Subject: Hrm...
To: None <netbsd-help@netbsd.org>
From: Richard Rauch <rkr@olib.org>
List: netbsd-help
Date: 09/10/2003 06:11:31
I wound up with a Mesa update recently.  A few things that I thought that
I'd float on the bit stream (this relates to NetBSD/i386 1.6, using alternately,
the Mesa that ships with the native XFree86, and the Mesa from pkgsrc):

 * The new GLU does not define GLU API revision.  This causes, e.g.,
   one of the Mesa demos (at least) to decide not to work.  The demo
   wants GLU 1.2 compatibility.  The GLU in pkgsrc claims to offer that,
   but doesn't provide the #define.  The GLU in /usr/X11R6/ seems to
   be for the 1.3 API (and 1.2 and 1.1...).

   Is this a bug, or is there a good reason why Mesa requires a less
   featureful, non-compliant GLU, but we can have a "better" one in
   /usr/X11R6/?

 * A toy program that causes the GLUT to crash used to produce a
   4MB .core file.  The program wasn't all that big, so that's a little
   shocking...

   With the new Mesa, the same program produces a 44MB(!!!) .core file.

   Yes, ten times as big.  I thought that my system might have locked
   up when the app crashed, since the app neither responded nor immediately
   vanished.  But, the mouse pointer still moved, and after a moment, its
   window vanished.  I was floored by the size of the .core file, though.

   This is the exact same program, relinked to the new Mesa
   (-{L,Wl,-R}/usr/pkg/lib before -{L,Wl,-R}/usr/X11R6/lib).

   (Err, yuck, that looks confusing, mixing csh {a,b} with GCC's
   weird -Wl,-R option.  (^&  But you know what I mean...)

   (No, I don't use the csh these days, but the syntax *is* convenient
   on occasion...)

 * Interestingly, if I link against the new Mesa, I seem to be able to
   omit the $(XLIBS).  (Maybe that's a clue to the .core file explosion?)

 * Mesa 5.x seems to be a little slower than the Mesa in the X server.
   (There are, of course, many reasons why this might be, including
    simply that the programs that I saw want to be tuned a little,
    or even that *correct* behavior in Mesa requires less efficient
    performance.  But the old version of Mesa, itself, never crashed
    for me and seemed correct enough...(^&)

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/