QApiTrace : invalid PNM header then segmentation fault

[hadrien-psydk]

I have a trace, when I replay it and click on a specific line, QApiTrace crashes. Here is the callstack:
#0 ?? ApiSurface::qimageFromRawImage (image=image@entry=0x0, lowerValue=lowerValue@entry=0, upperValue=upperValue@entry=1, opaque=opaque@entry=false, alpha=alpha@entry=false) (/home/hadrien/code-tiers/apitrace/gui/apisurface.cpp:163)
#1 0x422afc ApiSurface::contentsFromBase64(this=this@entry=0x7f95a95dd5c0, base64=...) (/home/hadrien/code-tiers/apitrace/gui/apisurface.cpp:43)
#2 0x42ffdf ApiTraceState::ApiTraceState(this=0x7f9594012b80, parsedJson=...) (/home/hadrien/code-tiers/apitrace/gui/apitracecall.cpp:533)
#3 0x46dbd4 Retracer::run(this=0x2606d20) (/home/hadrien/code-tiers/apitrace/gui/retracer.cpp:527)
#4 0x7f95c359e32f ??() (/usr/lib/x86_64-linux-gnu/libQtCore.so.4:??)
#5 0x7f95c330d182 start_thread(arg=0x7f95a95de700) (pthread_create.c:312)
#6 0x7f95c261630d clone() (../sysdeps/unix/sysv/linux/x86_64/clone.S:111)

The "image" argument is null so when the function leads to a crash on line 162 of apisurface.cpp :
int width = image->width;

In the error log, I have this line:
32599 - Message - api issue 1: FBO incomplete: driver marked FBO as incomplete [-1]

Which is probably the source of the crash.

[hadrien-psydk]

Here is a patch that avoids the crash and uses a replacement default image instead (a black square with the word "NULL" in the middle).

I did not find how to attach a file to a github ticket so here is an external link: http://mess.psydk.org/apitrace/avoid-crash-if-null-image.diff

Here is an illustration of the result:

[jrfonseca]

You patch is interesting (the NULL images actually don't look too bad) but I'd like to understand and fix glretrace not to dump these half-baked images.

Do you still have a trace that enables to repro this?

[jrfonseca]

I think the above round of commits glretrace should no longer dump these half-baked images.

Thanks for looking into this.