On Sun, 18 Jan 2026, LIU Hao wrote:
在 2026-1-17 21:03, Martin Storsjö 写道:
This looks mostly like an issue that should be fixed on the busybox-w32
side, but ideally we should coordinate with them around a fix for the
issue. (In the worst case, we could consider temporarily reverting this
addition here until it is sorted out on the busybox-w32 side. I'm unable to
include newer versions of mingw-w64 in releases of llvm-mingw until this is
sorted out...)
Does it make sense to build busybox-w32 with `-DHAVE_MKDTEMP` ?
As a concept, something like that could work for fixing the build
failures. That'd require the caller to decide whether to pass it or not
(e.g. for my builds, when using an older tagged version of mingw-w64, I'd
not set it, but when doing nightly builds with the latest mingw-w64, I'd
need to set it).
Concretely, with current busybox code, it wouldn't work as is, because
their platform.h actively undefs HAVE_MKDTEMP on this platform:
https://github.com/rmyorston/busybox-w32/blob/master/include/platform.h#L470-L475
But overall, if this is how mingw-w64 headers are going to be in the
future, then busybox-w32 probably will need to adapt to it in some way. I
guess it's just up to us to see if that can be done in some good enough
way (so that it works with both older and newer headers), or if we should
revert this to ease things for them. (In this case, the main desire for
the function was for our own testcases, right?)
// Martin
_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
