55967 - [FIX]handling of "align" attribute change is slow

Status: RESOLVED FIXED

(Core :: Layout, defect, P3)

Description

[buster]

currently, we always reframe when an [img|object|table] has it's align attribute
changed via the DOM (as with javascript, or in the editor.)  we should be smart
enough to only reframe when going from [no align attribute] to [any value for
align attribute] or vice versa.  we should just reflow if the align attribute
already exists and we're just changing it's value, ie "left" to "right"

This is spun off of bug 55250

Comment 1

[buster]

operation is infrequent enough that this isn't a dogfood or rtm issue.  marking
future.  strictly a performance improvement.

Comment 2

[Hixie (not reading bugmail)]

If "align" is set to an invalid value, for example "no", then I believe 
(untested) that we just treat it as if the align attribute is missing. Thus
when doing this optimisation, we should make sure that we reflow if the value
changes to a valid value, and reframe if it changes to an invalid value or is
removed altogether. (This all assumes that I am right about our handling of
invalid values, of course.)

Comment 3

[Kevin McCluskey (gone)]

Build reassigning Buster's bugs to Marc.

Comment 4

[Christopher Hoess (gone)]

->Misc Code

Comment 5

[Boris Zbarsky [:bzbarsky]]

We haven't pointlessly reframed on "align" changes since dbaron fixed bug
211308.  Now we compute the new style context and only reframe if the style
difference says to.

On the other hand, it looks like nsStyleDisplay::CalcDifference will reframe
when float changes from left to right.  I'll attach a patch to fix that, I guess.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Is that safe?

Comment 7

[Boris Zbarsky [:bzbarsky]]

Attachment: Testcase

Comment 8

[Boris Zbarsky [:bzbarsky]]

Attachment: Possible patch

Comment 9

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 135454 [details] [diff] [review]
Possible patch

David, what do you think?

Comment 10

[Boris Zbarsky [:bzbarsky]]

Taking, I guess.

Comment 11

[Boris Zbarsky [:bzbarsky]]

I'm not sure it's safe, to be truthful...  I haven't been able to break it so
far, though... and I don't believe we differentiate between left and right
floats in the frame tree, only during reflow.

Does nsBlockFrame cache left/right stuff somehow?

Comment 12

[Boris Zbarsky [:bzbarsky]]

Checked in.  This makes the testcase I posted about 30% faster.