23187 - document.write() is slow

Status: RESOLVED FIXED

(Core :: Layout, defect, P3)

Description

[ekrock's old account (dead)]

On WinNT 4.0 SP3, I get the following stopwatch timing results for loading a
page whose body contains only a for loop that does 1000 document.write()
statements:

Nav4.7 - 2 sec
IE5.5beta - 1 sec
M12 1999122023 - 15 sec

The performance of document.write() statements in SCRIPTs in the body must be
improved. Will attach test case.

Comment 1

[ekrock's old account (dead)]

Marking [TESTCASE].

Comment 2

[ekrock's old account (dead)]

Attachment: test case: for loop with 1000 document.write() statements

Comment 3

[troy]

Vidur, I'll leave it to you to decide whether this is because of DOM, the
content model, or the lack of batching for layout. If it is the later, then this
will be a good test case for Nisheeth

Comment 4

[vidur (gone)]

The latter is probably the biggest culprit. I'll let Nisheeth hold onto it as a
performance test.

Comment 5

[ekrock's old account (dead)]

Bulk moving [testcase] code to new testcase keyword. Sorry for the spam!

Comment 6

[Nisheeth Ranjan]

Accepting bug and setting milestone to M14...

Comment 7

[Nisheeth Ranjan]

Vidur, I'm re-assigning this to you so that you can close it out once you 
checkin your content sink changes.

Comment 8

[vidur (gone)]

It's much faster than it was, but is it fast enough? What say Krock? I checked 
in a fix to make it faster on 1/28/2000.

Comment 9

[vidur (gone)]

Moving this one along to a later milestone. I think the performance is much more 
reasonable now. The question is how good is good enough.

Comment 10

[Morten Wang]

pardon me for bumping in here.  I dropped by the BugAThon pages and found this 
bug, and document.write() speed is an odd hobby of mine.

over on http://home.sol.no/~warnckew/programming/javascript/document.write.html 
is some information about three tests I created about a year ago.  I altered the 
code slightly today so Mozilla M14 also runs it as it should.  I didn't shut 
down all applications before testing, but the result of a 150 table row test 
with the fastest output method (string concatenation directly in the write()) 
was:

Communicator 4.72: 2800-2890ms
Mozilla M14: 2410-2530ms
IE 5.0: 550-600ms

if you guys are looking to improve the speed further than the current I'll 
happily run it through all 9 tests I have (three different write methods and 
three types of content) to see how it works out.

Comment 11

[vidur (gone)]

It's interesting to see that we're comparable to 4.72 in your test - my guess is 
that the fact that it's a table is relevant. I think we should continue to look 
at document.write speed, so more data would definitely be helpful. Thanks!

Comment 12

[Morten Wang]

Attachment: write() test result, 2000-03-14

Comment 13

[Morten Wang]

I ran Mozilla, IE5 and NC4.72 through all tests today.  I wrote a test report in 
HTML and uploaded it as an attachment.

Compared to NC4.72 the result is good for the first test, but Mozilla is 
considerably slower on the two other tests.  IE5 nails every test, in my 
opinion.

What surprised me was the result of the second test when multiple 
document.write()s are used.  Mozilla suddenly spends 9 seconds, which is an 
speed decrease compared to test #1, while the other browsers have a speed 
increase.  M14 is still not as bad as IE4 though, which used 24 seconds on that 
test on a PII/266. :)

Overall Mozilla's performance is fair when the script is optimized for 
performance.

Comment 14

[Johnny Stenback (:jst)]

This bug has been marked "future" because the original netscape engineer
working on this is over-burdened. If you feel this is an error, that you or
another known resource will be working on this bug,or if it blocks your work
in some way -- please attach your concern to the bug for reconsideration.

Comment 15

[bht237]

The speed of any real world document.write() currently also depends on bug
61842, although the pure test case in THIS bug will probably not confirm this
because it uses a loop.
Bug 61842 deals with a shocking slowdown due to parsing of script that contains
HTML.
In other words: Even if the performance problem that this test case uncovers is
solved, then any real world application bug that use this bug for tracking might
not necessarily be resolved.

Comment 16

[vidur (gone)]

Johnny would like to hold on to this one.

Comment 17

[Markus Hübner]

what's the current status of this one?

Comment 18

[Pascal Chevrel:pascalc]

Here is two new testcases with a timer for this bug :

Test1 does 10000 document writes on body onLoad without HTML formatting:

for (i=0; i<10000; i++)
{ document.write (i+" ");
}

Test2 does 10000 document writes on body onLoad *with* HTML formatting:

for (i=0; i<10000; i++)
	{ 
	document.write ("<p>" + i + "</p>");
	}


Setup : athlon 1700XP-256MB-WinXP-build 2002122808

Results :
        Test1     Test2
Moz     0.969     12.141
IE6     0.375      0.437

The document.write loop is about 3x slower than IE6 with test1 but is about 28x
slower with test2 !

I believe that this second test (similar to the original testcase) is mainly
displaying a major layout performance issue which may not be related to
document.write.

Comment 19

[Pascal Chevrel:pascalc]

Attachment: Test1 testcase

Comment 20

[Pascal Chevrel:pascalc]

Attachment: test2 testcase

Comment 21

[Boris Zbarsky [:bzbarsky]]

test2 becomes at least 3 times faster if the proposed patch in bug 145425 is
checked in (the one that makes us not do an O(N^2) walk over all the crap you're
writing).

Comment 22

[Markus Hübner]

bz, can you tell how much perf gain it will have on the testcase in bug 
151461 ?

Comment 23

[Boris Zbarsky [:bzbarsky]]

None, as far as I can tell.

Comment 24

[Markus Hübner]

thx for checking!
then the perf bottleneck in bug 151461 is not document.write() - or?

Comment 25

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I would like to have a look at this. Stealing.

Comment 26

[Alfred Kayser]

Some current testing, on PentiumM 1.5Ghz, WinXp, 1GB RAM (Thinkpad T40):
             IE6.0        Deerpark 20050720
Test1        0.40         0.45
Test2        0.46         0.75

So differences are becoming much less!

Comment 27

[Alfred Kayser]

Actually, some re-testing (on 22/8/2005) results:
             IE6.0        Deerpark 20050817     Deerpark20050821
Test1        0.38         0.37                  0.38
Test2        0.73         0.78                  0.72 (!)

So, we are very very close, or can we just say there now the same?
Notes:
* Deerpark20050821 has bug 243486 fixed (Lots of small document.write spend a
lot of time in cache code).
* IE6.0 was really: 6.0.2800.1106.xpsp2.05031-1526

Marking this one as FIXED (by bug 243486)