On 2024-09-24 Tu 2:31 PM, Dave Cramer wrote:
Hi,
On 2024-02-13 12:49:33 -0500, Dave Cramer wrote:
> > I think I might have been on to something - if my human emulation of a
> > preprocessor isn't wrong, we'd end up with
> >
> > #define S_UNLOCK(lock) \
> > do { _ReadWriteBarrier(); (*(lock)) = 0; } while (0)
> >
> > on msvc + arm. And that's entirely insufficient - _ReadWriteBarrier() just
> > limits *compiler* level reordering, not CPU level reordering. I think it's
> > even insufficient on x86[-64], but it's definitely insufficient on arm.
> >
> In fact ReadWriteBarrier has been deprecated _ReadWriteBarrier | Microsoft
> Learn
> <https://learn.microsoft.com/en-us/cpp/intrinsics/readwritebarrier?view=msvc-170>
I'd just ignore that, that's just pushing towards more modern stuff that's
more applicable to C++ than C.
> I did try using atomic_thread_fence as per atomic_thread_fence -
> cppreference.com
> <https://en.cppreference.com/w/c/atomic/atomic_thread_fence>
The semantics of atomic_thread_fence are, uh, very odd. I'd just use
MemoryBarrier().
#define S_UNLOCK(lock) \
do { MemoryBarrier(); (*(lock)) = 0; } while (0)
#endif
Has no effect.
I have no idea if that is what you meant that I should do ?
Dave
Revisiting this:
Andrew, can you explain the difference between ninja test (which passes) and what the build farm does. The buildfarm crashes.
The buildfarm client performs these steps:
meson test -C $pgsql --no-rebuild --suite setup
meson test -t $meson_test_timeout $jflag -C $pgsql --logbase checklog --print-errorlogs --no-rebuild --suite regress --test-args=--no-locale
meson test -t $meson_test_timeout $jflag -C $pgsql --print-errorlogs --no-rebuild --logbase checkworld --no-suite setup --no-suite regress
foreach tested locale: meson test -t $meson_test_timeout $jflag -v -C $pgsql --no-rebuild --print-errorlogs --setup running --suite regress-running --logbase regress-installcheck-$locale
$pgsql is the build root, $jflag is setting the number of jobs
IOW, we do test suite setup, run the core regression tests, run all the remaining non-install tests, then run the install tests for each locale.
We don't call ninja directly, but I don't see why that should make a difference.
cheers
andrew
--
Andrew Dunstan
EDB: https://www.enterprisedb.com