Bug reported recently by both Emanuele Rocca and Joel Bertrand on the Debian BTS. They say:

On their sparc machine XineramaIsActive() returns True, while XineramaQueryScreens sets the number of screens to 0 and returns a NULL pointer. This bug makes gdm unusable, and lots of other applications calling Xinerama functions.

Emanuele tried the following program on sparc and i386:

  #include <stdio.h>
  #include <X11/Xlib.h>
  #include <X11/extensions/Xinerama.h>
  int main() {
    Display *disp = XOpenDisplay(NULL);
    int n_screens;
    XineramaScreenInfo *xscreens = XineramaQueryScreens (disp, &n_screens);
    printf("XineramaIsActive: %d\n", XineramaIsActive(disp));
    printf("Number of screens: %d\n", n_screens);
    if(xscreens == NULL)
      printf("XineramaQueryScreens returned a NULL pointer\n");
    else
      printf("XineramaQueryScreens returned something useful\n");
    XFree(xscreens);
    return 0;
  }

The program above prints the following on a sparc machine:
  XineramaIsActive: 1
  Number of screens: 0
  XineramaQueryScreens returned a NULL pointer

Conversely, on i386 the bug is not reproducible:
  XineramaIsActive: 1
  Number of screens: 1
  XineramaQueryScreens returned something useful

As a wild guess, I would say that the bug is caused by the server (possibly since RandR-1.2 arrived) assuming that all drivers support Xinerama while they probably don't. For instance, all sun* drivers I have looked at do not have any reference to Xinerama in their code (which makes me think XineramaIsActive should not return 1 on sparc), while the ATI driver has a RADEONProcXineramaDispatch.

It could also be related to Xserver 1.3 apparently refusing the disable the Xinerama extension.