Viewing Issue Advanced Details
ID Category [?] Severity [?] Reproducibility Date Submitted Last Update
07858 Crash/Freeze Minor Always Jan 27, 2021, 14:05 11 days ago
Tester Robbbert View Status Public Platform MAME (Official Binary)
Assigned To Resolution Open OS Windows 10 (64-bit)
Status [?] Confirmed Driver
Version 0.228 Fixed in Version Build 64-bit
Fixed in Git Commit Github Pull Request #
Summary 07858: CORE:: Use of -verbose causes a crash before start
Description Use of the verbose switch will cause many games to crash while in the process of starting up.
Steps To Reproduce With the official build, do

>mame64 batman -v

It will crash while starting the video device
Additional Information Tafoid has a list of several hundred other games that have the same issue.

I picked batman as it seemed to always happen with the official MAME.

This issue does not occur with my homebuilt MAME, possibly because of a different GCC version. (I'm using 10.1)
Github Commit
Flags
Regression Version
Affected Sets / Systems CORE:
Attached Files
 
Relationships
related to 07978Confirmed UI: File Manager crashes MAME in spectacular ways 
Notes
3
User avatar
No.18407
Tafoid
Administrator
Jan 27, 2021, 14:11
Some more examples:
https://mameload.mameworld.info/verbosecrash.txt
https://mameload.mameworld.info/verbosecrash2.txt

Similar to what was reported in late 2019 which was suspected to be some compiler bug that was exposed during code changes.
https://www.mameworld.info/ubbthreads/showthreaded.php?Cat=&Number=384425
User avatar
No.18413
cuavas
Administrator
Jan 28, 2021, 04:31
I ran gauntlet in a release build under valgrind, and it didn’t turn up any branches on uninitialised values or out-of-bounds accesses (I realise that was on Linux, so different calling convention, object format, etc. but it would still turn up a lot of issues). Given it’s jumping into the weeds from the well-exercised string formatting code, it looks a lot like a compiler bug.
User avatar
No.18804
cuavas
Administrator
11 days ago
Interestingly, -fomit-frame-pointer acts as a work-around. It really looks like something going wrong with exception unwinding.