Heikki Linnakangas <hlinnaka@iki.fi> writes:
> What do y'all think of adding a check for
> ALIGNOF_DOUBLE==MAXIMUM_ALIGNOF to configure.ac and meson.build? It's
> not a requirement today, but I believe AIX was the only platform where
> that was not true. With AIX gone, that combination won't be tested, and
> we will probably break it sooner or later.
+1, and then probably revert the whole test addition of 79b716cfb7a.
I did a quick scrape of the buildfarm, and identified these as the
only animals reporting ALIGNOF_DOUBLE less than 8:
$ grep 'alignment of double' alignments | grep -v ' 8$'
hornet | 2024-02-22 16:26:16 | checking alignment of double... 4
lapwing | 2024-02-27 12:40:15 | checking alignment of double... (cached) 4
mandrill | 2024-02-19 01:03:47 | checking alignment of double... 4
sungazer | 2024-02-21 00:22:48 | checking alignment of double... 4
tern | 2024-02-22 13:25:12 | checking alignment of double... 4
With AIX out of the picture, lapwing will be the only remaining
animal testing MAXALIGN less than 8. That seems like a single
point of failure ... should we spin up another couple 32-bit
animals? I had supposed that my faithful old PPC animal mamba
was helping to check this, but I see that under NetBSD it's
joined the ALIGNOF_DOUBLE==8 crowd.
regards, tom lane