Always use original ascent/descent for FontMetrics::floatAscent|floatDescent

Always use original ascent/descent for FontMetrics::floatAscent|floatDescent

[Xianzhu, 2017-04-07 18:34:28]

The CQ bit was checked by wangxianzhu@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-04-07 18:35:21]

Dry run: 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, 2017-04-07 20:01:12]

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

[commit-bot: I haz the power, 2017-04-07 20:01:13]

Dry run: Try jobs failed on following builders:
  linux_chromium_asan_rel_ng on master.tryserver.chromium.linux (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.linux/builders/linux_chromium_...)

[Xianzhu, 2017-04-07 20:13:12]

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

[Xianzhu, 2017-04-07 20:13:12]

I just noticed that some code in this follow-up CL is similar to
https://codereview.chromium.org/2392033002/#ps60001 to which you commented:

"We tried this before and it broke text line alignment for a bunch of tests
and a number of websites. It is possible that it'll work better now but look
very carefully at layout tests diffs and be prepared to revert it."

Try jobs show that this CL changes rendered output and fails a lot of layout
tests. I think we won't want to land this CL for M58, so
https://codereview.chromium.org/2803483002/ seems the best candidate for fixing
crbug.com/706298 for M58.

[eae, 2017-04-08 06:52:01]

Thanks!

On 2017/04/07 20:13:12, Xianzhu OOO 4.8 - 4.16 wrote:
> I just noticed that some code in this follow-up CL is similar to
> https://codereview.chromium.org/2392033002/#ps60001 to which you commented:
> 
> "We tried this before and it broke text line alignment for a bunch of tests
> and a number of websites. It is possible that it'll work better now but look
> very carefully at layout tests diffs and be prepared to revert it."

That change and the comment is about changing the ascent and decent values we
*use* to have subpixel precision. What you are proposing is about changing the
values we *store* and doing the rounding later, that should not have any test
results changes at all.

[Xianzhu, 2017-04-11 00:40:59]

On 2017/04/08 06:52:01, eae wrote:
> Thanks!
> 
> On 2017/04/07 20:13:12, Xianzhu OOO 4.8 - 4.16 wrote:
> > I just noticed that some code in this follow-up CL is similar to
> > https://codereview.chromium.org/2392033002/#ps60001 to which you commented:
> > 
> > "We tried this before and it broke text line alignment for a bunch of tests
> > and a number of websites. It is possible that it'll work better now but look
> > very carefully at layout tests diffs and be prepared to revert it."
> 
> That change and the comment is about changing the ascent and decent values we
> *use* to have subpixel precision. What you are proposing is about changing the
> values we *store* and doing the rounding later, that should not have any test
> results changes at all.

I don't quite understand. Can you explain more? To my understanding, this CL
does change what floatAscent/floatDescent stores and affects their usages. Also
there are a lot of layout test failures.