142631 - eliminate second-pass line reflow when flowing unconstrained

Status: RESOLVED FIXED

(Core :: Layout, defect, P3)

Description

[Chris Waterson]

I've got some #ifdef DEBUG_waterson stuff in nsBlockFrame::ReflowLine that
printf's messages whenever we do an unconstrained reflow _twice_.

Recall that when BRS_COMPUTEMAXWIDTH is set, the block frame will flow its lines
at an unconstrained width, and then reflow its lines again at the old width.
Ironically, many times the `old width' is unconstrained, too. (Presumably when
flowing nested tables.)

Anyway, I see my printf's fire all the time, e.g., on pages from amazon.com. If
we can eliminate that second reflow, we may shave off a bit of time.

Comment 1

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

How do you plan to do this?  Refactor into GetPrefSize (which I think we need to
do, but it's a big job)?

Comment 2

[Chris Waterson]

I don't propose to remove two-pass reflow _altogether_. I'm just proposing we
ought to avoid flowing the line twice at an unconstrained width; i.e., when
BRS_COMPUTEMAXWIDTH and BRS_UNCONSTRAINEDWIDTH are both already set on entry to
nsBlockFrame::ReflowLine.

Comment 3

[Chris Petersen]

Changing priority to P2.

Comment 4

[Markus Hübner]

Reassigning to [e-mail] for now until someone willing to take.

Comment 5

[Markus Hübner]

dbaron, is this covered by your arch rewrite - or is the impact of this not 
worth the effort?

Comment 6

[Boris Zbarsky [:bzbarsky]]

Is suspect this bug is now no more, with reflow branch landed.

Comment 7

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

Yeah, fixed by reflow branch landing.