I wrote:
> I have the AIX patches all staged here and was planning to push as
> soon as I finish polishing the commit message. I will try it on
> cfarm119 once that's done; if there are any corrections to be made
> I expect they'd be minor.
I confirmed that HEAD passes check-world on p9-aix1-postgres1
with both autoconf and meson builds; that was expected given
previous testing, but it shows I didn't mess up the final commit.
cfarm119 soon reminded me that I'd forgotten about the lgamma(NaN)
problem discussed in [1]. I'm inclined to go ahead and push the
patch shown there. Even though it's fixed in bleeding-edge AIX,
there's no reason to believe that the average AIX installation in
the field is any more up-to-date than cfarm119.
Also, reviewing the thread, I wondered why the pgstat_slru.c build
failure mentioned at [2] is not there anymore; it's not like we
did anything to that code since January. On closer investigation,
what's in unistd.h is actually
#ifdef _LARGE_FILES
#define ftruncate ftruncate64
#define truncate truncate64
#endif
and it seems that we are setting _LARGE_FILES in 32-bit builds but
not 64-bit ones. So not having that problem is an accidental
consequence of deciding not to support 32-bit AIX builds.
I wonder whether we should do something about it anyway.
regards, tom lane
[1] https://www.postgresql.org/message-id/573836.1769545598%40sss.pgh.pa.us
[2] https://www.postgresql.org/message-id/190704.1769449537%40sss.pgh.pa.us
