在 2026-5-5 21:29, Kirill Makurin 写道:
LIU Hao wrote:Those style changes are also subject to changes over time, and there's hardly an agreement.What do you thing about formatting at least winpthreads' tests? No one has touched them in more than a decade, so I think it won't be a big issue. I also can try to reduce amount of formatting changes to minimum in other files: - replace hard tabs with spaces for indentation - remove extra whitespace
These are fine.
I used "breaking changes" in that context to mean changes which will make it harder to backport things. I don't have plans for any user-visible API/ABI breaks. Some of changes I have in mind are: - add comments for most static functions, so people who read the code have an idea what they are doing - move some stuff around within source files, so related functions are located close to each other - rename some internal functions and structure field names to improve readability
Maybe these are fine, too. Honestly, I don't mind such changes.
One thing I realized is that winpthreads' pthread_spinlock implementation is half-working - it does not keep track of the lock owner, which means that a thread can release lock held by another thread. This is what causes some of pthread_spinlock/ tests to fail. Another thing I have in mind is to allow winpthreads to run on Windows versions as old as NT 3.1; currently it uses APIs which are available only since NT 4.0.
First of all, there's absolutely no reason to use spinlocks in user mode: https://www.realworldtech.com/forum/?threadid=189711&curpostid=189723
And second, are there really people who compile code with GCC for Windows NT 3.1? -- Best regards, LIU Hao
OpenPGP_signature.asc
Description: OpenPGP digital signature
_______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
