Hi,
Looks like fairywren is possibly seeing something I saw before and spent many
days looking into:
https://postgr.es/m/20220909235836.lz3igxtkcjb5w7zb%40awork3.anarazel.de
which led me to add the following to .cirrus.yml:
# Cirrus defaults to SetErrorMode(SEM_NOGPFAULTERRORBOX | ...). That
# prevents crash reporting from working unless binaries do SetErrorMode()
# themselves. Furthermore, it appears that either python or, more likely,
# the C runtime has a bug where SEM_NOGPFAULTERRORBOX can very
# occasionally *trigger* a crash on process exit - which is hard to debug,
# given that it explicitly prevents crash dumps from working...
# 0x8001 is SEM_FAILCRITICALERRORS | SEM_NOOPENFILEERRORBOX
CIRRUS_WINDOWS_ERROR_MODE: 0x8001
The mingw folks also spent a lot of time looking into this ([1]), without a
lot of success.
It sure looks like it might be a windows C runtime issue - none of the
stacktrace handling python has gets invoked. I could not find any relevant
behavoural differences in python's code that depend on SEM_NOGPFAULTERRORBOX
being set.
It'd be interesting to see if fairywren's occasional failures go away if you
set MSYS=winjitdebug (which prevents msys from adding SEM_NOGPFAULTERRORBOX to
ErrorMode).
Greetings,
Andres Freund
[1] https://github.com/msys2/MINGW-packages/issues/11864