Hi,
On Thu, 7 Mar 2002, Rainer Canavan wrote:
> > evidence of a more strict (or less strict, depending on your POV)
> > convention, we'll have to review pretty much all of the code to look for
> > similar potential problems.
>
> I don't know the spec (somebody should check),
IIRC it's a convention rather than part of the standard, but I don't know
for sure either.
> but for now, I'd consider
> this a bug in the irix libc...
[...]
Well, OK. The most critical places- underscored symbols with few
characters- should have been fixed now.
[...]
> It does, I just tried linux/x86.
In that case, it's likely to be an interesting bug. Please try starting up
the interpreter with '-D', then tell it to 'set debug_flag 1' and to 'go'.
It should now print every operation it executes.
It'd be best if you could do this on your other box as well (with the same
version of LSL3).
> > > This is an O2, which offers lots of
> > > visuals to choose from. FreeSCI's speed, however is _almost_
> > > right without MIT-SHM and most of the beautifications turned on.
> >
> > When auto-choosing (i.e. when -c is not used), FreeSCI first tries 32bpp
> > then 24bpp etc. Running with -c16 should work fine on Xsgi; it
> > might be faster and shouldn't look too much worse.
>
> Doesn't work. Visuals are depth 8, 12, 15 and 24, with depth==bpp for all
> except 15 (bpp=16) and 24 (bpp=24 or 32). I think freesci tries (depth=16),
> which, obviously, must fail.
Some work was done on this in CVS, a month or two ago. Have you tried this
with the CVS version yet?
> I almost forgot to mention the various quirks I encountered when building
> freesci: On my linux/x86 box, I had to manually add "-lXft -lXrender
> -lfreetype",
Did you try using -lXft on its own? We do check for Xft.h, but our
autoconf script for this isn't particularly sophisticated.
> and on the Irix box, -lreadline implies -lncurses, causing lots
> of warnings if one adds -lncurses. There seems to be a linker bug on Irix
> that forced me to edit configure and instead of $ac_link call
> ${CC-cc} conftest.$ac_ext -o conftest${ac_exeext} $CFLAGS $CPPFLAGS
> $LDFLAGS $LIBS in the curses test. The "conftest.$ac_ext" needs to go first,
> otherwise the linker fails ;-(
Hmm, this is generated code. I'll try to see if I can modify it according
to this.
> Oh, yes, and there is the fact that char is unsigned (sizeof(char)=1) by
> default with my sgi, thus
[...]
> - char file; /* Number of the resource file this resource is stored in */
> + signed char file; /* Number of the resource file this resource is
Ah, thanks for thinding that one! Will fix this.
[...]
> is needed to make the comparisons to SCI_RES_FILE_NR_PATCH functional.
> There may be more like this,
That's possible. While many of these have been fixed, thanks to a PPC
user, none of the core developers (IIRC) uses a system with this setup
regularly, so that's certainly something to look out for.
Thanks for your help!
llap,
Christoph
