slydiman wrote: Of course lldb waits for the vFile:close response before sending the vRun packet and lldb-server actually closes the file handle (all of them). No any leaks. Otherwise this workaround wouldn't work.
The behavior is the same on Linux and Windows targets. I launched 100 connections and 200 processes simultaneously on Windows (lldb-server gdbserver + a test app). I got 3..10 fails because of the error ERROR_SHARING_VIOLATION. After this patch I got 0..3 fails for 100 connections and 0 fails for 50 connections. After closing the copied file probably the system may cache it some time per process. The file may be blocked by the built-in antivirus for some time. It is hard to figure out the exact reason. We have a buildbot to run cross API tests in 8 threads with Linux Aarch64 target. All tests are green with the current (single thread) lldb-server. But we got randomly failed 50..60 tests with the multithreading version of lldb-server. Probably it is just little bit faster and system did not unlock the executable. We noticed that usually failed tests use simple and tiny executables. But this fact does not help to explain the reason of the problem. We got 100% green tests after this patch. https://github.com/llvm/llvm-project/pull/100659 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
