28677 - Replying to a message with a large number of recipients is to slow

Status: VERIFIED FIXED

(MailNews Core :: Composition, defect, P3)

Description

[Scott MacGregor]

On a release build from Friday February 18th.
If I reply to the hook which has at least 20 engineers on it, it can take over
10 seconds on my machine for the compose window to load.

The window itself comes up pretty fast but the envelope fields are all blank and
the whole app freezes for close to 10 seconds. After which, all the recipients
show up in the envelope pane.

During this freeze I do see the following dumped to the console:
"awtCopyNode" or something like that.

I suspect we are doing some inefficient dom tree copying or something that isn't
scaling well as the number of recipients in a message go up.

(Note for my test: there was no message body so the time is definetly in the
addresses. I clicked on the mailto url off the bonsai page for emailing the hook).

Also: my machine is PII-500 with 256MB of Ram. The 10 seconds may not sound like
much but if you scale this back to our target machine I'm sure we're talking
about a minute if not more.

Comment 1

[Scott MacGregor]

Nominating for beta1, adding perf keyword

Comment 2

[leger]

Putting on PDT+ radar for beta1. 

Comment 3

[Jean-Francois Ducarroz]

I will try to see if I can optimize the JS else I will rewrite part of in in C++

Comment 4

[Jean-Francois Ducarroz]

I have jut tried some JS opimization without be able to really boost the stuff. 
The only good point I found so far is to start laying down the recipients list 
before the window is fully loaded. It's more visual than really effective.

Comment 5

[Jean-Francois Ducarroz]

I tried a lot of stuff but every time I get serious side effect for only few 
seconds shaved. The addressing widget is very sensitive to any modification 
because of the way we create dynamically nodes. Typically, when you 
pre-consturct the whole tree children to speed up the insert node, the pre State 
mechanic of the select and input widget doesn't work anymore, 
that's means, we lose the content when scrolling the tree. I think it would be 
better to not touch it anymore for beta now that it works ok. Can I mark it B2?

Comment 6

[Phil Peterson]

Good call. No safe fix available means it waits for the next bus. PDT-

Comment 7

[Jean-Francois Ducarroz]

I have a fix for that that improve the time by a factor 10 for 100 recipients. 
With the fix, we are almost not depend on the number of recipients anymore. 
However, I cannot check it in because of a regression due to preState problem 
when using cloned nodes (bug 30916).

Comment 8

[Jean-Francois Ducarroz]

moving to M15

Comment 9

[Jean-Francois Ducarroz]

Fixed and checked in

Comment 10

[laurel]

Marking this verified using:
2000-04-11-09m15 commercial build NT 4.0
2000-04-11-09m15 commercial build linux rh6.0
2000-04-11-10m15 commercial build mac OS 9.0

It takes only a couple seconds on even my slower machine when replying to a
message with 50 addressees.  All addressees filled in appropriately in the
address block.

I know there's a bug about replying to hundreds of recipients, so someone will
be revisiting this overall issue ... ref bug #33715.