 B	0 B / +0.0%	923.376 ms	+378.8 ms / +69.6% (worse)	1.231 ms	-99.51 us / -7.5% (better)
Linux	cprintf-lto	8312 B	+1400 B / +20.3% (worse)	368 B	0 B / +0.0%	976.811 ms	+441.5 ms / +82.5% (worse)	1.320 ms	+39.13 us / +3.1% (worse)
Linux	fmtprintf	4844792 B	+3181368 B / +191.3% (worse)	501297 B	+2860 B / +0.6% (worse)	7.150 s	+3.606 s / +101.7% (worse)	3.019 ms	+68.27 us / +2.3% (worse)
Linux	fmtprintf-lto	3574864 B	+2073696 B / +138.1% (worse)	438141 B	+2018 B / +0.5% (worse)	16.176 s	+5.553 s / +52.3% (worse)	2.908 ms	-74.96 us / -2.5% (better)
Linux	println	673736 B	+604488 B / +872.9% (worse)	16823 B	+18 B / +0.1% (worse)	976.395 ms	+429.5 ms / +78.5% (worse)	1.656 ms	-12.66 us / -0.8% (better)
Linux	println-lto	189472 B	+129496 B / +215.9% (worse)	14231 B	+22 B / +0.2% (worse)	1.263 s	+471.3 ms / +59.6% (worse)	1.707 ms	+76.9 us / +4.7% (worse)
macOS	cprintf	50672 B	-17392 B / -25.6% (better)	4378 B	-51 B / -1.2% (better)	1.097 s	+187.9 ms / +20.7% (worse)	3.429 ms	+576.5 us / +20.2% (worse)
macOS	cprintf-lto	50432 B	-33856 B / -40.2% (better)	130 B	-63 B / -32.6% (better)	960.420 ms	-33.26 ms / -3.3% (better)	2.672 ms	-570.4 us / -17.6% (better)
macOS	fmtprintf	1773280 B	+268448 B / +17.8% (worse)	876696 B	+1212 B / +0.1% (worse)	6.462 s	+2.584 s / +66.6% (worse)	11.440 ms	+5.913 ms / +107.0% (worse)
macOS	fmtprintf-lto	1360752 B	+168048 B / +14.1% (worse)	761868 B	-87112 B / -10.3% (better)	15.895 s	+5.97 s / +60.2% (worse)	6.635 ms	-1.081 ms / -14.0% (better)
macOS	println	99136 B	-18080 B / -15.4% (better)	24149 B	-13416 B / -35.7% (better)	880.343 ms	+104.2 ms / +13.4% (worse)	4.015 ms	-246.5 us / -5.8% (better)
macOS	println-lto	83488 B	-35984 B / -30.1% (better)	21400 B	-13544 B / -38.8% (better)	1.122 s	+46.24 ms / +4.3% (worse)	4.824 ms	+570.2 us / +13.4% (worse)
Windows MinGW	cprintf	649728 B	+630272 B / +3239.5% (worse)	4550 B	0 B / +0.0%	1.804 s	+424.1 ms / +30.7% (worse)	4.078 ms	+52.4 us / +1.3% (worse)
Windows MinGW	cprintf-lto	43520 B	+25600 B / +142.9% (worse)	4486 B	0 B / +0.0%	1.870 s	+459.3 ms / +32.6% (worse)	4.129 ms	+230.5 us / +5.9% (worse)
Windows MinGW	fmtprintf	5383168 B	+3449856 B / +178.4% (worse)	599990 B	+960 B / +0.2% (worse)	7.873 s	+3.567 s / +82.9% (worse)	9.638 ms	+1.067 ms / +12.5% (worse)
Windows MinGW	fmtprintf-lto	4092928 B	+2135040 B / +109.0% (worse)	546230 B	-1088 B / -0.2% (better)	15.662 s	+4.989 s / +46.7% (worse)	9.014 ms	-66 us / -0.7% (better)
Windows MinGW	println	705024 B	+629248 B / +830.4% (worse)	25174 B	+32 B / +0.1% (worse)	1.817 s	+464.4 ms / +34.3% (worse)	7.610 ms	+631.4 us / +9.0% (worse)
Windows MinGW	println-lto	208896 B	+139776 B / +202.2% (worse)	22038 B	+48 B / +0.2% (worse)	2.101 s	+504.9 ms / +31.6% (worse)	8.503 ms	+1.68 ms / +24.6% (worse)
Windows MinGW 386	cprintf	600576 B	+557056 B / +1280.0% (worse)	5326 B	0 B / +0.0%	1.879 s	+493.3 ms / +35.6% (worse)	5.807 ms	+60 us / +1.0% (worse)
Windows MinGW 386	cprintf-lto	103424 B	+82432 B / +392.7% (worse)	5094 B	0 B / +0.0%	1.891 s	+436.9 ms / +30.0% (worse)	6.334 ms	+541.9 us / +9.4% (worse)
Windows MinGW 386	fmtprintf	4694528 B	+2798080 B / +147.5% (worse)	472286 B	-176 B / -0.03725% (better)	7.944 s	+3.439 s / +76.4% (worse)	11.769 ms	-395 us / -3.2% (better)
Windows MinGW 386	fmtprintf-lto	4115456 B	+1934336 B / +88.7% (worse)	450806 B	-468 B / -0.1% (better)	15.604 s	+5.029 s / +47.6% (worse)	12.030 ms	+328.2 us / +2.8% (worse)
Windows MinGW 386	println	651776 B	+555520 B / +577.1% (worse)	21474 B	0 B / +0.0%	1.884 s	+477.1 ms / +33.9% (worse)	10.616 ms	+905.5 us / +9.3% (worse)
Windows MinGW 386	println-lto	257536 B	+183296 B / +246.9% (worse)	19298 B	-36 B / -0.2% (better)	2.179 s	+535.1 ms / +32.5% (worse)	10.459 ms	+24.1 us / +0.2% (worse)
Windows MinGW ARM64	cprintf	659968 B	+641024 B / +3383.8% (worse)	4408 B	0 B / +0.0%	1.960 s	+354 ms / +22.1% (worse)	6.562 ms	-509.3 us / -7.2% (better)
Windows MinGW ARM64	cprintf-lto	43520 B	+25600 B / +142.9% (worse)	4340 B	0 B / +0.0%	1.973 s	+333.5 ms / +20.3% (worse)	6.943 ms	-267 us / -3.7% (better)
Windows MinGW ARM64	fmtprintf	5294080 B	+3474432 B / +190.9% (worse)	510680 B	+964 B / +0.2% (worse)	7.235 s	+2.848 s / +64.9% (worse)	13.272 ms	-34.1 us / -0.3% (better)
Windows MinGW ARM64	fmtprintf-lto	4267520 B	+2387456 B / +127.0% (worse)	476932 B	+756 B / +0.2% (worse)	13.971 s	+4.112 s / +41.7% (worse)	12.756 ms	-1.125 ms / -8.1% (better)
Windows MinGW ARM64	println	712192 B	+640000 B / +886.5% (worse)	23876 B	-8 B / -0.0335% (better)	1.962 s	+338.5 ms / +20.8% (worse)	11.922 ms	+48.4 us / +0.4% (worse)
Windows MinGW ARM64	println-lto	215552 B	+146944 B / +214.2% (worse)	21224 B	-8 B / -0.03768% (better)	2.216 s	+385.4 ms / +21.1% (worse)	11.975 ms	-245.9 us / -2.0% (better)
Windows MSVC	cprintf	891904 B	+771584 B / +641.3% (worse)	65798 B	0 B / +0.0%	1.365 s	+465.2 ms / +51.7% (worse)	2.392 ms	+66.8 us / +2.9% (worse)
Windows MSVC	cprintf-lto	289280 B	+169472 B / +141.5% (worse)	65734 B	0 B / +0.0%	1.233 s	+381.8 ms / +44.9% (worse)	2.799 ms	+486.1 us / +21.0% (worse)
Windows MSVC	fmtprintf	5683200 B	+4039680 B / +245.8% (worse)	695542 B	+960 B / +0.1% (worse)	5.399 s	+2.405 s / +80.3% (worse)	6.830 ms	-174.4 us / -2.5% (better)
Windows MSVC	fmtprintf-lto	4417024 B	+2781696 B / +170.1% (worse)	645782 B	-1296 B / -0.2% (better)	11.009 s	+3.691 s / +50.4% (worse)	6.936 ms	-52.7 us / -0.8% (better)
Windows MSVC	println	1016832 B	+822272 B / +422.6% (worse)	120838 B	+16 B / +0.01324% (worse)	1.214 s	+368.3 ms / +43.6% (worse)	5.748 ms	-20 us / -0.3% (better)
Windows MSVC	println-lto	528384 B	+335872 B / +174.5% (worse)	118390 B	+32 B / +0.02704% (worse)	1.400 s	+274.6 ms / +24.4% (worse)	5.708 ms	-786.5 us / -12.1% (better)
Windows MSVC 386	cprintf	512512 B	+502784 B / +5168.4% (worse)	3931 B	0 B / +0.0%	1.294 s	+227.6 ms / +21.4% (worse)	3.888 ms	+137.4 us / +3.7% (worse)
Windows MSVC 386	cprintf-lto	44032 B	+34816 B / +377.8% (worse)	3853 B	0 B / +0.0%	1.172 s	+292.5 ms / +33.2% (worse)	3.643 ms	+169.6 us / +4.9% (worse)
Windows MSVC 386	fmtprintf	4428800 B	+3224064 B / +267.6% (worse)	455676 B	-176 B / -0.03861% (better)	6.863 s	+3.741 s / +119.8% (worse)	8.092 ms	+594.5 us / +7.9% (worse)
Windows MSVC 386	fmtprintf-lto	3865600 B	+2624000 B / +211.3% (worse)	426379 B	-832 B / -0.2% (better)	10.994 s	+4.658 s / +73.5% (worse)	7.644 ms	-195.6 us / -2.5% (better)
Windows MSVC 386	println	565248 B	+528896 B / +1454.9% (worse)	20324 B	0 B / +0.0%	1.309 s	+467.1 ms / +55.5% (worse)	6.163 ms	-1.36 ms / -18.1% (better)
Windows MSVC 386	println-lto	197632 B	+161792 B / +451.4% (worse)	18485 B	-80 B / -0.4% (better)	1.859 s	+870.2 ms / +88.0% (worse)	8.142 ms	+1.654 ms / +25.5% (worse)
Windows MSVC ARM64	cprintf	661504 B	+649728 B / +5517.4% (worse)	4192 B	0 B / +0.0%	1.583 s	+364.4 ms / +29.9% (worse)	6.906 ms	-424.5 us / -5.8% (better)
Windows MSVC ARM64	cprintf-lto	47616 B	+36864 B / +342.9% (worse)	4084 B	0 B / +0.0%	1.590 s	+335.9 ms / +26.8% (worse)	6.691 ms	-243.5 us / -3.5% (better)
Windows MSVC ARM64	fmtprintf	5290496 B	+3903488 B / +281.4% (worse)	510616 B	+960 B / +0.2% (worse)	6.666 s	+2.761 s / +70.7% (worse)	13.376 ms	-1.109 ms / -7.7% (better)
Windows MSVC ARM64	fmtprintf-lto	4274176 B	+2868736 B / +204.1% (worse)	477604 B	+768 B / +0.2% (worse)	12.652 s	+3.621 s / +40.1% (worse)	13.427 ms	-1.031 ms / -7.1% (better)
Windows MSVC ARM64	println	714240 B	+669184 B / +1485.2% (worse)	23908 B	0 B / +0.0%	1.560 s	+310.3 ms / +24.8% (worse)	11.949 ms	-321.7 us / -2.6% (better)
Windows MSVC ARM64	println-lto	220160 B	+177664 B / +418.1% (worse)	21380 B	0 B / +0.0%	1.792 s	+371 ms / +26.1% (worse)	11.707 ms	-1.118 ms / -8.7% (better)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs base
Linux	BenchmarkLookupPCRandom	14.540 ns/op	-0.17 ns/op / -1.2% (better)
Linux	BenchmarkMergeCompilerFlags	198.400 ns/op	+4.6 ns/op / +2.4% (worse)
Linux	BenchmarkMergeLinkerFlags	137.700 ns/op	+11.4 ns/op / +9.0% (worse)
Linux	BenchmarkChannelBuffered	55.360 ns/op	-0.03 ns/op / -0.1% (better)
Linux	BenchmarkChannelHandoff	13503 ns/op	+480 ns/op / +3.7% (worse)
Linux	BenchmarkDefer	47.950 ns/op	+0.97 ns/op / +2.1% (worse)
Linux	BenchmarkDirectCall	1.515 ns/op	-0.07 ns/op / -4.4% (better)
Linux	BenchmarkGlobalRead	1.554 ns/op	+0.367 ns/op / +30.9% (worse)
Linux	BenchmarkGlobalWrite	7.757 ns/op	-0.1 ns/op / -1.3% (better)
Linux	BenchmarkGoroutine	24149 ns/op	-4986 ns/op / -17.1% (better)
Linux	BenchmarkInterfaceCall	6.261 ns/op	+0.393 ns/op / +6.7% (worse)
Linux	BenchmarkRuntimeGetG	2.976 ns/op	-0.027 ns/op / -0.9% (better)
macOS	BenchmarkLookupPCRandom	17.490 ns/op	+1.31 ns/op / +8.1% (worse)
macOS	BenchmarkMergeCompilerFlags	145.600 ns/op	+22 ns/op / +17.8% (worse)
macOS	BenchmarkMergeLinkerFlags	94.550 ns/op	+9.55 ns/op / +11.2% (worse)
macOS	BenchmarkChannelBuffered	33.910 ns/op	+5.47 ns/op / +19.2% (worse)
macOS	BenchmarkChannelHandoff	13346 ns/op	+3422 ns/op / +34.5% (worse)
macOS	BenchmarkDefer	43.700 ns/op	-5.09 ns/op / -10.4% (better)
macOS	BenchmarkDirectCall	1.194 ns/op	-0.227 ns/op / -16.0% (better)
macOS	BenchmarkGlobalRead	1.185 ns/op	-0.026 ns/op / -2.1% (better)
macOS	BenchmarkGlobalWrite	1.388 ns/op	-0.123 ns/op / -8.1% (better)
macOS	BenchmarkGoroutine	50452 ns/op	+10041 ns/op / +24.8% (worse)
macOS	BenchmarkInterfaceCall	4.884 ns/op	-0.195 ns/op / -3.8% (better)
macOS	BenchmarkRuntimeGetG	2.565 ns/op	-1.243 ns/op / -32.6% (better)
Windows MinGW	BenchmarkLookupPCRandom	13.160 ns/op	+0.01 ns/op / +0.1% (worse)
Windows MinGW	BenchmarkMergeCompilerFlags	637.700 ns/op	+13.7 ns/op / +2.2% (worse)
Windows MinGW	BenchmarkMergeLinkerFlags	553.400 ns/op	+10.7 ns/op / +2.0% (worse)
Windows MinGW	BenchmarkChannelBuffered	30.750 ns/op	-1.18 ns/op / -3.7% (better)
Windows MinGW	BenchmarkChannelHandoff	977.100 ns/op	-19.9 ns/op / -2.0% (better)
Windows MinGW	BenchmarkDefer	59.530 ns/op	+0.4 ns/op / +0.7% (worse)
Windows MinGW	BenchmarkDirectCall	1.550 ns/op	+0.003 ns/op / +0.2% (worse)
Windows MinGW	BenchmarkGlobalRead	1.548 ns/op	-0.312 ns/op / -16.8% (better)
Windows MinGW	BenchmarkGlobalWrite	2.471 ns/op	+0.012 ns/op / +0.5% (worse)
Windows MinGW	BenchmarkGoroutine	94889 ns/op	-8436 ns/op / -8.2% (better)
Windows MinGW	BenchmarkInterfaceCall	8.403 ns/op	+0.327 ns/op / +4.0% (worse)
Windows MinGW	BenchmarkRuntimeGetG	2.484 ns/op	+0.309 ns/op / +14.2% (worse)
Windows MinGW 386	BenchmarkLookupPCRandom	26.660 ns/op	+0.04 ns/op / +0.2% (worse)
Windows MinGW 386	BenchmarkMergeCompilerFlags	766.500 ns/op	-14.1 ns/op / -1.8% (better)
Windows MinGW 386	BenchmarkMergeLinkerFlags	691.200 ns/op	-12.9 ns/op / -1.8% (better)
Windows MinGW 386	BenchmarkChannelBuffered	41.360 ns/op	+2.62 ns/op / +6.8% (worse)
Windows MinGW 386	BenchmarkChannelHandoff	870.900 ns/op	-26.9 ns/op / -3.0% (better)
Windows MinGW 386	BenchmarkDefer	44.410 ns/op	-1.73 ns/op / -3.7% (better)
Windows MinGW 386	BenchmarkDirectCall	1.548 ns/op	-0.003 ns/op / -0.2% (better)
Windows MinGW 386	BenchmarkGlobalRead	1.555 ns/op	+0.002 ns/op / +0.1% (worse)
Windows MinGW 386	BenchmarkGlobalWrite	7.768 ns/op	-0.014 ns/op / -0.2% (better)
Windows MinGW 386	BenchmarkGoroutine	110064 ns/op	-105 ns/op / -0.1% (better)
Windows MinGW 386	BenchmarkInterfaceCall	8.379 ns/op	0 ns/op / +0.0%
Windows MinGW 386	BenchmarkRuntimeGetG	1.931 ns/op	+0.002 ns/op / +0.1% (worse)
Windows MinGW ARM64	BenchmarkLookupPCRandom	12.130 ns/op	+0.03 ns/op / +0.2% (worse)
Windows MinGW ARM64	BenchmarkMergeCompilerFlags	580.400 ns/op	-1.5 ns/op / -0.3% (better)
Windows MinGW ARM64	BenchmarkMergeLinkerFlags	546.800 ns/op	+10.2 ns/op / +1.9% (worse)
Windows MinGW ARM64	BenchmarkChannelBuffered	38.650 ns/op	+1.16 ns/op / +3.1% (worse)
Windows MinGW ARM64	BenchmarkChannelHandoff	2300 ns/op	-1035 ns/op / -31.0% (better)
Windows MinGW ARM64	BenchmarkDefer	57.280 ns/op	+0.64 ns/op / +1.1% (worse)
Windows MinGW ARM64	BenchmarkDirectCall	0.663 ns/op	+0.0735 ns/op / +12.5% (worse)
Windows MinGW ARM64	BenchmarkGlobalRead	0.663 ns/op	+0.0001 ns/op / +0.01508% (worse)
Windows MinGW ARM64	BenchmarkGlobalWrite	0.737 ns/op	+0.074 ns/op / +11.2% (worse)
Windows MinGW ARM64	BenchmarkGoroutine	62591 ns/op	-985 ns/op / -1.5% (better)
Windows MinGW ARM64	BenchmarkInterfaceCall	4.148 ns/op	+0.009 ns/op / +0.2% (worse)
Windows MinGW ARM64	BenchmarkRuntimeGetG	1.806 ns/op	+0.037 ns/op / +2.1% (worse)
Windows MSVC	BenchmarkLookupPCRandom	9.686 ns/op	+0.046 ns/op / +0.5% (worse)
Windows MSVC	BenchmarkMergeCompilerFlags	372 ns/op	+1.5 ns/op / +0.4% (worse)
Windows MSVC	BenchmarkMergeLinkerFlags	335.500 ns/op	+2.8 ns/op / +0.8% (worse)
Windows MSVC	BenchmarkChannelBuffered	24.500 ns/op	+2.07 ns/op / +9.2% (worse)
Windows MSVC	BenchmarkChannelHandoff	1108 ns/op	+113.3 ns/op / +11.4% (worse)
Windows MSVC	BenchmarkDefer	44.870 ns/op	-0.33 ns/op / -0.7% (better)
Windows MSVC	BenchmarkDirectCall	1.358 ns/op	+0.224 ns/op / +19.8% (worse)
Windows MSVC	BenchmarkGlobalRead	1.357 ns/op	+0.001 ns/op / +0.1% (worse)
Windows MSVC	BenchmarkGlobalWrite	2.166 ns/op	0 ns/op / +0.0%
Windows MSVC	BenchmarkGoroutine	56330 ns/op	-138 ns/op / -0.2% (better)
Windows MSVC	BenchmarkInterfaceCall	7.365 ns/op	+0.597 ns/op / +8.8% (worse)
Windows MSVC	BenchmarkRuntimeGetG	1.901 ns/op	+0.27 ns/op / +16.6% (worse)
Windows MSVC 386	BenchmarkLookupPCRandom	16.220 ns/op	+0.14 ns/op / +0.9% (worse)
Windows MSVC 386	BenchmarkMergeCompilerFlags	902.400 ns/op	-262.6 ns/op / -22.5% (better)
Windows MSVC 386	BenchmarkMergeLinkerFlags	925.600 ns/op	-167.4 ns/op / -15.3% (better)
Windows MSVC 386	BenchmarkChannelBuffered	37.130 ns/op	-0.65 ns/op / -1.7% (better)
Windows MSVC 386	BenchmarkChannelHandoff	580.600 ns/op	-21.4 ns/op / -3.6% (better)
Windows MSVC 386	BenchmarkDefer	27.440 ns/op	+0.11 ns/op / +0.4% (worse)
Windows MSVC 386	BenchmarkDirectCall	0.956 ns/op	-0.0112 ns/op / -1.2% (better)
Windows MSVC 386	BenchmarkGlobalRead	0.943 ns/op	-0.0219 ns/op / -2.3% (better)
Windows MSVC 386	BenchmarkGlobalWrite	8.784 ns/op	+0.215 ns/op / +2.5% (worse)
Windows MSVC 386	BenchmarkGoroutine	55360 ns/op	+325 ns/op / +0.6% (worse)
Windows MSVC 386	BenchmarkInterfaceCall	4.959 ns/op	+0.146 ns/op / +3.0% (worse)
Windows MSVC 386	BenchmarkRuntimeGetG	1.004 ns/op	-0.015 ns/op / -1.5% (better)
Windows MSVC ARM64	BenchmarkLookupPCRandom	12.210 ns/op	+0.16 ns/op / +1.3% (worse)
Windows MSVC ARM64	BenchmarkMergeCompilerFlags	554.200 ns/op	-7 ns/op / -1.2% (better)
Windows MSVC ARM64	BenchmarkMergeLinkerFlags	527.500 ns/op	-6 ns/op / -1.1% (better)
Windows MSVC ARM64	BenchmarkChannelBuffered	37.550 ns/op	-0.16 ns/op / -0.4% (better)
Windows MSVC ARM64	BenchmarkChannelHandoff	2549 ns/op	-362 ns/op / -12.4% (better)
Windows MSVC ARM64	BenchmarkDefer	59.450 ns/op	-1.59 ns/op / -2.6% (better)
Windows MSVC ARM64	BenchmarkDirectCall	0.664 ns/op	+0.0743 ns/op / +12.6% (worse)
Windows MSVC ARM64	BenchmarkGlobalRead	0.664 ns/op	+0.0007 ns/op / +0.1% (worse)
Windows MSVC ARM64	BenchmarkGlobalWrite	3.797 ns/op	+0.055 ns/op / +1.5% (worse)
Windows MSVC ARM64	BenchmarkGoroutine	56911 ns/op	-183 ns/op / -0.3% (better)
Windows MSVC ARM64	BenchmarkInterfaceCall	4.138 ns/op	-0.015 ns/op / -0.4% (better)
Windows MSVC ARM64	BenchmarkRuntimeGetG	1.811 ns/op	+0.04 ns/op / +2.3% (worse)

Timer runtime benchmarks

Platform	Operation and runtime	ns/op	vs base
Linux	AfterFuncZeroDelivery/Go	900.100 ns/op	-6.6 ns/op / -0.7% (better)
Linux	AfterFuncZeroDelivery/LLGo	45349 ns/op	+7320 ns/op / +19.2% (worse)
Linux	CreateStop/Go	294.100 ns/op	+6.3 ns/op / +2.2% (worse)
Linux	CreateStop/LLGo	2215 ns/op	+359 ns/op / +19.3% (worse)
Linux	RearmStopped/Go	115.800 ns/op	+1.1 ns/op / +1.0% (worse)
Linux	RearmStopped/LLGo	1398 ns/op	-121 ns/op / -8.0% (better)
Linux	ResetActive/Go	68.660 ns/op	+1.25 ns/op / +1.9% (worse)
Linux	ResetActive/LLGo	724.900 ns/op	+16.1 ns/op / +2.3% (worse)
Linux	ResetHeap1024/Go	67.160 ns/op	-0.07 ns/op / -0.1% (better)
Linux	ResetHeap1024/LLGo	182.300 ns/op	+8.3 ns/op / +4.8% (worse)
macOS	AfterFuncZeroDelivery/Go	606.100 ns/op	+96 ns/op / +18.8% (worse)
macOS	AfterFuncZeroDelivery/LLGo	103620 ns/op	+17413 ns/op / +20.2% (worse)
macOS	CreateStop/Go	182 ns/op	+12.4 ns/op / +7.3% (worse)
macOS	CreateStop/LLGo	833.700 ns/op	+417.3 ns/op / +100.2% (worse)
macOS	RearmStopped/Go	78.260 ns/op	+5.49 ns/op / +7.5% (worse)
macOS	RearmStopped/LLGo	576.600 ns/op	+227.1 ns/op / +65.0% (worse)
macOS	ResetActive/Go	53.960 ns/op	+6.67 ns/op / +14.1% (worse)
macOS	ResetActive/LLGo	246.200 ns/op	+86.2 ns/op / +53.9% (worse)
macOS	ResetHeap1024/Go	50.750 ns/op	-9.28 ns/op / -15.5% (better)
macOS	ResetHeap1024/LLGo	94.710 ns/op	+2.86 ns/op / +3.1% (worse)
Windows MinGW	AfterFuncZeroDelivery/Go	571.100 ns/op	+11 ns/op / +2.0% (worse)
Windows MinGW	AfterFuncZeroDelivery/LLGo	184505 ns/op	+1074 ns/op / +0.6% (worse)
Windows MinGW	CreateStop/Go	116.900 ns/op	-0.3 ns/op / -0.3% (better)
Windows MinGW	CreateStop/LLGo	433.100 ns/op	-113.9 ns/op / -20.8% (better)
Windows MinGW	RearmStopped/Go	31.480 ns/op	0 ns/op / +0.0%
Windows MinGW	RearmStopped/LLGo	266 ns/op	-12.4 ns/op / -4.5% (better)
Windows MinGW	ResetActive/Go	20.180 ns/op	+0.02 ns/op / +0.1% (worse)
Windows MinGW	ResetActive/LLGo	150.900 ns/op	-3.2 ns/op / -2.1% (better)
Windows MinGW	ResetHeap1024/Go	20.430 ns/op	-0.15 ns/op / -0.7% (better)
Windows MinGW	ResetHeap1024/LLGo	124.500 ns/op	-1 ns/op / -0.8% (better)
Windows MinGW 386	AfterFuncZeroDelivery/Go	947.100 ns/op	-12.6 ns/op / -1.3% (better)
Windows MinGW 386	AfterFuncZeroDelivery/LLGo	206680 ns/op	+2494 ns/op / +1.2% (worse)
Windows MinGW 386	CreateStop/Go	196 ns/op	-1.6 ns/op / -0.8% (better)
Windows MinGW 386	CreateStop/LLGo	531.600 ns/op	+25.7 ns/op / +5.1% (worse)
Windows MinGW 386	RearmStopped/Go	63.920 ns/op	+0.53 ns/op / +0.8% (worse)
Windows MinGW 386	RearmStopped/LLGo	372 ns/op	+33.1 ns/op / +9.8% (worse)
Windows MinGW 386	ResetActive/Go	39.130 ns/op	-0.07 ns/op / -0.2% (better)
Windows MinGW 386	ResetActive/LLGo	1034 ns/op	+85.7 ns/op / +9.0% (worse)
Windows MinGW 386	ResetHeap1024/Go	39.570 ns/op	+0.17 ns/op / +0.4% (worse)
Windows MinGW 386	ResetHeap1024/LLGo	187.400 ns/op	+0.4 ns/op / +0.2% (worse)
Windows MinGW ARM64	AfterFuncZeroDelivery/Go	669.500 ns/op	+4.5 ns/op / +0.7% (worse)
Windows MinGW ARM64	AfterFuncZeroDelivery/LLGo	151940 ns/op	+1629 ns/op / +1.1% (worse)
Windows MinGW ARM64	CreateStop/Go	212.600 ns/op	+12.8 ns/op / +6.4% (worse)
Windows MinGW ARM64	CreateStop/LLGo	364.800 ns/op	-1.6 ns/op / -0.4% (better)
Windows MinGW ARM64	RearmStopped/Go	70.560 ns/op	-0.03 ns/op / -0.0425% (better)
Windows MinGW ARM64	RearmStopped/LLGo	251.700 ns/op	-1.1 ns/op / -0.4% (better)
Windows MinGW ARM64	ResetActive/Go	31.030 ns/op	-0.11 ns/op / -0.4% (better)
Windows MinGW ARM64	ResetActive/LLGo	131.500 ns/op	+14.6 ns/op / +12.5% (worse)
Windows MinGW ARM64	ResetHeap1024/Go	31.100 ns/op	+0.09 ns/op / +0.3% (worse)
Windows MinGW ARM64	ResetHeap1024/LLGo	125.500 ns/op	-0.6 ns/op / -0.5% (better)
Windows MSVC	AfterFuncZeroDelivery/Go	381.300 ns/op	-4.1 ns/op / -1.1% (better)
Windows MSVC	AfterFuncZeroDelivery/LLGo	108304 ns/op	+1155 ns/op / +1.1% (worse)
Windows MSVC	CreateStop/Go	88.920 ns/op	-0.89 ns/op / -1.0% (better)
Windows MSVC	CreateStop/LLGo	334.200 ns/op	-6.1 ns/op / -1.8% (better)
Windows MSVC	RearmStopped/Go	24.450 ns/op	-0.03 ns/op / -0.1% (better)
Windows MSVC	RearmStopped/LLGo	223.700 ns/op	+5.9 ns/op / +2.7% (worse)
Windows MSVC	ResetActive/Go	14.780 ns/op	-0.02 ns/op / -0.1% (better)
Windows MSVC	ResetActive/LLGo	109.500 ns/op	-4.3 ns/op / -3.8% (better)
Windows MSVC	ResetHeap1024/Go	14.770 ns/op	-0.09 ns/op / -0.6% (better)
Windows MSVC	ResetHeap1024/LLGo	104.800 ns/op	+0.9 ns/op / +0.9% (worse)
Windows MSVC 386	AfterFuncZeroDelivery/Go	622.900 ns/op	-10.1 ns/op / -1.6% (better)
Windows MSVC 386	AfterFuncZeroDelivery/LLGo	109175 ns/op	+2301 ns/op / +2.2% (worse)
Windows MSVC 386	CreateStop/Go	160.300 ns/op	-2.4 ns/op / -1.5% (better)
Windows MSVC 386	CreateStop/LLGo	397.800 ns/op	+46.6 ns/op / +13.3% (worse)
Windows MSVC 386	RearmStopped/Go	62.470 ns/op	+0.14 ns/op / +0.2% (worse)
Windows MSVC 386	RearmStopped/LLGo	239.500 ns/op	-2.3 ns/op / -1.0% (better)
Windows MSVC 386	ResetActive/Go	29.760 ns/op	-0.08 ns/op / -0.3% (better)
Windows MSVC 386	ResetActive/LLGo	720.700 ns/op	-5.7 ns/op / -0.8% (better)
Windows MSVC 386	ResetHeap1024/Go	29.960 ns/op	+0.23 ns/op / +0.8% (worse)
Windows MSVC 386	ResetHeap1024/LLGo	116 ns/op	-4 ns/op / -3.3% (better)
Windows MSVC ARM64	AfterFuncZeroDelivery/Go	665.100 ns/op	-2.2 ns/op / -0.3% (better)
Windows MSVC ARM64	AfterFuncZeroDelivery/LLGo	164063 ns/op	+21082 ns/op / +14.7% (worse)
Windows MSVC ARM64	CreateStop/Go	197 ns/op	-0.3 ns/op / -0.2% (better)
Windows MSVC ARM64	CreateStop/LLGo	387.100 ns/op	+14.1 ns/op / +3.8% (worse)
Windows MSVC ARM64	RearmStopped/Go	70.590 ns/op	0 ns/op / +0.0%
Windows MSVC ARM64	RearmStopped/LLGo	270.800 ns/op	0 ns/op / +0.0%
Windows MSVC ARM64	ResetActive/Go	31.060 ns/op	-0.04 ns/op / -0.1% (better)
Windows MSVC ARM64	ResetActive/LLGo	132.800 ns/op	0 ns/op / +0.0%
Windows MSVC ARM64	ResetHeap1024/Go	31.050 ns/op	-0.04 ns/op / -0.1% (better)
Windows MSVC ARM64	ResetHeap1024/LLGo	137.200 ns/op	-0.9 ns/op / -0.7% (better)

Compared with f06143ba2aac measured in the same runner job.

[cpunion]

Restacked on the #2157 CI fix. All nine #2142-only commits remain patch-identical in git range-diff; the updated head inherits findRepoRoot and the focused DWARF/PCLN acceptance test passes.

[fennoai]

Review: DWARF default Go-compatible

Solid PR. The DWARF-by-default / stable-parameter-home / PC-line-site policy changes are well-factored and come with strong test coverage (native LLDB acceptance suite, TestNativeDWARFDefaults, TestDebugParameterHomes, TestParamHomeReusePreservesDebugStorage, and the pcln/link-option tables). The refactors are net improvements — the constructDebugAddr change drops a dead per-variable Load, and collectDebugAllocObjects replaces an O(n) linear scan (hasDebugAlloc) with an O(1) set lookup on the compile hot path.

No correctness or security blockers found. Inline notes below are minor. A few non-inline maintainability suggestions:

• ssa/di.go — DIParam vs DIParamWithHome: the two exported methods now delegate to the same diParam and differ only in whether the home Expr is returned/discarded. Consider collapsing to one method (callers can ignore the result) to prevent drift.
• cl/compile.go debugRef: the value-extraction block (v.X.(instrOrValue) → p.bvals[iv] / p.compileValue) is now duplicated across the stable-param and existing branches. Extracting a small helper would keep the two from diverging.
• cl/debug_alloc.go hasDebugAlloc: after this PR its only remaining caller is the test file (debugParams now uses debugAllocObjects). Consider removing it or marking it test-only so it doesn't read as leftover code.
• internal/build/link_options.go omitDWARFRequested: the four-level nested boolean (fixed target / wasm / darwin c-shared) is hard to audit; a named defaultsToOmitDWARF(conf) helper or a per-clause comment mapping to the design-doc rationale would help.
• cmd/llgo/debugtest/native/runtest.sh: paths ($script_dir, artifacts) are string-interpolated into LLDB script/command script import args, which is an eval boundary. Safe for CI-controlled paths, but passing them via env vars read in acceptance.py would be more robust. Not blocking.

[cpunion]

Addressed the five suggestions in the review body in 5955a9c86:

1. DIParam / DIParamWithHome: retained the existing public DIParam signature for API compatibility, and documented why. Adding a return value would change its method/function type even though direct callers can discard a result. Both wrappers already delegate to the same private diParam implementation, so parameter emission and home creation have one implementation.
2. Repeated debugRef value lookup: extracted debugRefValue, used by both the stable-home and ordinary debug-reference paths. Missing instruction values still return without rematerializing executable SSA; the same-value/same-block fast path is unchanged.
3. Test-only hasDebugAlloc: removed it and the redundant assertion. The test still verifies that the parameter is in the actual collectDebugAllocObjects set used by compilation, along with its parameter index.
4. Nested default-DWARF condition: expanded it into explicit-precedence, typed-default, fixed-target/Wasm, and Darwin c-shared checks, with short rationale comments. Existing ModeGen, -w, and -s behavior is unchanged.
5. Paths interpolated into LLDB Python: replaced interpolation with environment variables read by the Python acceptance entry point. LLDB executes a fixed runpy command; neither source nor artifact paths become Python source text.

Validation: the existing CL debug-metadata/alloca, SSA parameter/map-location, and build DWARF/default/linker-policy tests pass. Rebuilt the compiler and reran the real native panic/divide/nil plus O2 inline-frame/step-over acceptance with both the source path and temporary artifact directory containing spaces, single quotes, and double quotes; it reports NATIVE_DEBUG_ACCEPTANCE_OK. Shell/Python syntax checks and git diff --check also pass. This cleanup does not change the generated-DWARF or target-default policy.

[cpunion]

Fixed the Windows ARM64 MSVC cgofull link failure in b75bcdc28.

The generated __cgo_c_callback / __cgo_go_callback globals are function-address placeholders. Lowering replaced their uses with real functions but retained the old global declarations and debug attachments. With default DWARF, COFF consequently emitted .debug_info relocations to those undefined placeholders.

Lowering now removes the replaced global and its package/debug indexes. The operation is idempotent for alternate cgo source passes and protects shared zero-sized aliases. Real function addresses and DWARF remain enabled.

Validation:

• The new real-object regression fails on the previous implementation for ARM64 MSVC/GNU, amd64, and 386; all four pass with the fix, retaining .debug_info and the callback-address initializer.
• Full SSA tests and existing Cgo/GenExtern tests pass.
• A forced cgofull rebuild and actual execution on macOS arm64 with default DWARF passes Python, struct and errno checks; all four C/Go callback cases return 124.

The new Windows CI run is still required to confirm the original platform end to end. No test or DWARF generation was disabled.

[github-actions]

LLGo WebAssembly build benchmarks

08d0aef2e9e9 | workflow run | long-term charts

WebAssembly output sizes

Example, profile and compiler	Wasm module	vs base	Generated JS glue	vs base
cprintf/j32-emscripten/LLGo	154193 B	0 B / +0.0%	88742 B	0 B / +0.0%
cprintf/j32-goos-js/LLGo	152295 B	0 B / +0.0%	73165 B	0 B / +0.0%
cprintf/j64-emscripten-memory64/LLGo	141248 B	0 B / +0.0%	92630 B	0 B / +0.0%
cprintf/w32-goos-wasip1/LLGo	152848 B	0 B / +0.0%	0 B	0 B / 0.0%
cprintf/w32-wasi/LLGo	153560 B	0 B / +0.0%	0 B	0 B / 0.0%
fmtprintf/j32-emscripten/LLGo	3213428 B	0 B / +0.0%	132442 B	0 B / +0.0%
fmtprintf/j32-goos-js/Go	2526852 B	0 B / +0.0%	0 B	0 B / 0.0%
fmtprintf/j32-goos-js/LLGo	3192523 B	0 B / +0.0%	101635 B	0 B / +0.0%
fmtprintf/j64-emscripten-memory64/LLGo	2950033 B	0 B / +0.0%	139285 B	0 B / +0.0%
fmtprintf/w32-goos-wasip1/Go	2500019 B	0 B / +0.0%	0 B	0 B / 0.0%
fmtprintf/w32-goos-wasip1/LLGo	2347351 B	0 B / +0.0%	0 B	0 B / 0.0%
fmtprintf/w32-wasi/LLGo	2344013 B	0 B / +0.0%	0 B	0 B / 0.0%
j32-emscripten/LLGo	153428 B	0 B / +0.0%	88742 B	0 B / +0.0%
j32-goos-js/Go	1895533 B	0 B / +0.0%	0 B	0 B / 0.0%
j32-goos-js/LLGo	151768 B	0 B / +0.0%	73165 B	0 B / +0.0%
j64-emscripten-memory64/LLGo	140582 B	0 B / +0.0%	92630 B	0 B / +0.0%
reflectcall/j32-emscripten/LLGo	1543771 B	0 B / +0.0%	105908 B	0 B / +0.0%
reflectcall/j32-goos-js/Go	2191221 B	0 B / +0.0%	0 B	0 B / 0.0%
reflectcall/j32-goos-js/LLGo	1546222 B	0 B / +0.0%	90331 B	0 B / +0.0%
reflectcall/j64-emscripten-memory64/LLGo	1428426 B	0 B / +0.0%	111641 B	0 B / +0.0%
reflectcall/w32-goos-wasip1/Go	2205707 B	0 B / +0.0%	0 B	0 B / 0.0%
reflectcall/w32-goos-wasip1/LLGo	1276693 B	0 B / +0.0%	0 B	0 B / 0.0%
reflectcall/w32-wasi/LLGo	1274301 B	0 B / +0.0%	0 B	0 B / 0.0%
w32-goos-wasip1/Go	1909947 B	0 B / +0.0%	0 B	0 B / 0.0%
w32-goos-wasip1/LLGo	152495 B	0 B / +0.0%	0 B	0 B / 0.0%
w32-wasi/LLGo	153207 B	0 B / +0.0%	0 B	0 B / 0.0%

LLGo WebAssembly build measurements

Example and profile	Build	vs base
j32-emscripten	6.110 s	-202.9 ms / -3.2% (better)
j32-goos-js	5.964 s	-216.2 ms / -3.5% (better)
j64-emscripten-memory64	5.324 s	-441.3 ms / -7.7% (better)
reflectcall/w32-wasi	25.241 s	+1.391 s / +5.8% (worse)
w32-goos-wasip1	4.218 s	-53.05 ms / -1.2% (better)
w32-wasi	4.003 s	-376.7 ms / -8.6% (better)

Compared with f06143ba2aac measured in the same runner job.

[visualfc]

Review

The native DWARF default, O0 parameter homes, Darwin PC-line sites, cgo COFF cleanup, and test isolation look solid. A few issues around default-path cost and cgo revisit handling:

• P2 reuseParamHome is skipped for every function with a subprogram. Native DWARF is now default, so optimized builds lose param-home reuse too.
• P2 cgo's __cgo_* fallback still NewFuncs when the placeholder is already gone.
• P3 alloc-object DebugRefs are suppressed only at O0.
• P3 native panic/inline acceptance is not run on Windows CI.

Inline notes below.