On Jul 31 16:57, Jeremy Drake via Cygwin-patches wrote: > On Thu, 31 Jul 2025, Corinna Vinschen wrote: > > > On Jul 31 12:05, Jeremy Drake via Cygwin-patches wrote: > > > I noticed that dll_crt0_1 calls check_sanity_and_sync which performs some > > > checking on the per_process struct from the application, including if the > > > application's api_major is greater than the dll's. However, this is after > > > _cygwin_crt0_common already runs. I tested by downgrading to > > > 3.7.0-0.266 and running an executable that I had built with 267 (but not > > > using the new wrappers). It didn't crash during startup, but it did seem > > > to crash after forking (it was doing a posix_spawn). So maybe the > > > api_major check could catch this after the fact but before the corruption > > > caused any more issues. > > > > How so? That would be in the DLL, but you're running an old DLL which > > you can't change retroactively. OTOH, _cygwin_crt0_common already > > overwrites memory. > > Yes, this check happens after _cygwin_crt0_common has overwitten the > bounds of the __cygwin_cxx_malloc struct, but in my testing this isn't > immediately fatal, and the api_major check would abend the program with a > suitable message. I should test this with MSYS2, to make sure the memory > layout of the dll isn't different,
The memory layout is fly-by-night. Different compiler, different compiler versions, subtil, unrelated changes in the DLL... > and also because it's easier to get > older DLL versions (I only tried with the snapshot of 3.7.0 before the new > wrappers were added, but I can try with 3.6 3.5 3.4 and 3.3 there > pretty easily). Doesn't make much sense, IMHO, unless you really love spending time with testing old versions :) Corinna
