Running debug on Goland with -race flag.

Does it set -gcflags=all="-N -l" along with -race? I can reproduce the failure with that. Arguably that is a weird setting.

yes, I see this: -gcflags all=-N -l

I'll look into if we can fix that. But in the meantime, you can use either -race (for race detection) or -gcflags=all="-N -l" (for debugging) separately. I don't think you need to use them at same time.

Wow, where is this race fun center? I want to play!

Is it darwin only? runtime test on linux/arm64(raspi 4) with '-race' and '-N -l' all passed.

Change https://go.dev/cl/386715 mentions this issue: runtime: define racefuncenter and racefuncexit as ABIInternal

Change https://go.dev/cl/386716 mentions this issue: runtime: save some stack space for racecall on ARM64

Change https://go.dev/cl/386714 mentions this issue: reflect: call ABIInternal moveMakeFuncArgPtrs on ARM64

Change https://go.dev/cl/386719 mentions this issue: runtime: avoid cgo_unsafe_args for syscall.syscall functions on darwin/arm64

With the CLs above there should be no more nosplit overflow from the standard library on ARM64 with -race and -gcflags=all="-N -l".

However, I still suggest not to use -race and -gcflags=all="-N -l" separately, not at same time (it is possible using them at same time will not be supported).

why would we want to remove the ability to do race detection and debugging at the same time? From a dev perspective, I am more likely to run the -race with debugging than I am to swap back and forth. I am also unlikely to compile my production binaries with -race because I won't want the panic.