[Webkit-unassigned] [Bug 43516] REGRESSION: Huge number of memory leaks after enabling MathML

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=43516





--- Comment #18 from Alex Milowski <alex at milowski.com>  2010-08-11 09:29:51 PST ---
(In reply to comment #17)
> Generally speaking, destroying the render tree every time updateFromElement is called may be simple from an implementation point of view, but is likely to be quite inefficient. I haven’t looked further at the implementation, but this is one of the things Beth might be referring to when she says the architecture may need work.

Yet, this is only for the 'mo' element and only in certain situations. If it is not a stretched glyph, there is no glyph stack and the children do not need to be destroyed.  I don't think we've made those optimization checks.  We'd have to determine from the stretch height has changed such that the glyph is stretched differently and so the children are invalid.

Keep in mind that the 'mo' element only contains text children and is only expected to contain a single unicode character (the operator).  The children of this are either (A) a single RenderText instance or (B) a stack of glyphs implemented as a sequence of vertically stacked RenderBlock instances.

As I've said before, vertical glyph stacking is a standard technique here.  If there was a way to write vertical line boxes with the text progression direction being orthogonal (e.g. left to right versus the top-to-bottom stacking), we might be able to do things differently.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.