Prevent to measure the whole word when break-all

Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

This patch also revealed that the fix for crbug.com/603679 was not
correct. MidWordBreak must be recomputed when floats pushed the line
down and available width changed. This patch includes the fix.

BUG=622810, 603679
Committed: https://crrev.com/d24d249700deef0ed63bf12ebb6af3b28f8225fb
Cr-Commit-Position: refs/heads/master@{#403830}

[kojii, 2016-07-04 05:38:34]

Description was changed from

==========
WIP: break-all

BUG=
==========

to

==========
Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

BUG=622810
==========

[kojii, 2016-07-05 11:32:05]

Description was changed from

==========
Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

BUG=622810
==========

to

==========
Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

This patch also revealed that the fix for crbug.com/603679 was not
correct. MidWordBreak must be recomputed when floats pushed the line
down and available width changed. This patch includes the fix.

BUG=622810, 603679
==========

[kojii, 2016-07-05 11:32:26]

kojii@chromium.org changed reviewers:
+ eae@chromium.org

[kojii, 2016-07-05 11:32:27]

PTAL.

[kojii, 2016-07-05 11:37:59]

https://codereview.chromium.org/2077313002/diff/40001/third_party/WebKit/Sour...
File third_party/WebKit/Source/core/layout/line/BreakingContextInlineHeaders.h
(right):

https://codereview.chromium.org/2077313002/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/line/BreakingContextInlineHeaders.h:567: 
Up to this point in this function is just a move from handleText(). The
following logic is the additional one.

https://codereview.chromium.org/2077313002/diff/40001/third_party/WebKit/Sour...
third_party/WebKit/Source/core/layout/line/BreakingContextInlineHeaders.h:886: }
Doing this before rewindToMidWordBreak was the previous fix, but it was
incorrect. The new fix is to revert to do it after, but reset midWordBreak if
availableWidth was changed by floats, so that we loop again to recompute the new
break opportunity.

With this fix applied, word-break-all-wrap-with-floats.html fails without this
additional fix.

[eae, 2016-07-05 23:17:34]

Thanks for doing this.

LGTM

[eae, 2016-07-05 23:19:47]

The CQ bit was checked by eae@chromium.org

[commit-bot: I haz the power, 2016-07-05 23:19:53]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-07-05 23:24:15]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2016-07-05 23:25:59]

Description was changed from

==========
Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

This patch also revealed that the fix for crbug.com/603679 was not
correct. MidWordBreak must be recomputed when floats pushed the line
down and available width changed. This patch includes the fix.

BUG=622810, 603679
==========

to

==========
Prevent to measure the whole word when break-all

This patch applies the same heuristic fix as break-word to break-all.

Since the switch to the complex path, r385693 fixed the performance and
correctness issue of break-word and break-all in common cases. However,
it also introduced a performance regression when a word is really long,
such as minimized JS.

This can be fixed by taking over the ShapeResult to the next lines as
Gecko does, but it will require larger overhaul of the line breaker and
need more time to bake.

As an interim fix, r387974 introduced a heuristic fix for break-word.
This patch applies the same heuristic to break-all.

This patch also revealed that the fix for crbug.com/603679 was not
correct. MidWordBreak must be recomputed when floats pushed the line
down and available width changed. This patch includes the fix.

BUG=622810, 603679

Committed: https://crrev.com/d24d249700deef0ed63bf12ebb6af3b28f8225fb
Cr-Commit-Position: refs/heads/master@{#403830}
==========

[commit-bot: I haz the power, 2016-07-05 23:25:59]

Patchset 3 (id:??) landed as
https://crrev.com/d24d249700deef0ed63bf12ebb6af3b28f8225fb
Cr-Commit-Position: refs/heads/master@{#403830}