3405 - [PP]m3s: Mac crashes going to webpages not on local disk

Status: VERIFIED FIXED

(Core Graveyard :: Tracking, defect, P1)

Description

[Mike Pinkerton (not reading bugmail)]

Mac crashes when visiting any webpage not on the local drive. It dies in a
callback probably in an interrupt handler. THe machine is toast at this point.

Comment 1

[chris hofmann]

danm suggests this is machine dependend and we started to see
different behavior following scc's checkin...

any more ideas on where to start the investigation on this?

Comment 2

[Mike Pinkerton (not reading bugmail)]

I saw this in the morning before scc's checkins.

Comment 3

[Pierre Saslawsky]

I heard that the Unix folks had a problem that was fixed after scc's checkin too.
Maybe they have more info...

Comment 4

[sar]

Linux had a problem of crashing on the second url loaded. When we pulled to get
scc's checkin stuff, we also got errant checkins by troy and buster. It is
concievable to me that some of those changes could have had an effect on the bug
we were seeing.

After clobbering and rebuilding, the apprunner can now visit >2 url's and be
happy about it. I don't know what fixed it.

troy touched the following 5 files in mozilla/layout/html/style/src/
nsCSSFrameConstructor.cpp
nsCSSFrameConstructor.h
nsTableFrame.cpp
nsTableRowGroupFrame.cpp
nsTableRowGroupFrame.h

tomorrow, when I have a clean pull and build again, we will see if we have the
same problem. It is possible that the act of clobbering and rebuilding could
have fixed it as well. I really don't know.

But this is unrelated to the mac crashing bugs

Comment 5

[don]

Changed component to Apprunner.

Comment 6

[mcmullen]

Chris Saari is already investigating this, it seems.  Reassigning to him.

Comment 7

[Simon Fraser [no longer active]]

OK, we have a fix for this one now. It was the result of an apparently innocuous
warnings fix on March 3, in mozilla/lib/mactime.c. The warnings fix changed
macro expansion in a way that caused a routine to call itself recursively; the
offending macro is
#define ctime(t)	Macctime(t)
in xp_mcom.h

So the crash occurred because tight recursion blew the stack very quickly.

Comment 8

[Simon Fraser [no longer active]]

sdagley is checkin in the fix right now.

Comment 9

[chris hofmann]

Yes!  great find...   Thanks for the hard work in this one.

Comment 10

[chris hofmann]

Yes!  great find...   Thanks for the hard work in this one.

Comment 11

[mcmullen]

Great.  But two things:

1.  There are three files called mactime.c in the tree.  One of them is in NSPR,
and is in use.

2.  It's "daylight-saving time".  (It's not a man-eatings tiger, but a man-eating
tiger, see).

Comment 12

[Simon Fraser [no longer active]]

Yeah, and kudos to the person who #defined an existing routine name to something
else. That's always a bad idea. The real fix for this problem would have been
to change that #define.

Comment 13

[Eli Goldberg]

Verified fixed using 3.9.99 build on Mac OS (Apprunner checked).

Broke side-issue raised by Simon/John into 3559.

Comment 14

[leger]

Moving all Apprunner bugs past and present to Other component temporarily whilst
don and I set correct component.  Apprunner component will be deleted/retired
shortly.