211634 - [FIX]Add GetOwnerDocument on nsIContent or nsGenericElement

Status: RESOLVED FIXED

Description

[Boris Zbarsky [:bz] (if a patch has no decent message, automatic r-)]

See comments in bug 211440; the idea is to add a

already_AddRefed<nsIDocument> GetOwnerDocument()

to nsGenericElement (or even nsIContent, depending on what the call patterns
look like...).

I'll do a sweep over the tree looking for GetDocument callers and see whether
there is enough call to put it on nsIContent, but I suspect that it'll be enough
to put it on nsGenericElement for now.

Comment 1

[Johnny Stenback (:jst)]

How about we start by putting in on nsGenericElement and if there's enough
callers after that's done, let's take it to nsIContent...

Comment 2

[Boris Zbarsky [:bz] (if a patch has no decent message, automatic r-)]

Created attachment 127842 [details] [diff] [review]
Patch

One issue arose here.  The leaf HTML classes all decl nsIDOMNode, so they all
declare a local GetOwnerDocument with a different return type and signature. 
Unfortunately, it hides the GetOwnerDocument we want.  Hence the qualification
with the parent class name at those callsites...

Comment 3

[Boris Zbarsky [:bz] (if a patch has no decent message, automatic r-)]

Er... the first change in content/base/src/nsImageLoadingContent.cpp should not
be there -- please ignore it (and I will not be checking it in).

Comment 4

[John Keiser (jkeiser)]

Comment on attachment 127842 [details] [diff] [review]
Patch

Everything here looks fine with the generic "if it compiles" caveat.

There are a few places, such as GetInnerHTML, SetInnerHTML, and some of the
leaf elements, where you replace a simple nodeInfo->GetDocument() with
GetOwnerDocument().  I think this should be safe and probably runs more quickly
(given how cache is accessed) so it's OK with me.  My assumption is that
content's document can never be different than its owner document.  If this
rule can be broken, then you are probably *still* right but I'm not as sure
about the InnerHTML cases.

Comment 5

[Johnny Stenback (:jst)]

Comment on attachment 127842 [details] [diff] [review]
Patch

sr=jst

Comment 6

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

How come nsINodeInfo::GetDocument isn't inline? Or at least non-virtual?

Comment 7

[Boris Zbarsky [:bz] (if a patch has no decent message, automatic r-)]

I talked to jst about that today, actually.  There are non-layout users of
nsINodeInfo, so inlining and devirtualization would need to be done somewhat
carefully...

Also, the member lives on nsNodeInfo, not nsINodeInfo.  Which means that to
devirtualize or inline we would first have to merge the two classes (which we
should do).

Comment 8

[Boris Zbarsky [:bz] (if a patch has no decent message, automatic r-)]

Fix checked in.