I've been trying out the Solaris compiler (Oracle Developer Studio 12.6,
on Solaris), mainly to test out some C11 support details. Observations:
- That compiler has a certain known idiosyncratic way of dealing with
some inline functions, which lead to commits such as 3e9ca526012 and
discussions such as [0] (ironically the first thread you find when
searching for "solaris compiler dead"). So people did use this at some
point. But this has been broken again since commit 4908c5872059. So
apparently, no one has used this compiler successfully since at least
PG17. (This issue is easy to fix with some #ifndef FRONTEND, as
before.) (The 32-bit build appears to have been uncompilable since
commit c8b2ef05f48 in PG16. This is fixed by 8-byte-datums-everywhere
in master.)
[0]:
https://www.postgresql.org/message-id/20180928150700.h4ikpnh2fibib...@alap3.anarazel.de
- At least in my installation (which is supposed to have all the latest
versions), the compiler runs out memory when trying to compile gram.c,
even without optimizations on. (You can get past this by compiling that
one file with gcc.)
- This compiler is not supported in Meson at all. (meson setup run
fails very early.) I also don't see any recent relevant attempts in the
Meson issue tracker.
- No buildfarm coverage since who knows when (possibly since [0]?).
- Hundreds of compiler warnings (They are not necessarily wrong, but it
shows that no one has taken care to tune the warnings or the code in any
way recently.)
- Shaky C11 support:
* I was trying to test the ICE_P() macro discovered by David Rowley [1]
-- I doesn't work. (It returns zero even for constants, so at least it
fails in the right direction.)
[1]:
https://www.postgresql.org/message-id/CAApHDvoOnd4PBNV0qyJVLmbWvWUjztzcMH5xY2AGp5Vov6XU3Q%40mail.gmail.com
* Strangely, the compiler claims to support __builtin_constant_p(), but
our configure test for that fails for other mysterious weird reasons.
* This compiler rejects _Alignas() inside structure definitions, even
though it appears to support pretty much everything else, so it seems
kind of incomplete.
* It gives "typedef redeclared" warnings for my patch set [2], but I
suppose there would be a way to turn those off via warning toggling options.
[2]:
https://www.postgresql.org/message-id/flat/10d32190-f31b-40a5-b177-11db55597...@eisentraut.org
I mean, at least the AIX enthusiasts popped up right away when we broke
their platform. Here, it seems, truly no one cares.
So, can we declare that we don't support this compiler anymore?
(gcc is readily available and seems to work alright, so I'm not saying
the OS is no longer viable.)
