Preparations for running skimage tool on bots.

Preparations for running skimage tool on bots.

skimage_main:
Specify the full name of the expectations file to write on the command line,
rather than the folder to write 'results.json'. This is consistent with
writeJsonSummaryPath in GM, as well as the format for reading the file.
Further, it allows more flexibility to name the file.
Do not force opaque when writing bitmaps. skimage does its comparison on
the raw bitmaps. The images are written for visual comparison, and mostly
transparent bitmaps (like resources/plane.png) aren't meaningful when
forced opaque.
Provide an option to to write only mismatches to a directory. Handled
similar to the writePath.
Report a failure if there is no expectations file. This allows writing
to the mismatchPath so we can create an initial baseline.
Pass SkBitmap by const&.

skimage_self_test:
Include the name of the json file in the parameter passed to
createExpectationsPath. The name has been changed to be more specific.

.gitignore:
Ignore the new name of the file created by skimage_self_test.

run.sh:
Run skimage_self_tests.py

BUG=http://code.google.com/p/skia/issues/detail?id=1241
R=epoger@google.com

Committed: https://code.google.com/p/skia/source/detail?r=9474

[epoger, 2013-06-05 14:21:45]

looks good for the most part; please grab me when you have 10 minutes to discuss
the stuff below

https://codereview.chromium.org/15813016/diff/1/tools/skimage_main.cpp
File tools/skimage_main.cpp (right):

https://codereview.chromium.org/15813016/diff/1/tools/skimage_main.cpp#newcod...
tools/skimage_main.cpp:106: return SkImageEncoder::EncodeFile(outName, bm,
SkImageEncoder::kPNG_Type, 100);
Regarding the fact that we no longer force pixels to be opaque... let's please
discuss (in person) the problems written up at
https://code.google.com/p/skia/source/browse/trunk/gm/gmmain.cpp?r=9298#413 ,
and make sure we understand what the behavior will be.

https://codereview.chromium.org/15813016/diff/1/tools/tests/run.sh
File tools/tests/run.sh (right):

https://codereview.chromium.org/15813016/diff/1/tools/tests/run.sh#newcode163
tools/tests/run.sh:163: COMMAND="python tools/tests/skimage_self_test.py"
If it's not too much trouble, I think it would be a good idea to put this change
in a separate CL.  (In case you need to roll out some other part of the CL, it
would be too bad if this was rolled out too.)

[scroggo, 2013-06-05 15:30:38]

https://codereview.chromium.org/15813016/diff/1/tools/skimage_main.cpp
File tools/skimage_main.cpp (right):

https://codereview.chromium.org/15813016/diff/1/tools/skimage_main.cpp#newcod...
tools/skimage_main.cpp:106: return SkImageEncoder::EncodeFile(outName, bm,
SkImageEncoder::kPNG_Type, 100);
On 2013/06/05 14:21:45, epoger wrote:
> Regarding the fact that we no longer force pixels to be opaque... let's please
> discuss (in person) the problems written up at
> https://code.google.com/p/skia/source/browse/trunk/gm/gmmain.cpp?r=9298#413 ,
> and make sure we understand what the behavior will be.

As we discussed in person, since we never compare in the following situation:

Compare A to B where A was generated by C code and B was originally generated by
C code, but then written to PNG and decoded from PNG (which is lossy due to the
fact that Skia's colors are premultiplied).

Instead, a checksum is created from B and written to a file, and a checksum is
created from A and compared to that checksum.

https://codereview.chromium.org/15813016/diff/1/tools/tests/run.sh
File tools/tests/run.sh (right):

https://codereview.chromium.org/15813016/diff/1/tools/tests/run.sh#newcode163
tools/tests/run.sh:163: COMMAND="python tools/tests/skimage_self_test.py"
On 2013/06/05 14:21:45, epoger wrote:
> If it's not too much trouble, I think it would be a good idea to put this
change
> in a separate CL.  (In case you need to roll out some other part of the CL, it
> would be too bad if this was rolled out too.)

Done. Moved to https://codereview.chromium.org/15774017

[epoger, 2013-06-05 15:32:22]

lgtm

[scroggo, 2013-06-07 12:43:18]

Committed patchset #2 manually as r9474 (presubmit successful).