cmd/compile: building with `-asan` generates many more allocations than with other instrumentation modes

[mknyszek]

I'm not really sure what the trouble is yet, but this does seem to be true given how many allocation count tests failed on asan but not with race or msan.

For example, tests failed on the new msan builder: https://ci.chromium.org/ui/p/golang/builders/ci/gotip-linux-amd64-msan-clang15/b8732829196319772801/overview

This mostly aligns with skips for race mode.

But with asan, there are many more: https://ci.chromium.org/ui/p/golang/builders/ci/gotip-linux-amd64-asan-clang15/b8732829195285565617/test-results?sortby=&groupby=

These tests will be skipped for now, so this is low priority. But anyone interested in improving the performance of asan mode might consider looking here first.

Discovered during #70054.

[gabyhelp]

Related Issues and Documentation

• cmd/compile: //go:norace for ASAN and other instrumentation #64310 (closed)
• all: skip allocation counting tests with -asan and -msan
• x/build: add MSAN and ASAN builders #70054
• cmd/dist: add asan/msan smoke tests

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[gopherbot]

Change https://go.dev/cl/622855 mentions this issue: all: skip and fix various tests with -asan and -msan

[mknyszek]

In triage, I am told this may be totally intentional. ASAN doesn't work quite so well with stack pointers, so it intentionally forces a lot more to the heap, like every pointer converted to an unsafe.Pointer.

[gopherbot]

Change https://go.dev/cl/623957 mentions this issue: [release-branch.go1.23] all: skip and fix various tests with -asan and -msan

[gopherbot]

Change https://go.dev/cl/654835 mentions this issue: context: skip TestAllocs when using ASAN