Re: Anyone have firefox working on i386?

[Anne Bennett <anne%porcupine.montreal.qc.ca@localhost>]

I'm going to summarize what I have so far on this problem; I don't
want to chase too many wild geese (and I'm hoping that my own goose
isn't cooked!).  On Feb 11 I wrote that firefox36 works on my i386
system, but firefox (6.0.2 from pkgsrc) fails to open a window,
and complains:

    Failed to initialize GEM.  Falling back to classic.
    failed to create drawable

The system runs 5.1_STABLE, patched regularly, the X server is the
one that comes with the O/S, the graphics card is "Intel 82945G/P
Integrated Graphics Device (rev. 0x02)", and it uses "i915drm0"
with "EXA" acceleration.  It cannot enable DRI2 because this would
require "UXA" acceleration.

I don't generally do anything fancy with graphics, but I can use
the GIMP and firefox36 just fine.

Other people report successfully using a similar system with a
post-3 firefox.

The system *is* able to run firefox 4 when booted into MythBuntu,
which has a more recent X server (which may or may not be relevant).

Also, I can run a post-3 firefox on another host (Linux) and display
it locally via ssh X forwarding; sadly, this is unusably slow.
But it does suggest that firefox is *not* doing anything weird in
terms of direct access to the graphics card, which seems an unlikely
scenario in any case (but I've been grasping at straws!).

Avenues explored so far:

- The message "Failed to initialize GEM.  Falling back to classic."
  Also appears when I start "glxgears", which then seems to work
  fine, so is probably a red herring.

- The message "failed to create drawable": Greg Troxel reports
  seeing it but firefox 8 still runs.  A Google search suggests
  that it comes from Pango.

- It was suggested that my X configuration might be incorrect or
  too fancy.  I've tried running with the plain config obtained
  via "Xorg -configure"; it makes no difference.

- I thought maybe the missing DRI2 might somehow be at fault,
  but I can't seem to enable it (turning on UXA results in a
  non-functional X).  But why would this bother firefox 6 and not
  firefox 3?

- This isn't caused by a personal configuration to firefox, or
  misbehaving plugins: if I move the .mozilla directory aside, then
  I see two sets of the error messages reported above (and a fresh
  .mozilla is created and populated), but I get no further ahead.

- I resorted to "ktruss" to try to identify the problem, but it's
  hard for me to differentiate between non-serious errors and serious
  errors.  These seemed perhaps problematic:

    4675  1 xulrunner-bin write(0x5, 0xbfbfd270, 0x4c) = 76
            "X error occurred in GLX probe, error_code=9,
            request_code=128, minor_"
          (Google search comes up with a page at Mozilla that says
          "nothing to worry about. This just means that you won't get
          graphics hardware acceleration because your system doesn't
          support "recent" GLX features")


    1663  1 xulrunner-bin sendto(0x17, 0xb971f69f, 0x13, 0x400, 0, 0) = 19
              "NEGOTIATE_UNIX_FD\r\n"
    1663  1 xulrunner-bin poll(0xbfbfd32c, 0x1, 0xffffffff) = 1
    1663  1 xulrunner-bin recvfrom(0x17, 0xb8c35000, 0x1000, 0, 0, 0) = 25
              "ERROR "Unknown command"\r\n"
          (The above command seems to be part of dbus; a "truss" of
          the functional firefox36 shows the same interaction, so this
          is a red herring too.)


What's left?  Could it be that at some point an ill-conceived "make
replace" in pkgsrc put libraries out of synch with each other in some
subtle way?  Perhaps I should try rebuilding or replacing all of my
packages, but that would be painful.  :-(

For now, I'm going to try comparing the ktruss output from both
firefoxes, and hope something strikes me as significant...


Anne.
-- 
Ms. Anne Bennett, as a private citizen:  anne%porcupine.montreal.qc.ca@localhost
Also reachable more officially at work:  anne%encs.concordia.ca@localhost