Closed Bug 11634 Opened 27 years ago Closed 25 years ago

Mozilla refuses to run on Linux libc5 (and others) with user threads

Categories

(SeaMonkey :: General, defect, P3)

Other
Linux
defect

Tracking

(Not tracked)

RESOLVED WORKSFORME

People

(Reporter: tenthumbs, Assigned: tenthumbs)

References

Details

(Keywords: platform-parity)

Attachments

(1 file)

If I build without pthreads, Mozilla just hangs; always in a select call. I have seen reports that this also happen on glibc Linux as well as other platforms. It started with the Necko landing. Are user threads no longer supported?
While testing of NSPR is typically done with native threads only (pthreads, Win32 threads, etc), the user-level threads should also work. However, there are restrictions to be followed when using user-level threads - because the user-level threads are scheduled by NSPR, any blocking call made by the application will block the entire process - integration with GUI libraries, such as Motif, needs special attention. The GUI library may not be multi-threaded at all, or may be ported to native threads only. You can try to run the NSPR test programs before testing Mozilla with user-level threads. Then you can trace system calls and see if the select call is timing out. Some gdb macros for debugging user-level threads were posted to the mozilla.nspr newsgroup recently; you can list stacks of threads to see what is happening in the application.
Status: NEW → ASSIGNED
Attached file test results —
All the tests succeed except possibly priotest and ranfile. These two appear to never end. They are not hung because they're doing something. It's not clear to me how long they should run so I can't tell if there's a problem here. As for Mozilla, attached are some results from a build from yesterday (August 15). They show that it hangs is a select call. The stack traces seem rather bogus but I suppose there may be some gdb weirdness. This is on a Linux libc 5 with no pthreads at all.
*** Bug 11925 has been marked as a duplicate of this bug. ***
Here's some new data. Looking in /proc/<apprunner_pid>/fd, you can see that descriptors 8 and 9 appear to be two ends of a pipe. Descriptor 10 looks to be a unix socket, at least according to netstat. Presumably fd 10 is talking to the X server. lr-x------ 1 10 users 64 Aug 16 13:46 8 -> [0000]:79998 l-wx------ 1 10 users 64 Aug 16 13:46 9 -> [0000]:79998 lrwx------ 1 10 users 64 Aug 16 13:46 10 -> [0000]:80007 Active UNIX domain sockets (w/o servers) Proto RefCnt Flags Type State I-Node Path unix 2 [ ] STREAM CONNECTED 80007 unix 2 [ ] STREAM 80009 /tmp/.X11-unix/X0 If all this is accurate, then the question is who is writing to fd 9 or fd 10? Another question is whether it should select for errors on fd 10?
The NSPR test program 'priotest' may not complete on some platforms because of mismatched thread priorites. The 'ranfile' test program should eventually complete. I don't think these test results explain why the apprunner hangs at startup.
Hardware: PC → Other
Summary: Mozilla refuses to run on Linux libc5 (and others) with user threads → [PP]Mozilla refuses to run on Linux libc5 (and others) with user threads
Putting on [PP] radar.
There isn't any information available so far to show that a NSPR bug is the cause of this problem; this problem first needs to be debugged in terms of the application events excpected at the point at which the hang occurs. For example, if Mozilla hangs during the startup sequence, you may want to study the module during the initialization of which the hang occurs. Or, if the hang occurs later on, say when fetching a URL, you may want to find out when in the sequence of I/O events the hang occurs. The NSPR scheduler blocking in a select call is a normal event; this happens when no thread is runnable (as all threads could be waiting for I/O completion, user-input event, synchronization event, etc.) How exactly do you want to proceed with debugging this problem? It isn't clear what should be done at the NSPR level.
With the new ability to switch toolkits at run-time, new information appears. With the gtk toolkit Mozilla dies in the usual place; with the xlib toolkit it actually runs until a window appears. (It segfaults there but I'm not sure that's related.) It would appear that this is perhaps, just maybe, not NSPR's fault. I'm not willing, however, to absolve NSPR completely until an explanation surfaces for why a pthreads version seems to work. More investigation is in progress.
Is there any more information available on this issue? There hasn't been an update in quite a while.
Unfortunately, no. I have had very little time recently to look into this. The problem still exists as of October 18 which is the last time I built Mozilla.
Assignee: srinivas → tenthumbs
Status: ASSIGNED → NEW
Component: NSPR → ActiveX Wrapper
Product: NSPR → Browser
Version: 3.0 → other
I am reassigning this bug to you as there is no information related to a bug in NSPR; leaving this bug report assigned to the NSPR module will lead people to believe that this bug is being worked on by the NSPR module owner/peers.
Status: NEW → ASSIGNED
Keywords: pp
Summary: [PP]Mozilla refuses to run on Linux libc5 (and others) with user threads → Mozilla refuses to run on Linux libc5 (and others) with user threads
Changing component from ActiveX to Browser-General
Component: ActiveX Wrapper → Browser-General
enthumbs@cybernex.net: You are assigned to this bug. Can you close this bug or add a comment that you are working on this bug ? (Your Last comment : 1999-10-29 08:25) Thanks Matthias
Status: ASSIGNED → RESOLVED
Closed: 25 years ago
Resolution: --- → WORKSFORME
This bug seems out of date no answer from tenthumbs@cybernex.net marking worskforme (please reopen if you disagree)
Product: Browser → Seamonkey
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: