39238 - leaks and slowdown when reloading about:blank

Status: RESOLVED WORKSFORME

(Core :: DOM: Core & HTML, defect, P2)

Description

[Marko Macek]

if I enter "about:blank" into location bar and keep pressing Enter, mozilla will
leak memory and become slower and slower.

Reproduced on linux and win98. (nightly builds)

On linux profiled with jprof:

It get's slower because JS_GC takes longer and longer. If I remove
calls to JC_GC in SetNewDocument and SetupNewViewer, the speed becomes constant
(and 25% faster).

Comment 1

[Jerry Baker]

Confirmed in Windows 2000. Although the slowdown is only noticeable because of 
the elapsed time displayed in the status bar. The memory footprint of Mozilla 
was increased by 4MB after pressing ENTER more than 200 times.

I'm sure this is a dupe, but I do not know where.

Comment 2

[Marko Macek]

On Linux it is quite easy for me to get to several seconds per load of about:
blank. GC seems to be noticeably slower on Linux than under Win98 on the same
machine.

Comment 3

[Bozhan Boiadzhiev]

NS 4 look better whitout this thing
now it looks like IE

Comment 4

[Phil Schwartau]

This bug is mentioned as a variant of bug 42321, which has recently
been reassigned from JS Engine to DOM Level 0. 

Would also like to redirect this to DOM Level 0 to be consistent -

Comment 5

[Johnny Stenback (:jst)]

Nominating for nsbeta3.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I may have a fix for this leak... I need to check to see if it's fixed in my
tree.  (It might have been the DOCTYPE node leak...)  That leak would explain it
taking more time in GC, since it was a leaked root that left lots more JS
objects around...

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Did you see lots of leaks according to leak logging (setting XPCOM_MEM_LEAK_LOG
= 1), or were these leaks things that were held too long and all cleaned up at
the end?

I leak a little image and network stuff when loading about:blank, but that seems
to be it.  Were you seeing more?  (Or are those XPCOM objects holding on to lots
of other data?)

Comment 8

[Nisheeth Ranjan]

Marking nsbeta3- for now, given David's comment about leaking only a few objects 
on about:blank load.  If you want us to re-evaluate based on new information, 
please remove the nsbeta3- marker from the status whiteboard and add a comment.  
Thanks!

Comment 9

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I am still seeing a slight slowdown and a little leaking memory.

Comment 10

[Judson Valeski]

The leaks we're seeing here could be leaking other objects. If we're aware of 
leaks I'd like to see them persued. asking for re-consideration.

Comment 11

[Nisheeth Ranjan]

Marking nsbeta3- for now.

Comment 12

[ekrock's old account (dead)]

Removing [egk].

Comment 13

[Johnny Stenback (:jst)]

This should be a lot better by now, can someone actually still reproduce the
slowdown by reloading about:blank?

Comment 14

[Johnny Stenback (:jst)]

I don't have any time to spend on this untill sometime later, moving to Future.

Comment 15

[Jerry Baker]

Mass removing self from CC list.

Comment 16

[Jerry Baker]

Now I feel sumb because I have to add back. Sorry for the spam.

Comment 17

[mikol]

Attachment: Test Case: Reloading an iframe w/setInterval.

Comment 18

[mikol]

I was investigating a similar problem today. Reloading an iframe with
setInterval causes Mozilla 1.1 Build 20020826 to leak a few K of memory every
couple of seconds on Windows and Linux. After creating a reproducible test case
(see Attachment #101171 [details]), I downloded Mozilla 1.2 Build 2002093010 for Windows
and used the test case. Build 2002093010 does not exhibit the memory leak I saw
with Build 20020826.

Comment 19

[Johnny Stenback (:jst)]

Mass-reassigning bugs to [e-mail]

Comment 20

[Cees T.]

Reloading leaks memory - bug 81446.

Possibly related: File | Quit takes a huge amount of time (exit or shutdown slow) - bug 39742.

Comment 21

[Steve Chapel]

Is anyone still able to reproduce this bug? Firefox 2 and Firefox 3 don't even seem to allow the user to reload about:blank, so what product(s) does this bug still apply to?

Comment 22

[AndrewM]

(In reply to comment #21)
> Is anyone still able to reproduce this bug? Firefox 2 and Firefox 3 don't even
> seem to allow the user to reload about:blank, so what product(s) does this bug
> still apply to?

Given that you can't reload about:blank anymore, I'd say this is INVALID.

Comment 23

[Steve Chapel]

I was able to test with a SeaMonkey 2 trunk build on Windows XP. After loading about:blank, I was able to reload by pressing Ctrl-R. After several minutes of holding down Ctrl-R, memory use went up and down a little, but didn't continually rise as would be the case with a memory leak. This one is WFM.