[no merge see #60505] gcc-16 with gdc patches for arm

[sgn]

[skip ci][ci skip]

Based on #60505

This is available for @oreo639 to cherry pick

[oreo639]

Thanks for the patch. I am still running into the crashes. (although after looking into it, it might not be related to the gcc 16 time_t patches)

(You need to be compiling on 32-bit musl to encounter the ICE, cross compiling won't encounter the issue.)

The line in question afaict from the gdb disas: https://github.com/dlang/dmd/blob/6b49d066a27f70ea935213f8148e675b7d1af996/druntime/src/core/time.d#L2882-L2883

The crashing is happening in stage1 though. Initially, I thought it was an issue with the time_t patch in gcc 16 since it was occurring in core.time.TickDuration._sharedStaticCtor49 (which is a part of druntime), and I thought gcc was compiling druntime for stage 1 gdc (checking now, I'm pretty sure just uses druntime from, gcc-11 in this case), and the crashing wasn't happening with previous versions of gcc (including gcc 14) for whatever reason.

It does seem like TickDuration got removed from dmd upstream though.

gcc16/ldc output: https://gist.github.com/oreo639/45ccd0fb222524346ffb480c15cd8d8b

EDIT: I doubled checked, clock_getres(CLOCK_MONOTONIC, &ts) in qemu does return {.tv_sec = 0, .tv_nsec = 1} as expected (same when running it natively). We don't have any musl time_t patch there so it probably is caused by that.

EDIT 2: Adding a patch to gcc11 to just use 1_000_000_000 if tv_nsec is zero does allow gcc 16 to get past stage1 although it still fails to link here: https://gitlab.com/gnutools/gcc/-/blob/032c099d7a8703e6611344a4ffaf940939825566/libphobos/libdruntime/core/sys/posix/time.d#L499-518 (presumably the -fversion flag didn't get pulled in somewhere) I will continue tomorrow. (Compiling gcc through qemu is very slow)