A strang bug on 64 bit systems (Linux only?)

bjorn on 2007/09/15 00:17 of Hole in the Head Studios - We might have lost our minds...

So I've noticed a strange bug with a few of the games, which I'm fairly certain the creators did not see. I'm running Ubuntu 7.06, 64 bit. I just tested one of the games exhibiting this bug in my 32 bit chroot (ubuntu 6.06) and did not experience it, so I'm thinking it's probably a 64 bit issue:

The symptoms of this bug are that all game objects seem to be miss-positioned, piling up in the upper right corner of the screen. I'm not sure what the common thread here is, but I experience this bug on at least the following games:

twisted zombie, twisted twister, Hampster Smash, and murmel labyrinth.

Is anyone else seeing this problem and/or have any idea what all these games have in common that the others do not?

(log in to comment)

Comments

alex on 2007/09/15 03:49:

I can reproduce similar problems (upper-left corner, or upper-edge) with Python 2.5 on Gentoo 64-bit.

I'm going to take a stab and guess that the problem is with PyOpenGL (3.0.0a6 under Py2.5 causing problem, 2.0.1.09 under Py2.4 is ok).

Tigga on 2007/09/15 13:10:

I don't think it is PyOpenGL. A friend described the same problem as this with a game that was only using pygame/numpy. He was on a 64-bit system though I hadn't connected it to the problem.

dmoisset on 2007/09/15 13:24:

twisted zombie does not use OpenGL, so I guess the problem is elsewhere

craigm on 2007/09/17 22:31:

I can vouch that it's not an issue with PyOpenGL, because Twisted Twister does not use OpenGL at all. Perhaps there's something I'm doing wrong with either the random number generator, or something else.

alex on 2007/09/18 11:08:

I've narrowed it down to a Python 2.4/2.5 regression in pygame.Surface.get_rect:: 

% cat test.py
import sys
import pygame
s = pygame.Surface((10, 10))
print 'Python', sys.version
print 'PyGame', pygame.__version__
print 'Result', s.get_rect(bottom=100)


% python2.4 test.py
Python 2.4.2 (#1, Oct 30 2005, 19:35:03) 
[GCC 3.4.3 20041125 (Gentoo 3.4.3-r1, ssp-3.4.3-0, pie-8.7.7)]
PyGame 1.7.1release
Result <rect(0, 90, 10, 10)>

% python2.5 test.py
Python 2.5b3 (r25b3:51041, Aug 13 2006, 17:18:17) 
[GCC 3.4.4 (Gentoo 3.4.4-r1, ssp-3.4.4-1.0, pie-8.7.8)]
PyGame 1.7.1release
Result <rect(0, 0, 10, 10)>

Both versions PyGame installations were built from the same (cleaned) source. Both Pythons are 64-bit (sizeof(c_void_p) == 8). Same problem can be reproduced in PyGame 1.8.0pre1 (the source at this function is unchanged). There is no problem directly manipulating `bottom` in a Rect.

Next step is probably to start tracing through surf_get_rect in gdb...

alex on 2007/09/18 11:33:

Solved; surface.c:1356

int pos=0;

should be 

long pos=0;

(The fact that it works in Python 2.4 is a stack fluke). By examination of related compile warnings, there are similar bugs throughout the rect, mixer, event, image and time extensions.

JuanjoConti on 2007/09/19 16:28:

Wow, weird bug. Some people told me that they have that problem running the game whith Python 2.5 but it was fine when running it with Python 2.4. I am not 100% sure but I think they have a 64 bits Kubuntu installation. Greets!