Crash on OSX - Transmission 14481 - EXC_CRASH (SIGABRT)

[victorhooi]

Hi,

I have Transmission build 14481 installed on OSX 10.9, and I've started receiving intermittent crashes today - 3 of them so far.

I've provided the crash dumps as taken from Console.app:

• https://gist.github.com/victorhooi/c59a5e6d5f1fa09c044a
• https://gist.github.com/victorhooi/a042554437e8c50c581b
• https://gist.github.com/victorhooi/32c37ddfe5e8511201d3

I'm happy to raise a ticket, just wanted to check if this was a known gotcha, like the last one I raised (SIGABRT appears to be too generic a search term on Trac - there's quite a few results).

Please let me know if you need anything else.

CheerVcicttor

[mike.dld]

Looks related to recent changes in r14479. I'll take a look when I'm home unless someone else will get ahead of me.

[mike.dld]

Hmm, I'm not getting it. I thought those were invalid checks, since 0 is a perfectly valid value for a socket and -1 is the invalid one. If you look through the old code, udp_socket and udp6_socket variables (specifically) are being compared as `s < 0` (announcer-udp.c) `s >= 0`, `s > 0`, `s < 0` (tr-dht.c), `s >= 0`, `s < 0` (tr-udp.c), `s` (tr-utp.c), which doesn't make any sense taking into account that those same variables are only being assigned `-1` or the result of calling `accept()` or `socket()`, which in turn return -1 on error (not taking Windows into account here to not add more fun).

[mike.dld]

@victorhooi, could you try previous revisions to see when this all started?

[victorhooi]

I went to build 14480 - seem to get the same crashes:

https://gist.github.com/victorhooi/cd131a3b02d9cb839f94

I went to build 14479 - I ended up getting a different crash, when I tried to do a "Verify Local Data" on some of my recent downloads.

https://gist.github.com/victorhooi/548c83006cf1e64b3cf2

I'll keep an eye on 14479, and see if the original SIGABRT crash recurs, I guess.

[victorhooi]

Crash on 14479 as well:

https://gist.github.com/victorhooi/2b8f664f8779640f0971

[mike.dld]

What about 14478? I wasn't able to reproduce myself yet BTW.

[mike.dld]

I see I suggested "!tor->isRunning" without its counterpart "tor->isStopping". Sorry about that. Maybe you could re-open #5407 and add that bit.

Done (r14483).

I also found one place I didn't change in the commit in question that lead to `-EINVAL` being treated as valid socket descriptor value. Fixed in r14482. @victorhooi, please try out the latest nightly to see if it helps.