GNOME Bugzilla – Bug 778720
Crash in SoupCacheInputStream when cancelling a message load
Last modified: 2017-12-12 13:15:30 UTC
The gtk widget port (Bug 765516) introduced the use of a stand-along libsoup session for loading avatars from Gravatar, but the current implementation is triggering a libsoup crasher bug (Bug 771912). We need to work around this before releasing 0.12.
Fix for this pushed to master as commit 4ca86af, will sit on this a bit before marking it as closed.
Federico, just FYI. I think you were seeing some crashes with related to this?
Yes, but I can't remember if and how I was able to trigger the crash.. perhaps when loading big conversations? Yes, it seems so. Just got a crash with latest master while loading a 150 messages conversation: DB QUERY STEP "SELECT id, fields, date_field, date_time_t, from_field, sender, reply_to, to_field, cc, bcc, message_id, in_reply_to, reference_ids, subject, header, body, internaldate, internaldate_time_t, rfc822_size, preview, flags FROM MessageTable WHERE id=?" elapsed=2,595242 ** libsoup:ERROR:soup-cache-input-stream.c:195:soup_cache_input_stream_write_next_buffer: assertion failed: (priv->output_stream && !g_output_stream_is_closed (priv->output_stream)) Aborted (core dump creato) Let me know if you need a backtrace or anything else.
Cheers. Don't worry about the backtrace, I've seen it again since as well. It probably won't help, but I did push a fix just then as commit f3fd81b to stop using a deprecated version of a method in the code path involved in this crash. Next time you update and rebuild, can you try loading that large conversation again and see if it crashes again?
I've tried again with the 150 messages conversation and no crash this time! I would say that the crash percentage was about 95% of times. Now it's dropped to zero.
Bumping things that aren't likely to make 0.12.0.
Hey all, I just pushed a WIP branch that should fix the issue to git as wip/778720-soup-avatar-crash. Please try this out and let me know if it fixes the issue for you. I can't seem to reproduce it on long conversations any more.
I still have my 183 messages thread, but I cannot reproduce the crash since end of February (see comment #5). Should you ask to the mailing list?
(In reply to Federico Bruni from comment #8) > I still have my 183 messages thread, but I cannot reproduce the crash since > end of February (see comment #5). > > Should you ask to the mailing list? Oh maybe, I'm pretty sure the fix is good though. I've been seeing the crashes more frequently since I'm usually running under gdb with G_DEBUG=fatal-criticals set, and haven't seen any with this new fix applied. Just pushed to master as commit b2f0eca along with a fix for a more obscure problem in commit b627fd4. Will let the m bake for a while then merge to geary-0.12.
Thee commits have been cherry picked to geary-0.12 as commit 60fed7a and commit 0a036e0.