http://bugs.winehq.org/show_bug.cgi?id=52386 --- Comment #11 from Bernhard Übelacker <bernhardu@mailbox.org> --- Created attachment 81734 --> http://bugs.winehq.org/attachment.cgi?id=81734 debugging-wine-11.14.txt After some patches to LLVM got accepted I tried to check where we are here. I tested quite a few llvm-mingw releases since this bug got opened. Added notes of it in attached file. It looks since LLVM 20.1.0 the instruction size decoder understands enough instructions to intercept most needed functions of wine-11.14, the winehq debian packages. At least the ucrt-i686 variant is able to detect the use-after-free.cc test case. Wine seems to have gained more compatibility to Windows as all msvcrt-i686 builds show now: "((__interception::real_memcpy)) != (0)". This is also shown when running with a native Windows. Following seems to have more details: https://github.com/mstorsjo/llvm-mingw/issues/552 In the end llvm-mingw dropped sanitizer support in msvcrt builds. There appeared a few more "unhandled instruction", for which I placed another pull request to LLVM: https://github.com/llvm/llvm-project/pull/214707 The x86_64 case, which is more subject of #50993, I did give only a short test with recent llvm-mingw build and it looks working too. (Given the system has enough memory.) When I build the testcase with llvm-mingw-20211002-msvcrt-i686 it seems to fail here also with native Windows. A build by llvm-mingw-20211002-ucrt-i686 works with native Windows. Should therefore this bug be resolved as notourbug? -- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.