Re: [tuxracer-main] Trouble getting tuxracer to start.

[Steve B. <sjb...@ai...>]

Oh - I feel <sick> - something very nasty just happened.

OK - it's been a while since this thread fizzled out - so I'll quickly
recap:

Ryan wrote:

> Ive installed the data in /usr/local/share/tuxracer and compiled (and
> installed) tuxracer-0.12 with no errors (maybe just a few warnings) and when
> i type 'tuxracer' from, say an xterm, I get...
>
> root@scrumpy:~# tuxracer
> BUG IN DYNAMIC LINKER ld.so: dl-version.c: 210: _dl_check_map_versions:
> Assertion `needed != ((void *)0)' failed!

I expressed doubt that it could be a problem with tuxracer - and indeed,
just doing an 'ldd' on the tuxracer executable was enough to produce
the exact same message.

I wondered whether Ryan's older 1.9.11 linker was maybe not playing
nicely with his ultra-modern 2.2.14 kernel - or whatever compiler
revision came with his Debian Linux.

There was also this strange compiler error:

Ryan said:

> c++ -DHAVE_CONFIG_H -I. -I. -I..     -Wall -pedantic -O2 -c
> course_quad.cpp
> In file included from tuxracer.h:46,
>                 from course_quad.cpp:2:
> /usr/local/include/GL/glut.h:202: warning: declaration of `exit(int)'
> throws different exceptions

Finally, Ryan said:

> OK, well yes I have Debian potato which is the latest stable (I belive) but
> I have never had any problems with it before with the dynamic linker.
> Basicaly I installed the base system and then in a more slackware like
> fashion did everything else as tarballs.  The compiler is the latest stable
> of gcc (which of the top of my head is 2.9.75 or something like that)  and
> my glibc is i belive 2.2 or 2.1   and my kernel is 2.2.14     now the only
> one of those things that actually came with the debian install was the
> glibc.   At any rate, is it possible to update my ld.so and not have major
> library dependancy errors? I can see how it would be a problem, but that's
> just me.   Hmm also I think I will need to take a look a GLUT perhaps.

OK - so that brings us up to date.

Now, I've just upgraded both of my PC's to SuSE 6.4 - which uses 2.2.14
kernel (same as Ryan's) and ldd version 2.1.3.

My PC's are called 'batcave' and 'waynemanor':

  batcave: 450MHz AMD with a GeForce card and 64Mb RAM.
  waynemanor: 266MHz AMD with a Voodoo-3 and 64Mb RAM.

They are running more or less identical SuSE 6.4 loads - installed off
the same CD-ROMs.

Now, I compiled a program of my own (an early version of my upcoming
'tuxkart' game) on batcave - and it ran *perfectly*.  When I tried
to run it on waynemanor (without recompiling it - the two machines
share disk space via NFS) I got the EXACT SAME LINKER ERROR AS RYAN.

On batcave:

  batcave% ldd /u/tuxkart/src/tuxkart
        libglut.so.3 => /usr/lib/libglut.so.3 (0x40015000)
        libGLU.so.1 => /usr/lib/libGLU.so.1 (0x40048000)
        libGL.so => /usr/lib/libGL.so (0x40065000)                              
        ....

But on waynemanor:

  waynamanor% ldd /u/tuxkart/src/tuxkart
        BUG IN DYNAMIC LINKER ld.so: dl-version.c: 210:
        _dl_check_map_versions: Assertion `needed != ((void *)0)'
        failed!

So:

1) We can definitely eliminate a bug in tuxracer - since tuxracer
   and tuxkart don't share any code. (I suspected as much before)

2) We can eliminate Ryan's strange GLUT problem and the old version
   of ldd he seems to have - although both are 'fishy'.  My machines
   are both running the exact same (and bang up to date) GLUT and
   ldd.

3) Actually, we can eliminate everything except the CPU types I have
   in my two PC's - they are virtually identical machines with virtually
   identical software loads.

What I now suspect is the little-know (but bloody annoying) bug in
some AMD K6-2 CPU's.  I've never seen it impact ldd before - but
it regularly causes peculiar and intermittant compilation errors.
You run a 'make', the compiler dies with some internal error message,
you re-run it and it 'goes away'.

However, the nature of the glitch in the CPU is such that in principal,
almost any program could fall foul of it - although it's an astronomically
rare problem (no Windoze program has ever been reported to fail for this
reason - and the only Linux program that ever reported a bug is the
gcc compiler).

Oh - and the problem only shows up when you have >32Mb RAM.

Well, perhaps this problem is showing up with the linker under
2.2.14 kernels - and perhaps it's not intermittant for some programs.

So - the big question is - what CPU does Ryan have - and at what
clock rate - and with how much RAM?

If it turns out that this is indeed the problem, there are only
three possible fixes (AFAIK):

1) Change your CPU chip.  AMD did promise to replace affected
   CPU's at no charge - but when I tried to take them up on
   that offer, they claimed that the bug only affects old
   CPU's running at 133MHz and that they'd fixed it in 233MHz
   parts.  There are a couple of web sites out there where
   people claim they see it on 233 and 266MHz parts - but AMD
   won't play ball.

2) Downgrade your memory to 32Mb.  Uck!

3) There is a kernel patch (for 2.1 series kernels I believe)
   that does a *DISGUSTING* kludge to try to reduce the
   probability of the bug.  I don't know if that works with
   2.2.14 - and it seemed to me that the kludge had the possibility
   of slowing memory allocation down noticably.


-- 
Steve Baker                  http://web2.airmail.net/sjbaker1
sjb...@ai... (home)  http://www.woodsoup.org/~sbaker
sj...@ht...      (work)