42107 - need to fix string bundle init workaround

Status: VERIFIED FIXED

(Core Graveyard :: Embedding: APIs, defect, P3)

Description

[Judson Valeski]

currently, in an embedding situation the embeddor must init a string bundle 
prior to using embedding routines in order to avoid thread safety assertions. We 
need to remove this requirement.

Comment 1

[Judson Valeski]

over to ruslan. 

Ruslan, this is your string init proxy stuff from socket thread.

Comment 2

[ruslan]

Done

Comment 3

[Judson Valeski]

waterson re-intro'd this last night by pulling the late init. we need to fix 
this asap as it kills many embedding folks.

waterson's removal of the lateinit
http://bonsai.mozilla.org/cvsview2.cgi?diff_mode=context&whitespace_mode=show&fi
le=nsSocketTransportService.cpp&root=/cvsroot&subdir=mozilla/netwerk/base/src&co
mmand=DIFF_FRAMESET&rev1=1.41&rev2=1.42

Comment 4

[Chris Waterson]

Use a proxied call! The LateInit() solution was broken and re-entered the 
service manager in an invalid way.

Comment 5

[ruslan]

Proxied call is not going to work, as a socket transport thread is using an 
event queue in a different way.

Comment 6

[Chris Waterson]

So we all talked about this a bit tonight; just thought I'd write down what was
said for posterity's sake. A couple of ideas on how to fix:

1. Get rid of string bundle usage in Necko altogether; implement some kind of
message interface that places the burden of error code-to-message resolution on
the caller.

2. Make the service manager support a "late init" or "post create" callback that
will allow a service to do any work after it's been officially be stuffed into
the hashtable. (Cons: leaves service in an undefined state between creation and
"post-create" running; doesn't solve the problem in a general way, e.g., may
still need post-post-create and post-post-post-create ad infinitum.)

Comment 7

[ruslan]

I vote for both solutions :-)

Comment 8

[Warren Harris]

We talked some more, and I think (a) string bundles should be thread-safe, and 
(b) necko probably shouldn't use them.

To make string bundles be thread-safe, the chrome registry will have to be 
thread safe.

Comment 9

[ruslan]

Should we make a dependent bug nsbeta+ or remove this from nsbeta2 per Gagan's 
request.

Comment 10

[Judson Valeski]

this is serious. embeddors have to init the string bundle service on their
thread at init time to get around this.

Comment 11

[Warren Harris]

I'm working on this (by removing string bundles from necko altogether -- see 
bug 40729). Reassigning to myself.

Comment 12

[Warren Harris]

Turns out this bug is blocking Mitch's bug 40159 which is nsbeta2+.

Comment 13

[Warren Harris]

Fixes checked in to make necko not depend on string bundles directly.

Comment 14

[Paul Wyskoczka]

Updating QA Contact

Comment 15

[Michael Dunn]

Correction: Changing QA contact for the Embed API bugs to David Epstein.

Comment 16

[David Epstein]

string bundle init no longer required for embedding.