mingw-w64: Update to 03d8a40f57

[lazka]

mingw-w64/mingw-w64@a2d1921...03d8a40

drop the revert since that should be fixed via mingw-w64/mingw-w64@c9eca28

[rouault]

I strongly believe this update has caused a regression in the GDAL regression test suite.
Last good build in https://github.com/OSGeo/gdal/actions/runs/12505586623/job/34889173672 used git-12.0.0.r446.ga2d19218d-1 and first failed in https://github.com/OSGeo/gdal/actions/runs/12515627580/job/34913475983 uses -git-12.0.0.r458.g03d8a40f5-1
One failure is about long filenames starting with \\?\ (not sure which calls to the Windows API has regressed there)
The other failure is more straightforward. It is about a failure renaming "tmp/yy\u4E2D\u6587.\u4E2D\u6587" on top of itself using _wrename()

[Biswa96]

Could you provide any minimal sample code to reproduce the issue?

[lazka]

just to confirm, looking at the build logs the only package versions differences are due to this mingw-w64 change, so seems likely.

[rouault]

I figured the root cause. With git-12.0.0.r446.ga2d19218d-1, the below code prints 0x700, whereas with >= git-12.0.0.r458.g03d8a40f5-1, it prints 0x600. The GDAL code base has code paths that require __MSVCRT_VERSION__ >= 0x601 to use some functions of the Win32 API

test.c:

#include <windows.h>
#include <stdio.h>

int main()
{
    printf("0x%X\n", __MSVCRT_VERSION__);
    return 0;
}

old working mingw64 install:

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ pacman -Q | grep git
mingw-w64-x86_64-crt-git 12.0.0.r369.g0d4221712-2
mingw-w64-x86_64-headers-git 12.0.0.r369.g0d4221712-2
mingw-w64-x86_64-libmangle-git 12.0.0.r369.g0d4221712-1
mingw-w64-x86_64-libwinpthread-git 12.0.0.r369.g0d4221712-1
mingw-w64-x86_64-tools-git 12.0.0.r369.g0d4221712-1
mingw-w64-x86_64-winpthreads-git 12.0.0.r369.g0d4221712-1
mingw-w64-x86_64-winstorecompat-git 12.0.0.r369.g0d4221712-1

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ gcc test.c -o test.exe

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ ./test.exe
0x700

latest mingw64 install:

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ pacman -Q | grep git
mingw-w64-x86_64-crt-git 12.0.0.r459.g63f3f2846-1
mingw-w64-x86_64-headers-git 12.0.0.r459.g63f3f2846-1
mingw-w64-x86_64-libwinpthread-git 12.0.0.r459.g63f3f2846-1
mingw-w64-x86_64-winpthreads-git 12.0.0.r459.g63f3f2846-1

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ gcc test.c -o test.exe

evenr@DESKTOP-J297O1T MINGW64 /c/dev
$ ./test.exe
0x600

[rouault]

ok, so this seems to be related to commit mingw-w64/mingw-w64@90d3f33
Probably that the build recipe here should use --with-default-msvcrt=msvcr70 to restore the previous behavior ?

[lazka]

ok, so this seems to be related to commit mingw-w64/mingw-w64@90d3f33

Here is the discussion behind it: https://sourceforge.net/p/mingw-w64/mailman/mingw-w64-public/thread/20241018181538.bahgwxwolvx2n6cj%40pali/#msg58829839

Probably that the build recipe here should use --with-default-msvcrt=msvcr70 to restore the previous behavior ?

That would break the mingw-w64 internal checks then, since we are not using msvcr70. We could revert that linked commit temporarily, but removing the checks for __MSVCRT_VERSION__ on your side, as you did in the PR, is probably the best way forward.

[rouault]

but removing the checks for __MSVCRT_VERSION__ on your side, as you did in the PR, is probably the best way forward.

ok, I'll stick with that then since it doesn't seem to have obvious inconvenience (if one excepts the fact that current released versions of GDAL would be slightly broken when built against current mingw-w64, until our next release, but I guess that's life)

thanks for your assistance

[lazka]

There is some talk upstream to maybe move the default to 0x06FF instead, to avoid issues like this.

I've also created boostorg/align#21 for boost to remove some old checks as well there.

[ognevny]

the repo seems abandoned, latest commit was created 2 years ago :(