frasercrmck wrote: > Yes, the article you link to is highly relevant. Without this PR we're > hitting the problem described in example #4. > > We can promote the issue to a build failure by choosing a custom command > which can't run concurrently with itself: > > ```cmake > cmake_minimum_required(VERSION 3.2) > > add_custom_command( > OUTPUT gen > # Try to grab a lock and fail immediately if we can't: > COMMAND flock -x -n gen.lock sleep 1 > COMMAND cmake -E echo Hello > gen > ) > > add_custom_target( > my-all-1 ALL DEPENDS gen > ) > > add_custom_target( > my-all-2 ALL DEPENDS gen > ) > ``` > > With the above `make` will succeed, but `make -j2` will fail. The issue is > not exposed with Ninja. This is the same behavior I see in libclc today, > except more jobs are needed. Adding an intermediate target and depending on > both the new target and the file does seem to fix all of our problems. > > I've added the file dependencies back in > [b707fc7](https://github.com/llvm/llvm-project/commit/b707fc7dea2e5e57bc3e4f01e27b9ca7bbf0e398), > and I think it fixes the build issue while preserving the incremental build > behavior you described. Could you please give it a try?
Brilliant, thanks. I have played around and see that incremental builds appear to work as intended. I've tried deleting and `touch`ing various files and full rebuilds and partial rebuilds do what I'd expect. Thanks also for pointing out I've already read about this exact issue (example #4) and missed it! It's a real shame there isn't a better framework for testing CMake and build generators themselves. https://github.com/llvm/llvm-project/pull/95018 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
