[gs-bugs] [Bug 692998] Regression between 1.0rc1 and latest git: pdf document with a lot of vector graphics is rendered incorrectly.

[bugzilla-daemon at ghostscript.com]

http://bugs.ghostscript.com/show_bug.cgi?id=692998

--- Comment #7 from Mike Gorchak <mike.gorchak.qnx at gmail.com> 2012-04-22 17:56:02 UTC ---
(In reply to comment #6)
> Yes. The uninitialised variable things are all to do with the exception
> handling, and the fact that variables can become undefined when a setjmp
> happens.

I have good news for you and bad for me :)

It seems that source code is fine. I developed and tested phmupdf on:

AMD AthlonXP 3800+
Intel Core2Duo E6750
Intel Atom N470
Intel Pentium B950 (one of the newest low end cpu) on Fujitsu AH531 notebook.

This problem appears only on last CPU - Pentium B950 and do not depends on
mupdf source. I upgraded gcc to 4.4.2 - nothing changed. I tried to narrow down
the problem with optimization and have found, that produced code is broken,
when using
no optimization options and -O, -O1 options. -O2 fixes this issue. I also tried
to pass optimization options manually and have found that option -falign-jump
fixes the problem.

So I do not know right now if this is a QNX or gcc issue, or even CPU issue.
But it is definitely not a mupdf issue :)

Excuse my bothering you with this miracle issue. I do not even know how to
report about this issue to gcc team, maybe someone have such CPU to test mupdf
on it using Linux...

-- 
Configure bugmail: http://bugs.ghostscript.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.