33329 - mozilla hangs when start-psm fails

Status: VERIFIED WORKSFORME

(Core Graveyard :: Security: UI, defect, P3)

Description

[Dan Mosedale (:dmosedale, :dmose)]

Red Hat Linux build from the tip:

	 --enable-x11-shm --enable-optimize --disable-debug CFLAGS="-g"

When start-psm fails, the browser hangs completely.  It should instead fail
gracefully and continue to work so that data and session info in other windows
is not lost.

Comment 1

[Doug Turner (:dougt)]

dmose,

did you have psm installed?  did a dialog every appear asking you for the 
location of psm?  

Did you see anything appear in the console?  Specifically "*** Failure setting 
up Cartman! \n"

Comment 2

[Dan Mosedale (:dmosedale, :dmose)]

I did have it installed, and start-psm was executing, but psm itself was failing
because of a dynamic-linking problem.  bug 33331 contains more info about the
specific problem.

Comment 3

[Dan Mosedale (:dmosedale, :dmose)]

And no, a dialog did not appear, because it had already found {start-,}psm, and
the only thing on stderr was related to the dynamic linking error after fstat()
not being found in libnspr3.so.

Comment 4

[Doug Turner (:dougt)]

dmose, just to reiterate, you did install psm via xpinstall?

It might be a problem with the cmt client lib's CMT_EstablishControlConnection() 
call.  Attached is a diff that fixes(?) a couple problems.  

a) Instead of calling system(), it might be better to call nspr's 
PR_CreateProcessDetached()

b) it removes the wait after launch code (which looked wrong for unix) and just 
does exactly what the comment states "Now try to connect to the psm server.  We 
will try to connect a maximum of 30 times and then give up."

dmose, could you try this diff?  Mark can you review/comment on this diff?

Comment 5

[Doug Turner (:dougt)]

Attachment: changes to how psm is launched

Comment 6

[Mark Welch]

I'm concerned about this diff as written, because it creates a dependency on 
NSPR within the PSM client library itself. How hard would it be to move the NSPR 
bits into a separate routine, such as we do now with the socket, read/write, 
etc. methods? Perhaps a Launch method?

Comment 7

[Doug Turner (:dougt)]

Oh, sure.  I didn't think about that dependancy.  

dmose, could you just apply the patch which removes the unix specify wait while 
launch code to see if that does it?

mark, maybe it is better if I just launch psm from my code and not use 
EstablishControlConnection?

Comment 8

[Dan Mosedale (:dmosedale, :dmose)]

Yes, I used XPinstall.  Applying just that part of the patch brings up a new
error:

/mozilla-bin: error in loading shared libraries:
/u/dmose/s/browser-main/mozilla/obj-i686-pc-linux-gnu/dist/bin/libcmt.so:
undefined symbol: Sleep

The place I see Sleep() defined is as part of nsIThreadManager which is from the
OJI world.

Comment 9

[Doug Turner (:dougt)]

over to [e-mail].

Comment 10

[Mark Welch]

I'm leaving Netscape, so I'm reassigning all my pending bugs to David Drinan.
I've added "[from [e-mail]]" to the status whiteboard for each of
these bugs for easier searching.

Comment 11

[John Unruh]

Setting milestone to M16.

Comment 12

[Phil Peterson]

PDT would like to know how likely a user is to see start-psm fail before making 
a +/- call for nsbeta2

Comment 13

[leger]

Putting on [nsbeta2-] radar. Not critical to beta2.  Putting on nsbeta3 radar 
for a lookee then :-)

Comment 14

[Mike]

M16 has been out for a while now, these bugs target milestones need to be 
updated.

Comment 15

[John Unruh]

Marking worksforme. I haven't seen this problem.

Comment 16

[John Unruh]

Verified.

Comment 17

[John Unruh]

Mass changing Security:Crypto to PSM

Comment 18

[John Unruh]

Mass changing Security:Crypto to PSM