Re-enable MINGW32-specific compile options

[dscho]

This reverts a mistake of mine: My change to introduce an "else ifeq" in #5439 chain missed that there is an else clause that should be used for MINGW32 and MSys (to add large-address-awareness and use 32-bit time_t) but not for CLANGARM64 nor for MINGW64.

[jeremyd2019]

Why would you still want 32-bit time_t? And why would you want those flags for msys?

[dscho]

Why would you still want 32-bit time_t? And why would you want those flags for msys?

Good question. I instinctively wanted to reinstate what I had mistakenly removed. Now, there's this:

#if defined (_WIN32) && !defined (_WIN64) && !defined (__MINGW_USE_VC2005_COMPAT) && !defined (_UCRT)
#ifndef _USE_32BIT_TIME_T
#define _USE_32BIT_TIME_T
#endif
#endif

But that's mingw-w64's headers, and historically I did not mean to rely on those for the MINGW builds.

I guess the more important idea was to maintain backwards-compatibility because I thought that Git's on-disk representation might be careless enough to use time_t. But now that I looked, it seems as if only the packed_git structure uses this type, and I don't think that this is read back verbatim from a cached version on disk (which is the scenario I want to be careful about: When Git reads data that is potentially written by an older Git version).

So I guess for the 32-bit time_t, it is basically just tradition. And since Git for Windows, as per #5394 and related tickets, will abandon 32-bit builds in 2029 (i.e. well in advance of 2038), I am not worried about the 32-bit time_t, either.

A more important issue here, then, is the large address aware stuff. I cannot find the documentation rn but I vaguely remember that without this option, you cannot allocate regions larger than 2GB on i686, and we very much want to be able to do that.

[jeremyd2019]

ok. yes, the large address aware flag does opt a process into addresses with the most significant bit set. It is not necessary to set it for msys though, it seems to be set by default already.

[dscho]

ok. yes, the large address aware flag does opt a process into addresses with the most significant bit set. It is not necessary to set it for msys though, it seems to be set by default already.

It also does not do any harm and my thinking was the this else clause would catch whatever obscure setup some random (MINGW-based?) developer might have.