http://bugs.winehq.org/show_bug.cgi?id=60177 Wehrwolfmann <wehrwolfmann@gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Regression SHA1| |120db075b5e8863c1ded907e38a | |88662d5ff8e39 CC| |rbernon@codeweavers.com --- Comment #3 from Wehrwolfmann <wehrwolfmann@gmail.com> --- Bisected. The first bad commit is: 120db075b5e8863c1ded907e38a88662d5ff8e39 winex11: Call window_set_wm_state instead of map_window. Its direct parent, 50a51d9aa11 ("winex11: Remove unnecessary calls to map_window helper"), was built and tested separately and is good, so the boundary is that single commit and not a range. How I ran it: 10 builds between Release 10.16 (good) and Release 10.17 (bad), both endpoints rebuilt from source with the same script rather than taken from packages. Every step configures with --enable-archs=i386,x86_64 --disable-tests, installs into a private DESTDIR, creates a fresh prefix, installs the same MS Agent 2.0 files and runs the reproducer from the report. No step had to be skipped: 7 bad, 3 good, no build failures. I cannot yet point at the line that breaks it. What I can add is that the commit does two things at once in dlls/winex11.drv/window.c: it inlines map_window() into X11DRV_WindowPosChanged, and it drops the "if (!is_managed( hwnd )) return;" guard in make_owner_managed(), which now runs earlier and under a different condition. MS Agent is an out-of-process COM server whose marshaling goes through hidden OLE windows, and on the broken versions IsWindow() for the client window returns FALSE inside AgentSvr while the client's notify sink is never unmarshaled - which is why a change to when windows become managed looks related. That last part is a guess; the bisect result itself is not. Happy to test a patch or run any further build you want. -- 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.