Fix size of rotated text with FreeType.

Fix size of rotated text with FreeType.

This generalizes and shares the solution found for a similar issue with
GDI. The issue is that the text size is applied early, and the rest of
the transformation late. This allows us to isolate and independently
control the text size from the rest of the transformation.

R=reed@google.com

Committed: https://code.google.com/p/skia/source/detail?r=14201

[bungeman-skia, 2014-03-26 20:03:24]

This fixes the issue where rotating text with FreeType would cause the height of
the text to change, causing weird effects.

[reed1, 2014-03-26 20:19:57]

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.cpp
File src/utils/SkMatrix22.cpp (right):

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.cpp...
src/utils/SkMatrix22.cpp:34: G->setAll(c, -s, 0,
SkMatrix::setSinCos(-s, c)

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.h
File src/utils/SkMatrix22.h (right):

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.h#n...
src/utils/SkMatrix22.h:16: void Sk2x2Givens(SkVector h, SkMatrix* G);
SkComputeGivensMatrix ?

const SkVector& baseline

[reed1, 2014-03-26 20:21:05]

maybe move matrix22 into core for now, to make DEPS rolls easier

[bungeman-skia, 2014-03-27 21:02:31]

On 2014/03/26 20:21:05, reed1 wrote:
> maybe move matrix22 into core for now, to make DEPS rolls easier

This (at Patch Set 4) causes improvement (which means changes) in 74 layout
tests and 3 gms. Not sure what the current limit is on suppressing vs flagging,
this seems to be somewhere in between.

[reed1, 2014-03-27 21:07:00]

still would like a more readable name to the matrix helper.

adding experts on evaluating rebaseline -vs- suppress

[bungeman-skia, 2014-04-08 21:33:38]

For my own sanity, I'm setting this up as a suppression.

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.cpp
File src/utils/SkMatrix22.cpp (right):

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.cpp...
src/utils/SkMatrix22.cpp:34: G->setAll(c, -s, 0,
On 2014/03/26 20:19:57, reed1 wrote:
> SkMatrix::setSinCos(-s, c)

Done.

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.h
File src/utils/SkMatrix22.h (right):

https://codereview.chromium.org/213153006/diff/60001/src/utils/SkMatrix22.h#n...
src/utils/SkMatrix22.h:16: void Sk2x2Givens(SkVector h, SkMatrix* G);
On 2014/03/26 20:19:57, reed1 wrote:
> SkComputeGivensMatrix ?
> 
> const SkVector& baseline
> 

Done.

[reed1, 2014-04-09 12:14:05]

lgtm

[f(malita), 2014-04-09 14:09:37]

On 2014/04/08 21:33:38, bungeman1 wrote:
> For my own sanity, I'm setting this up as a suppression.

74 suppressions should be ok short term (we can handle rebaselining with the
auto-bot).

[bungeman-skia, 2014-04-09 14:20:39]

On 2014/04/09 14:09:37, Florin Malita wrote:
> On 2014/04/08 21:33:38, bungeman1 wrote:
> > For my own sanity, I'm setting this up as a suppression.
> 
> 74 suppressions should be ok short term (we can handle rebaselining with the
> auto-bot).

The issue with the autobot is that it only handles NeedsRebasline requests in
Blink's TestExpectations file. As a result, if I suppress in Skia's test
expectations, the test expectations need to be moved around or else I need to
handle the rebaseline manually. This way I can just check it in (I already have
the suppression in Blink), wait for all the rolls, then pull the suppression and
add the NeedsRebaseline in one go, and then just have the autobot take care of
it. At least that's what I'm planning on trying out.

[Stephen Chennney, 2014-04-09 14:29:19]

On 2014/04/09 14:20:39, bungeman1 wrote:
> On 2014/04/09 14:09:37, Florin Malita wrote:
> > On 2014/04/08 21:33:38, bungeman1 wrote:
> > > For my own sanity, I'm setting this up as a suppression.
> > 
> > 74 suppressions should be ok short term (we can handle rebaselining with the
> > auto-bot).
> 
> The issue with the autobot is that it only handles NeedsRebasline requests in
> Blink's TestExpectations file. As a result, if I suppress in Skia's test
> expectations, the test expectations need to be moved around or else I need to
> handle the rebaseline manually. This way I can just check it in (I already
have
> the suppression in Blink), wait for all the rolls, then pull the suppression
and
> add the NeedsRebaseline in one go, and then just have the autobot take care of
> it. At least that's what I'm planning on trying out.

That's the right thing to do. I see no reason why it would fail unless there are
existing expectations. But in that case you comment out the existing one and
re-add it when the rebaseline has happened (or maybe we are now robust to that
problem).

[bungeman-skia, 2014-04-15 15:52:13]

Committed patchset #6 manually as r14201 (presubmit successful).