Go version
go version go1.26-devel_bffe7ad9f1 Wed Sep 17 06:43:57 2025 -0700 darwin/arm64
Output of go env in your module/workspace:
AR='ar'
CC='clang'
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_ENABLED='1'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
CXX='clang++'
GCCGO='gccgo'
GO111MODULE=''
GOARCH='arm64'
GOARM64='v8.0'
GOAUTH='netrc'
GOBIN='/Users/tom/.local/share/mise/installs/go/1.25.0/bin'
GOCACHE='/Users/tom/Library/Caches/go-build'
GOCACHEPROG=''
GODEBUG=''
GOENV='/Users/tom/Library/Application Support/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFIPS140='off'
GOFLAGS=''
GOGCCFLAGS='-fPIC -arch arm64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -ffile-prefix-map=/var/folders/r1/myg4s_g127d6pjrb1rbry1_80000gn/T/go-build559902963=/tmp/go-build -gno-record-gcc-switches -fno-common'
GOHOSTARCH='arm64'
GOHOSTOS='darwin'
GOINSECURE=''
GOMOD='/dev/null'
GOMODCACHE='/Users/tom/go/pkg/mod'
GONOPROXY=''
GONOSUMDB=''
GOOS='darwin'
GOPATH='/Users/tom/go'
GOPRIVATE=''
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/Users/tom/.local/share/mise/installs/go/1.25.0'
GOSUMDB='sum.golang.org'
GOTELEMETRY='local'
GOTELEMETRYDIR='/Users/tom/Library/Application Support/go/telemetry'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/Users/tom/.local/share/mise/installs/go/1.25.0/pkg/tool/darwin_arm64'
GOVCS=''
GOVERSION='go1.26-devel_bffe7ad9f1 Wed Sep 17 06:43:57 2025 -0700'
GOWORK=''
PKG_CONFIG='pkg-config'
What did you do?
We have a binary that has around ~1GB of DWARF data when built. This takes ~4 minutes, 3 of which is spent inside the linker.
What did you see happen?
When passing the -w argument to the linker to skip including DWARF data the linker runtime is reduced to ~30 seconds.
After some profiling (see attached linker-cpu.prof.gz), ~25% of linker runtime is spent inside runtime/memmove.
One of the biggest culprits seems to be WriteSym here:
|
func (out *OutBuf) WriteSym(ldr *loader.Loader, s loader.Sym) []byte { |
|
if !ldr.IsGeneratedSym(s) { |
|
P := ldr.Data(s) |
|
n := int64(len(P)) |
|
pos, buf := out.writeLoc(n) |
|
copy(buf[pos:], P) |
|
out.off += n |
|
ldr.FreeData(s) |
|
return buf[pos : pos+n] |
Specifically, with this binary copy() is called 5,402,448 times and moves a total of 832 MB.
What did you expect to see?
Building binaries that bring in large volumes of DWARF data should not take ~3 minutes to build.
Go version
go version go1.26-devel_bffe7ad9f1 Wed Sep 17 06:43:57 2025 -0700 darwin/arm64
Output of
go envin your module/workspace:What did you do?
We have a binary that has around ~1GB of DWARF data when built. This takes ~4 minutes, 3 of which is spent inside the linker.
What did you see happen?
When passing the
-wargument to the linker to skip including DWARF data the linker runtime is reduced to ~30 seconds.After some profiling (see attached linker-cpu.prof.gz), ~25% of linker runtime is spent inside
runtime/memmove.One of the biggest culprits seems to be
WriteSymhere:go/src/cmd/link/internal/ld/outbuf.go
Lines 307 to 315 in bffe7ad
Specifically, with this binary
copy()is called5,402,448times and moves a total of832 MB.What did you expect to see?
Building binaries that bring in large volumes of DWARF data should not take ~3 minutes to build.