search

wasilibs / wazero-helpers

MIT License
2 stars 1 forks source link

Data race in reallocate #4

Closed anuraaga closed 1 month ago

anuraaga commented 1 month ago

I noticed this data race in Rellocate in go-re2. It's not using this package yet but I think it would still have the same issue.

https://github.com/wasilibs/go-re2/actions/runs/9343300382/job/25712674994

I think because malloc takes a lock in Wasm, this isn't a real issue and just that the Go race detector isn't aware of wazevo-compiled locks but probably still good to allow the Go race detector to be safe, microoptimization of Grow shouldn't be too important.

@ncruces Wondering if any thoughts on an implementation, a mutex in LinearMemory, or just use atomic ops?

ncruces commented 1 month ago

Yeah, I would hope a mutex isn't really needed, but it's probably the simplest thing to do.

And since wazero is already doing atomicStoreLengthAndCap, a mutex around Grow might be needed; simple atomics might not quiet the race detector. I wouldn't want to spaghettify this with a CAS loop.