https://bugs.winehq.org/show_bug.cgi?id=47751
Bernhard Übelacker bernhardu@mailbox.org changed:
What |Removed |Added ---------------------------------------------------------------------------- Distribution|--- |Debian Regression SHA1| |d5a372abbba2e174de78855bdd4 | |a004b56cdc006
--- Comment #1 from Bernhard Übelacker bernhardu@mailbox.org --- A git bisect leads to commit d5a372abbba2e174de78855bdd4a004b56cdc006. The function _CurrentScheduler__Id is tested in msvcrt110 tests, but for some reason I could not find a crash in any testbot logs, also not with a noop patch to msvcrt110. While I am seeing this issue on Debian 10, which is used by testbot too.
I think the issue might be in debian build and mine, that the variable context_tls_index ends up in the .text section of msvcr120.dll.so, which might be readonly.
Before the above mentioned commit that variable is in the .data section of msvcr120.dll.so.
For some reason the DEFINE_VTBL_WRAPPER lines are related. Moving these to the end of the file makes the variable be again in the .data section in current git head.
But as testbot does not suffer from this, I am not sure if this is an issue with my setup or compiler versions, or such. Doesn't testbot use the Debian 10 compiler versions?