Segfault in alarm_query/tglmp_encrypt_send_message/init_enc_msg from telegram-purple about tgl HOT 5 CLOSED

This sounds a lot like a duplicate of #49, so I suggest closing this issue.

from tgl.

How is it a duplicate? The backtrace looks very different.

(I haven't noticed this issue lately, mostly because the few times i did use telegram-purple it was inside valgrind, where I just ignored any errors that weren't my fault)

from tgl.

Keep in mind that the upper stacktrace doesn't crash at the point of fault.

So we're looking at the lower stacktrace, which says:

==4104== Invalid read of size 8
==4104==    at 0x1242EA1A: alarm_query (queries.c:94)
==4104==    by 0x1242EDA4: alarm_query_gateway (queries.c:142)
==4104==    by 0x1240F77B: timer_alarm (tgp-timers.c:34)

alarm_query Invalid read of size 8 is exactly what I was looking at when I was investigating this bug, which mostly manifests as a use-after-free (freed after the first execution, then the second alarm triggers and accesses free'd memory).

And I just noticed: Since this bug report is older, ideally this would be marked as the original, and the other one as the duplicate. However, #49 contains a lot of background information explaining why and what happens.

Anyway, it is resolved on the test branch, so when @vysheng does another merge into master it'll become public.

from tgl.

Fair enough. Also I'm fine with this one being the duplicate since it's less detailed. Thanks for your work debugging this!

from tgl.

This seems like a bug that has been fixed a long time ago. It was caused by certain timers not being freed properly when going offline. It didn't cause any issues when using libtgl with libevent, but it did cause issues when using it with purpe_timer.

I'd suggest closing this issue, it is already resolved.

from tgl.