On 2018-04-07 14:07:05 -0400, Tom Lane wrote:
> Andres Freund <and...@anarazel.de> writes:
> > As Daniel pointed out in:
> > https://postgr.es/m/fb948276-7b32-4b77-83e6-d00167f8e...@yesql.se the
> > pg_atomic_flag fallback implementation is broken. That has gone
> > unnoticed because the fallback implementation wasn't testable until now:
> > ...
> > The attached fixes the bug and removes the edge-cases by storing a value
> > separate from the semaphore. I should have done that from the start.
> > This is an ABI break, but given the fallback didn't work at all, I don't
> > think that's a problem for backporting.
> > Fix attached. Comments?
> pademelon says it's wrong.
> 2018-04-07 13:39:34.982 EDT [1197:89] pg_regress/lock LOG: statement: SELECT
> TRAP: UnalignedPointer("(((uintptr_t) ((uintptr_t)(ptr)) + ((sizeof(*ptr)) -
> 1)) & ~((uintptr_t) ((sizeof(*ptr)) - 1))) != (uintptr_t)(ptr)", File:
> "../../../src/include/port/atomics.h", Line: 177)
Yea, I just saw that.
Afaict it's "just" an over-eager / wrong assert. I can't for the heck of
it think why I wrote (9.5 timeframe)
where the bigger ones all have asserts alignment to their own size. I
assume I did because some platforms want to do atomics bigger than a
single int - but then I should've used sizeof(ptr->value). So far
pademelon is the only animal affected afaict - let me think about it for
a bit and come up with a patch, ok?