Improve word-break: break-all and word-wrap: break-word

Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793, 479370
Committed: https://crrev.com/d12b58ef0005d24dff2047f566f5ff5a5f2e45fb
Cr-Commit-Position: refs/heads/master@{#385693}

[kojii, 2016-03-07 16:13:45]

Description was changed from

==========
WIP: break-all2

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

to

==========
WIP: break-all2

BUG=380667, 591793
==========

[kojii, 2016-03-22 02:24:36]

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

[kojii, 2016-03-22 02:24:37]

I suppose you didn't mean this one, but JFYI.

This one isn't ready to land yet, but if we want a temporary fix that breaks
less for joining scripts, I think I can try to finish this one.

[eae, 2016-03-22 16:45:41]

Yes, this is the one.

[kojii, 2016-03-27 12:17:32]

Description was changed from

==========
WIP: break-all2

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the folow, it became hard to preserve the existing
behavior. This patch matches the behavior to Gecko.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

[kojii, 2016-03-27 12:18:44]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the folow, it became hard to preserve the existing
behavior. This patch matches the behavior to Gecko.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, it became hard to preserve the
existing behavior. This patch matches the behavior to Gecko.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

[kojii, 2016-03-27 12:20:40]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, it became hard to preserve the
existing behavior. This patch matches the behavior to Gecko.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, it became hard to preserve the
existing behavior. This patch matches the behavior to Gecko because
that was the easiest to implement in the current flow.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

[kojii, 2016-03-27 13:47:34]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word".

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, it became hard to preserve the
existing behavior. This patch matches the behavior to Gecko because
that was the easiest to implement in the current flow.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above. Also these tests depends on the width of
  Hebrew characters and the computed value of "5ex" that they depends
  on platforms now. Its platform independency before the patch is
  coincidence.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width in the following examples):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, preserving the current behavior was
not easy. This patch matches the behavior to Gecko because that was the
easiest to implement in the current flow.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above.
  I tried to fix the test, but they have some platform dependencies
  that I could not produce expected values that pass on all platforms.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

[kojii, 2016-03-27 14:02:32]

PTAL. One failure in imported/web-platform-tests/dom does not seem to be
related, I'll re-run bots later.

Terribly sorry in advance, I've got a bad fever for 3 days. Back today, but I'm
leaving for vacation tomorrow until 4th Apr.

So could you please decide whether I should work after I'm back, land now
(assuming review looks good ;), or someone takes over to finish?

I suspect there could be some regressions in break-all given our low test
coverage of the property, in break-word given the behavior change, hopefully no
crashes but hard to be sure, so I might need to ask that someone to look after
them if they need immediate fixes.

Sorry for the bad timing...

[leviw_travelin_and_unemployed, 2016-03-28 20:31:45]

Howdy Kojii!

I'm actually having a lot of trouble trying to understand the change you called
out in your description:

"This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width in the following examples):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor"

Can you try describing it another way? It seems like you're describing a
situation where we'd break differently on the 1st line vs the 2nd? Anyways, it
seems like an important issue so I want to make sure I understand your point.
I'm excited about the patch (and your vacation :)

[kojii, 2016-04-05 07:57:37]

On 2016/03/28 at 20:31:45, leviw wrote:
> Howdy Kojii!
> 
> I'm actually having a lot of trouble trying to understand the change you
called out in your description:
> 
> "This patch includes one line breaking behavior change. When
> "word-wrap: break-word", and the line does not have other break
> opportunities, and the word ends at the right end, and a space follows
> (assume 5ch width in the following examples):
> 
> Gecko:
> Lorem
> [SP]
> Ipsum
> [SP]
> dolor
> 
> WebKit:
> Lorem[SP]
> Ipsum[SP]
> dolor
> 
> Blink:
> Lorem[SP]
> Ipsum
> [SP]
> dolor"
> 
> Can you try describing it another way? It seems like you're describing a
situation where we'd break differently on the 1st line vs the 2nd? Anyways, it
seems like an important issue so I want to make sure I understand your point.
I'm excited about the patch (and your vacation :)

I was totally wrong about the condition; it's not about whether the first line
or not, but about one space or 2+ spaces. I thought this is a minor change and
didn't pay enough attention, sorry about that.

So when the following text flows into 5ch width:
  Lorem[SP]Ipsum[SP][SP]dolor ABCDEF
Gecko: Wraps spaces regardless of the # of spaces.
WebKit/Trident: Overflows spaces regardless of the # of spaces.
Blink: Overflows one space, wraps if two or more spaces.
Test is here: http://jsbin.com/quwopu/edit?html,output

I'll look into this further, but can you advice which behavior we prefer? I
guess WebKit/Trident?

[kojii, 2016-04-05 17:19:13]

On 2016/04/05 at 07:57:37, kojii wrote:
> 
> I'll look into this further, but can you advice which behavior we prefer? I
guess WebKit/Trident?

or preserve the current behavior, of course. I think PS15 preserves the current
Blink behavior, though Mac bot is broken and cannot confirm on Mac yet.

[kojii, 2016-04-05 18:30:34]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

This patch includes one line breaking behavior change. When
"word-wrap: break-word", and the line does not have other break
opportunities, and the word ends at the right end, and a space follows
(assume 5ch width in the following examples):

Gecko:
Lorem
[SP]
Ipsum
[SP]
dolor

WebKit:
Lorem[SP]
Ipsum[SP]
dolor

Blink:
Lorem[SP]
Ipsum
[SP]
dolor

Due to the change in the code flow, preserving the current behavior was
not easy. This patch matches the behavior to Gecko because that was the
easiest to implement in the current flow.

Test rebaselines are as follows (results are in PS11):
editing/selection/click-left-of-rtl-wrapping-text.html
editing/selection/modify-up-on-rtl-wrapping-text.html
  One failure turns to pass, and one pass turns to failure due to the
  change described above.
  I tried to fix the test, but they have some platform dependencies
  that I could not produce expected values that pass on all platforms.
fast/forms/textarea/basic-textareas-quirks.html
fast/forms/textarea/basic-textareas.html
  Mostly due to the chagne described above. There are some cases where
  one more/less characters can fit.
fast/loader/text-document-wrapping.html
  The width of text runs change by 1px.
fast/text/midword-break-after-breakable-char.html
  The width of text runs change by 1px, and one break point change.

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793
==========

[kojii, 2016-04-05 19:20:51]

PTAL.

PS16 preserves the current behavior. Maybe 2+ space case is a bug and the
intention is WebKit/Trident, but that could be a separate patch. I've added a
ref test for the specific case.

Bots not finished yet, and there seems to be a flaky test, but I'm going to bed
today.

[eae, 2016-04-05 20:47:20]

LGTM, thanks kojii!

[kojii, 2016-04-06 18:45:03]

The timeout of
imported/web-platform-tests/dom/nodes/Document-contentType/contentType/contenttype_txt.html
was thought as flaky but turned out to be real. It contains 4k of "a" without
spaces in <pre> (i.e., word-wrap: break-word) and the test has display: none to
<iframe>, in which case the <pre> in the <iframe> has availableWidth=0.

Put some optimizations, but still the new logic measures until the end of the
word multiple times, taking ~100secs on debug build, so marked it as Timeout.
We'll need long word support to help this test, but probably this test is too
far from the real.

[kojii, 2016-04-07 07:04:15]

The CQ bit was checked by kojii@chromium.org

[kojii, 2016-04-07 07:04:15]

The patchset sent to the CQ was uploaded after l-g-t-m from eae@chromium.org
Link to the patchset: https://codereview.chromium.org/1766243003/#ps360001
(title: "Fix whitespace-in-pre.html broke by the fix for contenttype_txt.html")

[commit-bot: I haz the power, 2016-04-07 07:04:35]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1766243003/360001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1766243003/360001

[commit-bot: I haz the power, 2016-04-07 07:17:27]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-04-07 07:17:29]

Try jobs failed on following builders:
  chromium_presubmit on tryserver.chromium.linux (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.linux/builders/chromium_presub...)

[kojii, 2016-04-07 07:28:58]

The CQ bit was checked by kojii@chromium.org

[kojii, 2016-04-07 07:28:59]

The patchset sent to the CQ was uploaded after l-g-t-m from eae@chromium.org
Link to the patchset: https://codereview.chromium.org/1766243003/#ps380001
(title: "Remove Mac10.9 from TestExpectations for crbug.com/601166")

[commit-bot: I haz the power, 2016-04-07 07:29:14]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1766243003/380001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1766243003/380001

[kojii, 2016-04-07 07:29:57]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793, 479370
==========

[commit-bot: I haz the power, 2016-04-07 08:27:51]

Committed patchset #20 (id:380001)

[commit-bot: I haz the power, 2016-04-07 08:29:51]

Description was changed from

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793, 479370
==========

to

==========
Improve word-break: break-all and word-wrap: break-word

This patch improves:
1. Performance of "word-wrap: break-word" and "word-break: break-all".
2. "word-wrap: break-word" no longer breaks shaping scripts when it
   should not (crbug.com/380667).

This patch does not give a complete support when a shaped word is
broken by these properties. More accurate break using the safe-to-break
feature in HarfBuzz and rendering support will be in future patches.

Before this patch, BreakingContext::handleText() measures all
substrings to measure the correct width of joining scripts. For a
"word", this means it measures "w", "wo", "wor", and "word". This
method does not work when a shaped words can be shorter by adding more
characters.

With this patch, Blink tries normal line breaking. Then at the break
point, Blink rewinds the break point to the largest number of glyphs
that can fit.

BUG=380667, 591793, 479370

Committed: https://crrev.com/d12b58ef0005d24dff2047f566f5ff5a5f2e45fb
Cr-Commit-Position: refs/heads/master@{#385693}
==========

[commit-bot: I haz the power, 2016-04-07 08:29:53]

Patchset 20 (id:??) landed as
https://crrev.com/d12b58ef0005d24dff2047f566f5ff5a5f2e45fb
Cr-Commit-Position: refs/heads/master@{#385693}