On Wed, 4 Jun 2025, Jeremy Drake via Cygwin-apps wrote: > There are a number of test failures, particularly in llvm. A bunch of > them are JIT related segfaults, so I kind of suspect some sort of > low-hanging fruit fix that could knock out several of them at once. I > haven't gotten a chance to look into that yet though.
I had a look, and found a dlfcn screwup that addressing brought failed tests down from 110 to 50. (It turns out somebody at some time knew that Cygwin required RTLD_DEFAULT as the handle passed to dlsym to search all modules for the symbol, so overrode a return to be RTLD_DEFAULT on Cygwin, but unfortunately RTLD_DEFAULT is NULL and elsewhere there was a check if the handle was NULL it was assumed to not have been set). Go figure. I'll have to work through a PR and see how they want to fix it. https://github.com/llvm/llvm-project/pull/143072 I found another test-specific issue for something that never worked on Windows, but still included an exeext substitution, which seems to be actively harmful on Cygwin. That brought it down to 39 failures. https://github.com/llvm/llvm-project/pull/143077
