Move macOS overrides for st_mtim & friends into a generic macOS source

[barracuda156]

Nim Version

2.2.6

Description

macOS uses st_mtimespec instead of st_mtim. However for some reason correct macOS definitions are hidden in a x86-specific source:

Nim/lib/posix/posix_macos_amd64.nim

Lines 213 to 216 in 548b1c6

	when defined(osx):
	st_atim* {.importc:"st_atimespec".}: Timespec ## Time of last access.
	st_mtim* {.importc:"st_mtimespec".}: Timespec ## Time of last data modification.
	st_ctim* {.importc:"st_ctimespec".}: Timespec ## Time of last status change.

This leaves powerpc build broken.

Current Output

gcc -isystem/opt/local/include/LegacySupport -w -fmax-errors=3 -fno-strict-aliasing -arch ppc -Ic_code -c c_code/3_16/@pos.nim.c -o c_code/3_16/@pos.nim.o
c_code/3_16/@pos.nim.c: In function 'nosgetLastModificationTime':
c_code/3_16/@pos.nim.c:959:59: error: 'struct stat' has no member named 'st_mtim'
  959 |         nimln_(292);    result = toTime__pureZos_u27(res_1.st_mtim);
      |                                                           ^
Command failed: PATH=./bin:$PATH ./build.sh --os darwin --cpu powerpc
Exit code: 1

Expected Output

Known Workarounds

The fix is already there, apparently, but in a wrong place.

Additional Information

No response

[barracuda156]

@Araq Any reason this file is included only for 64-bit?

Nim/lib/posix/posix.nim

Lines 107 to 108 in b901a80

	elif (defined(macos) or defined(macosx) or defined(bsd)) and defined(cpu64):
	include posix_macos_amd64

Perhaps posix_macos_amd64 should be made just posix_macos, and if something is actually specific to either x86_64 or 64-bit, that can be excluded for non-matching cases.

[bptato]

A few months ago I started (but never finished...) reworking posix_macos_amd64 to properly reflect the POSIX int type zoo on each BSD-derived system, and I was wondering about the same thing. Surely ABI differences per architecture aren't significant enough to warrant a separate include (that is, I think it's better handled inline).
It's also weird that we have a separate posix_openbsd_amd64, really that should be merged with posix_macos_amd64 too.

[barracuda156]

@bptato From a quick look I did not see anything explicitly x86_64 in posix_macos_amd64 (maybe some types which are references from system headers will differ between 32- vs 64-bit, but AFAIK not many are, and those which are – are known, like vm statistics). I wonder how this works on i386 currently, or powerpc64-openbsd…

[bptato]

maybe some types which are references from system headers will differ between 32- vs 64-bit

Yes, that's the part where you have to be careful, but from what I've seen it's not very common.
I'll double check the BSD headers some time.

I wonder how this works on i386 currently, or powerpc64-openbsd…

Even on Linux, i386 falls into the posix_other branch, which mostly just defines everything to int. That's almost always wrong but usually it still works by accident.
ppc64/OpenBSD uses the "macos_amd64" include which should work fine for the most part (although there are a couple inaccuracies). At least I run riscv64/OBSD without issues, and it's a similar case.

[barracuda156]

@Araq Could you say what is the difference between defines osx, macosx and macos?

[bptato]

ref.

Nim/compiler/options.nim

Lines 664 to 667 in 2dbdf08

	of "macintosh":
	result = conf.target.targetOS in {osMacos, osMacosx, osIos}
	of "osx", "macosx":
	result = conf.target.targetOS in {osMacosx, osIos}

"osx" and "macosx" seem to be identical.
"macos" might just be a typo - my first guess was classic mac, but that's listed as "macintosh"...

[Araq]

"macos" is/was classic mac but it's not supported officially. (Might be used incorrectly too)

[barracuda156]

@Araq Oh wow, that even I cannot test. Possibly some folks from MacRumors or @classilla could, but first we need to address more general issues for powerpc.