Re: segfault in Xnest

Le ven 14/11/2003 à 21:16, Eric Anholt a écrit :
> On Fri, 2003-11-14 at 03:04, Benjamin Dauvergne wrote:
> > I wanted to test the xcompmgr and so i compiled xserver with
> > --enable-composite, but it seems some structure initialization is
> > lacking. Here is the output of valgrind:
> > *********
> > ==3187== Use of uninitialised value of size 4
> > ==3187==    at 0x80AE662: compCheckWindow (compwindow.c:36)
> > ==3187==    by 0x8074703: TraverseTree (window.c:223)
> > ==3187==    by 0x80AE83F: compCheckTree (compwindow.c:64)
> > ==3187==    by 0x80AEFB2: compCreateWindow (compwindow.c:299)
> > ==3187==
> > ==3187== Jump to the invalid address stated on the next line
> > ==3187==    at 0x0: ???
> > ==3187==    by 0x8074703: TraverseTree (window.c:223)
> > ==3187==    by 0x80AE83F: compCheckTree (compwindow.c:64)
> > ==3187==    by 0x80AEFB2: compCreateWindow (compwindow.c:299)
> > ==3187==    Address 0x0 is not stack'd, malloc'd or free'd
> > Erreur de segmentation
> > *********
> > compwindow.c:36 is this line, it's GetWindowPixmap that is
> > uninitialized:
> > *********
> > PixmapPtr   pWinPixmap = (*pScreen->GetWindowPixmap) (pWin);
> > *********
> > 
> > I'm using CVS head with jhbuild.
> 
> I was told on IRC that the composite extension requires render, which is
> unsupported in xnest.  It doesn't check that render's actually enabled,
> so it crashes.  I would love for somebody to fix the render support, but
> I don't know what's needed.
I have traced a bit more the problem. It seem that the only place where
the field GetWindowPixmap of the Screen structure is defined, is in
fb/fbscreen.c. It depends upon the defines FB_OLD_SCREEN which itself is
defined in fb/fb.h if TEKX11 is defined, but TEKX11 is never defined
either in a Makefile or a header. And the Render extension never touch
GetWindowPixmap.

What does the other servers do to not segfault when composite extension
is activated? I don't understand.

@+