Mango The Fourth
Mango The Fourth
That doesn't seem to be it, you're right, the `payloadLen` before that is already completely off.
So afaiu this means that either the parsing of the header on the `ghc-events` side or the writing of the header on `ghc` RTS side goes wrong. (At least the...
Also don’t they acquire the global eventBuf lock before writing?
> I think it's highly likely that somehow we try to write both a heap sample and a time sample at the same time to the eventlog and the -threaded...
So there must be a context switch somewhere in `dumpCensus`, even with the non-threaded RTS?
Is it possible that this is happening because the time profile is running asynchronously? (see `initTimer` -> `initTicker` -> `createAttachedOSThread` with `handle_tick` -> `handleProfTick` -> `traceProfSampleCostCentre`) maybe there should be...
so maybe it would work if we'd just keep the `ACQUIRE_LOCK` stuff in existence, even in the non-threaded runtime? :eyes:
@TeofilC I don't think that you observation is generally right - I was going to try if I can completely circumvent the issue by using `-threaded` or using only one...
@mpickering - `i10` doesn't fix the issue, this made it so that I got one of the eventlogs through, so I guess it decreases the probability for failure but most...
it also happens when only using `-hy`, no time profiling. Is time profiling still doing things when we only request heap profiling with `-l`?