That disaster....
Smells like the infamous netscape doesn't do locking with Xlib properly
bug...
Keith Packard and I looked into something that smells the same (triggered
frequently on my machine by a Java app) about 18 months ago, and tried to get
Netscape to fix it... They unfortunately never have.
The fault was netscape reentering Xlib without telling Xlib to
run with locking. Needless to say, things get confused, and
the result was netscape looping.
The bug is a heisenbug: can occur anytime, but certain timing tends to
trigger it differently. If it is the same bug, the difference is just
that the server is running at a different speed than it used to...
So if it is the same bug, there ain't a thing we can do about it...
It is buried in Netscape, a dead product.
Sigh...
- Jim
--
Jim Gettys
Cambridge Research Laboratory
Compaq Computer Corporation
[EMAIL PROTECTED]
> Sender: [EMAIL PROTECTED]
> From: Jeremy White <[EMAIL PROTECTED]>
> Date: 14 Feb 2002 22:12:45 -0600
> To: [EMAIL PROTECTED]
> Subject: [Xpert]Request for help on possible 4.1.0 bug
> -----
> We're trying to track down a regression that occurs when using
> Quicktime in conjunction with Netscape 4.72. It works
> on X 3.3.6, but fails on 4.1.0.
>
> The failure behavior is that Netscape, while loading a
> web page with Quicktime content, goes into a nearly
> infinite spin loop, calling select() again and
> again. The socket it is calling select
> on is connected to /tmp/.X11-unix/X0.
>
> An strace of the app as it runs sheds some clues,
> but I was hoping to ask for further help from
> the experts.
>
> Specifically, the behavior is that we see a
> write() to the X socket which returns an
> EAGAIN. Now, when I run strace on other,
> success, cases, I never see a write() return
> EAGAIN (I only see writev return EAGAIN). Shortly after
> the write() occurs, the app goes into the
> select() spin loop.
>
> I was hoping to ask for help in several ways.
> First, as a newbie to this list, please feel free to correct me.
> If I should post elsewhere, or if there is a FM I should
> go read; I will cheerfully go off and do more
> homework.
>
> Second, I was hoping to get help pinpointing
> where in the source a write() to the
> socket with a potentially buggy EAGAIN
> handler might be, so I can see if I can
> spot a bug. To that end, I was hoping someone
> could teach me how to decode the X protocol.
>
> I have from the strace log the header of the
> request written, but I don't know how to decode it;
> again, I'd appreciate pointers to TFM or
> source code.
>
> I have attached the relevant fragment of the
> strace log; the write call in particular is
> on line 5 (you can see the select() spin loop
> starting at line 27).
>
> Thanks for any help you can offer,
>
> Jeremy
>
> Enclosed text file: strace.fragment (9 KBytes)
_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert
