105726 - Core dump when going to home.netscape.com on HP-UX

Status: VERIFIED FIXED

(SeaMonkey :: General, defect, P1)

Description

[Shannon Diener]

Mozilla on HP-UX core dumps when going to http://home.netscape.com. 

I believe this started sometime this week and is only seen on the trunk.

Reproducible: Always
Steps to Reproduce:
1.launch Mozilla
2.go to http://home.netscape.com


Actual Results:  Mozilla core dumps

Expected Results:  http://home.netscape.com loads successfully

Comment 1

[Shannon Diener]

Attachment: stack trace

Comment 2

[Jim Dunn]

adding brendan since it looks like he just did 
a big checkin on 10/17 at 3am... Brendan?

Shannon can set you up/point you to a debug build.
wdb on hpux really rocks...

Comment 3

[Jim Dunn]

specifically he changes jsarena.c
Hey brendan did you really mean
*(JSArena ***)a->avail = ap on line 157????

Comment 4

[Brendan Eich [:brendan]]

Yes, I meant that.  Works great, except for this report.  I will help debug --
where and when?

/be

Comment 5

[Brendan Eich [:brendan]]

*** Bug 106568 has been marked as a duplicate of this bug. ***

Comment 6

[Brendan Eich [:brendan]]

Mine, something stupid I did.  Wonder why it works on the tier1 platforms?

/be

Comment 7

[Brendan Eich [:brendan]]

Attachment: proposed fix

Comment 8

[Brendan Eich [:brendan]]

I ass-u-med that the align parameter to all arenas satisfied pointer alignment
requirements, but it need not (it's 1 for the JS_NewPrinter-initialized arena
pool in a JSPrinter).  This tends to break RISCs and not CISCs, hence the sorry
effect on HP and Sun.

Please confirm that the patch cures all ills on both platforms, and I'll get it
reviewed and checked in.  Thanks,

/be

Comment 9

[Shannon Diener]

Looks good on HP-UX ... can now go to home.netscape.com and abcnews.go.com.

Thanks Brendan!

Comment 10

[Mike Shaver (:shaver emeritus)]

Comment on attachment 55347 [details] [diff] [review]
proposed fix

mmm, slop.  sr=shaver

Comment 11

[Brendan Eich [:brendan]]

Shaver, thanks for the fast sr -- sorry I didn't get the next patch in sooner,
as I realized late last night that the first patch adds unnecessary slop (can
there be such a thing?) if pool->mask is >= (JS_ALIGN_OF_POINTER - 1).  So the
second patch parameterizes those FOOTER macros with pool to compute a 0
FOOTER_MASK for pools with alignment sufficient for a pointer (as a->avail is
aligned to the pool's modulus).

So transfer your sr= and I'll get an r= and we can get on with our lives.  Thanks,

/be

Comment 12

[Brendan Eich [:brendan]]

Attachment: less sloppy proposed fix

Comment 13

[John Bandhauer]

Comment on attachment 55398 [details] [diff] [review]
less sloppy proposed fix

r/sr=jband

Comment 14

[Mike Shaver (:shaver emeritus)]

Comment on attachment 55398 [details] [diff] [review]
less sloppy proposed fix

sr/r=shaver.

Comment 15

[Brendan Eich [:brendan]]

Attachment: patch I checked in: comments, cosmetics (SLOT => SIZE), and *minimal* slop (see comment)

Comment 16

[Brendan Eich [:brendan]]

Fix is in.  Reviewers, sorry I didn't present the third patch from the get-go.

/be

Comment 17

[Barrett Lindstrom]

Verified on trunk build # 2001102821