NWNX_Events Segfault on Quickchat

[Seph K]

Hi everyone, new to these forums. I ran into an issue with the current llinux version of nwnx_events, whenever a character uses a particular emote/quickchat, either from pressing v, etc on the keyboard or through the subradial menu, the server crashes. Not all emotes cause a crash, but the ones that do will always cause a crash. The one I first noticed was the "greet" emote.

My first thought was that I had screwed something up when compiling, so I recompiled > Segfault. Then I downloaded the lastest build from Elven's stickied post of nwnx_events.so > segfault. So then I just started trying all of the builds of nwnx_events and found that all builds after #118 (May 08, 2013) have this problem, at least on my end. The May 08 Build, and any build before then works fine and does not segfault.

P.S. I tried to do a backtrace when the first segfault happened, but as I am new to linux, and had never done it before, I was extremely unsuccessful.

[elven]

Build 118 was debian squeeze, build 120 is debian wheezy, which pulled in a new version of gcc and appendages.

I've also been seeing reliable crashes with SCO/RCO in odbc when using the new compiler. I think they've 'optimised' asm generation somehow, but I didn't investigate. I'd suggest using build #118 until this crap is fixed .. sorry for the mess.

[virusman]

Thanks for taking the time to narrow down the cause!
Apparently, NWNX has major issues with GCC >=4.6. Try to avoid GCC 4.6 or 4.7 until we've tested everything. Last known working version is GCC 4.4.
_________________
In Soviet Russia, NWN plays you!

[eeriegeek]

I haven't looked into this recently, but based on the debugging I was doing last year, I think the gcc folks may have changed the generated function call stack frame exit code. I think that (possibley) rewritting the hooks from embedded assembly (relying on the gcc stack handling) to pure assembly (with the old exit syntax) MAY solve at least some of the issues with new gcc versions.

[leo_x]

I've been using NWNX built with gcc4.7 for awhile on my test server. The machinery in nx_hook definitely still works fine, so does d_redirect style (at least in the case of nwnx_optimizations). So, while I have no clue what's causing the issue (and even re-implemented part of nwnx_events because of it), it seems particular to only some plugins.
_________________
the awakening (PW Action)

[virusman]

I think it only applies to the older plugins that heavily use assembly in their hook code. Recent plugins rely on C declarations and should be working fine.
ODBC, Events, Chat, Profiler, Resman may be affected.
_________________
In Soviet Russia, NWN plays you!

[virusman]

I've just pushed a change to the git repo. Please try recompiling from the latest revision.
_________________
In Soviet Russia, NWN plays you!

[Seph K]

now it doesn't want to compile, says error 2 in defenses>funcs/GetAcByType, or something along those lines.

[Seph K]

Sorry was in a hurry earlier, here's more info:

[Seph K]

By selectively editing makefiles and adding -maccumulate-outgoing-args to the individual plugins I got additional errors sequentially:

[virusman]

Hmm, try to add -fno-omit-frame-pointer to the flags.
_________________
In Soviet Russia, NWN plays you!

[virusman]

What's your OS and GCC version? I can't reproduce the problem on Debian Wheezy. Everything compiles fine.
_________________
In Soviet Russia, NWN plays you!

[Seph K]

I tried compiling on fedora 18 with GCC 4.7.2-8, then I also tried on Xubuntu 12.04.2 with GCC 4.6.3-1ubuntu5

[Seph K]

The -fno-omit-frame-pointer option seems to keep the compiler happy.

[virusman]

I've disabled optimization for autoconf/make compile path. cmake didn't pass -O2, so it should be working anyway.
_________________
In Soviet Russia, NWN plays you!