672830 - Very slow autocomplete on 7.0a2, os x

Status: RESOLVED FIXED

(Thunderbird :: Message Compose Window, defect)

Description

[David Ascher (:davida)]

For a couple of days I've noticed exceedingly slow autocomplete in the composition window: 2-4 seconds of delay between keystroke and autocomplete popup showing up.

I'm using:

Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:7.0a2) Gecko/20110720 Thunderbird/7.0a2

I thought it was ldap, but disabling the ldap addressbook and restarting didn't fix it.

Wondering whether some of the addressbook de-rdfification could have had perf impact?

Stats:

 - personal address book: 1621 entries
 - collected addresses: 1376 entries
 - OS X address book: 1098 entries

Comment 1

[Mike Conley (:mconley) (:⚙️)]

David:

Going to test a hypothesis - that this performance issue is for the OSX address book implementation.

Can you please go into your EarlyBird config editor, and change ldap_2.servers.osx.uri to equal the empty string as opposed to moz-abosxdirectory:///?

Does this seem to bypass the slowness?

-Mike

Comment 2

[Mike Conley (:mconley) (:⚙️)]

(You'll need to restart, too - if the alteration works, the OSX address book will be missing in the address book manager).

-Mike

Comment 3

[David Ascher (:davida)]

I confirm that disabling OS X address book helps perf (and re-enabling it hurts it again).  I'll note that even w/o the OS X address book, there's a delay which feels larger than the usually-recommended-by-UI-guidelines of 200-400msec.  I don't know if that's a regression though.

Comment 4

[Mike Conley (:mconley) (:⚙️)]

David:

Ok, I'll look into this - but it might have to wait until after my August 4th EDS integration add-on demo.

Let me know if that's not alright,

-Mike

Comment 5

[David Ascher (:davida)]

tried to indicate w/ flags that this seems high-priority for 7.0final.

Comment 6

[David :Bienvenu]

I wonder if we cached the cards previously somehow...

Comment 7

[Mark Banner (:standard8)]

We did use to cache the cards, I know I checked that we cached ABs, when the de-RDF patch landed, but admittedly I didn't check cards.

Comment 8

[Mike Conley (:mconley) (:⚙️)]

davida:

> I'll note that even w/o the OS X address book, there's a delay
> which feels larger than the usually-recommended-by-UI-guidelines
> of 200-400msec.  I don't know if that's a regression though.

Could you install a copy of TB 6.0 and compare their behaviour with the OSX address book disabled?  I've got the feeling that the slowdown is in nsAbOSXDirectory.mm, but I want to know for sure.

-Mike

Comment 9

[David Ascher (:davida)]

I'll give it a shot.

Comment 10

[Mike Conley (:mconley) (:⚙️)]

I've generated about ~1000 contacts for testing, and the evidence is pretty strong that the slow-down is localized to the OSX address book implementation.

The slow-down in auto-complete from having ~1000 contacts in the OSX address book is much, much greater than the slow-down from having ~1000 contacts in the Personal Address Book.

I'll attach my randomly generated contacts if anyone else wants to try to confirm or refute.

Comment 11

[Mike Conley (:mconley) (:⚙️)]

Attachment: ~1000 randomly generated contacts for importing into OSX address book

Comment 12

[Mike Conley (:mconley) (:⚙️)]

Attachment: ~1000 randomly generated contacts in LDIF for importing into Thunderbird

Comment 13

[Mike Conley (:mconley) (:⚙️)]

Attachment: Patch v1

Looks like we were being a bit dumb about nsAbOSXDirectory query instances.  This patch significantly sped up auto-complete for me, with ~1000 contacts in my OSX address book.

Try results will be available here:  http://build.mozillamessaging.com/tinderboxpushlog/?tree=ThunderbirdTry&rev=09fe19b6790e

Comment 14

[Mike Conley (:mconley) (:⚙️)]

And by "significantly", I mean, without the patch, I was waiting for about 4 - 5 seconds for results.  With the patch, results come in within 1 second or so.

Comment 15

[Mike Conley (:mconley) (:⚙️)]

Comment on attachment 550815 [details] [diff] [review]
Patch v1

The try build failed on the OSX 64 bit machine, but I think it was an unrelated failure / fluke.  Ready for review.

Comment 16

[Mike Conley (:mconley) (:⚙️)]

David:

Any luck trying this patch out?  Does it seem to make things snappier?

-Mike

Comment 17

[Mike Conley (:mconley) (:⚙️)]

Bump

Comment 18

[Mark Banner (:standard8)]

Checked in: http://hg.mozilla.org/comm-central/rev/fd30dc96a8ca

Comment 19

[Mark Banner (:standard8)]

Checked into aurora: http://hg.mozilla.org/releases/comm-aurora/rev/cb2c7a7d3b65