292903 - the bfcache should save and restore all resolved standard objects

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect, P3)

Description

[Brendan Eich [:brendan]]

bryner's code in nsGlobalWindow::SaveWindowState uses CopyJSProperties which
uses JS_Enumerate to list the enumerable ids in the window object.  What we
really want are the enumerables, which include DOM standard objects, plus the
non-enumerable standard core language classes and utilities (Date, parseInt, etc.).

This gives the desired transparency, so users can decorate String.prototype,
e.g., and find the added methods on fast-back.

Since the DOM code uses JS_ResolveStandardClass to lazily initialize the
standard objects, we want a JS_EnumerateResolvedStandardClasses API.  I'll
change the bfcache code to use this new API, as well as add the API, all under
this bug, if that's ok.

/be

Comment 1

[Brendan Eich [:brendan]]

Easy enough to write a testcase.  You could even automate the navigation using
an iframe and theFrame.history.back(), etc.

/be

Comment 2

[Brendan Eich [:brendan]]

Attachment: proposed fix

This patch fixes XML-related class name resolution to (a) use
standard_class_atoms not standard_class_names, so they're included in
enumerations; (b) split out a few init hooks for the QName variants, per bug
295606.  It also implements JS_EnumerateResolvedStandardClasses, as promised. 
Next patch will use that API from the bfcache code.

/be

Comment 3

[Mike Shaver (:shaver emeritus)]

Comment on attachment 186362 [details] [diff] [review]
proposed fix

r=shaver.

Comment 4

[Brendan Eich [:brendan]]

Attachment: patch to use the new JS API

Comment 5

[Brendan Eich [:brendan]]

Comment on attachment 186362 [details] [diff] [review]
proposed fix

I checked this in earlier today.

/be

Comment 6

[Johnny Stenback (:jst)]

Comment on attachment 186446 [details] [diff] [review]
patch to use the new JS API

sr=jst

Comment 7

[Brian Ryner (not reading)]

Comment on attachment 186446 [details] [diff] [review]
patch to use the new JS API

r=me if you're sure |props| will always be non-null under normal conditions

Comment 8

[Brendan Eich [:brendan]]

(In reply to comment #7)
> (From update of attachment 186446 [details] [diff] [review] [edit])
> r=me if you're sure |props| will always be non-null under normal conditions

Null or false return from a JS API that takes a cx first argument and allocates
memory or does anything that might throw a JS error/exception means that the API
call failed, always.

Checked in, without some printfs I added to verify it worked, and with a fix to
jsapi.c, to avoid resolving all the not-yet-resolved standard class names!

/be