On Saturday 21 June 2025 13:47:12 Martin Storsjö wrote: > On Fri, 20 Jun 2025, Pali Rohár wrote: > > > On Thursday 19 June 2025 20:52:59 Martin Storsjö wrote: > > > On Sun, 15 Jun 2025, Pali Rohár wrote: > > > > > > > Both msvcrt and UCRT ABI uses for off_t-based function without any > > > > suffix > > > > in their name the 32-bit long off_t type for both 32-bit and 64-bit > > > > architectures. > > > > > > The non-English word order makes this sentence very hard to parse. Right > > > now, this sentence reads "X uses for Z Y" - which isn't how such things > > > are > > > said in English, and makes it very hard to spot which part of the long > > > sentence belongs to the "for Z" part and which part is the "Y" part. The > > > proper word order would be "X uses Y for X". > > > > Feel free to change any wording. I'm fine if anybody else improve > > English in commit messages. > > Sure. I just wanted to point it out in case you want to keep it in mind for > writing future messages, as some of these are really hard to decipher as > they are right now.
Thanks for info. Mostly I'm trying to write commit message the best I can, but english is not my native language and that is visible. > > > Other than that, I didn't really spot any issue with it by reading it > > > through. And a run through the CI looked fine. > > > > > > But given the fallout from the stat patches, I wonder if we should have > > > some > > > sort of test that tries to call all the functions that are involved in > > > this > > > patchset, that we could test build - ideally for all CRTs, but in practice > > > at least for msvcrt+ucrtbase+ucrt, to make sure we don't miss any. > > > > > > // Martin > > > > I will definitely prepare mingw-w64-crt/testcases/* for those _*stat* > > functions, which can catch these issue. > > > > I can also send a separate patch which will add test cases for calling > > all of these functions. > > > > The remaining part is to add and enable CI testing of these mingw-w64 > > integrated test suite, which can be run by the "make check" command. > > I actually did an attempt at adding a CI job that builds these, recently. > But the existing tests seem quite broken right now - "make check" fails on a > number of them right now. I have locally already reproduced the brokenness and I'm already preparing patches which should address some of problems. Tests are incorrectly hooked into automake. > See https://github.com/mstorsjo/mingw-w64/commits/ci-testsuite for my > attempt, and > https://github.com/mstorsjo/mingw-w64/actions/runs/15764626737/job/44438487120 > for the results of running it (which does require you to be logged in on > github to see). > > And secondly, even if they are fixed up, these tests in the current setup > doesn't work well for e.g. UCRT configurations. As mingw-w64-crt/Makefile.am > specifies building with -D__MSVCRT_VERSION__=0x600 (which is right for the > installed files), this ends up breaking testing of UCRT configurations (as > we'd still link against the toolchain default libmsvcrt.a, but build with a > different CRT setting). For the testcases, we'd want to build everything > with just the default toolchain settings, both compilation and linking wise. > > // Martin This is one of the issue that tests are incorrectly hooked into the automake. Tests should not be compiled with -D_CRTBLD and other CRT specific flags (like forcing the __MSVCRT_VERSION__ version). I'm preparing right now fixes for this. _______________________________________________ Mingw-w64-public mailing list Mingw-w64-public@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
