Add to skdiff: test images for bitwise equality before testing for pixel-wise equality

Add to skdiff: test images for bitwise equality before testing for pixel-wise equality

- we now check for bitwise equality of files before bothering with pixel 
  comparisons (this will help with formats for which we don't have good
  decoders, like PDF)
- unparsable files are now reported as differing, unless they are bitwise equal
  (before, they were always reported as matching!)
- "largest area mismatch" is now based on same-size image pairs only;
  previously, if any image pairs had mismatching size, it was 100%
- removed repetitive "image size mismatch, so no diff to display" messages
- changed format of leftmost table cells to be more readable

BUG=http://code.google.com/p/skia/issues/detail?id=473
Committed: https://code.google.com/p/skia/source/detail?r=4027

[epoger, 2012-05-21 19:28:39]

If you want to visually compare the new output from the old one, you can view:

old one:
http://skia.googlecode.com/svn/trunk/tools/tests/skdiff/test1/output-expected...

new one:
http://www.corp.google.com/~epoger/codereview/6208089/tools/tests/skdiff/test...

[TomH, 2012-05-21 19:37:10]

LGTM

https://codereview.appspot.com/6208089/diff/1/tools/skdiff_main.cpp
File tools/skdiff_main.cpp (right):

https://codereview.appspot.com/6208089/diff/1/tools/skdiff_main.cpp#newcode319
tools/skdiff_main.cpp:319: return SkData::NewFromMalloc(bufferStart,
bytesInFile);
Does this take care of ownership of the malloc'd buffer?

[reed1, 2012-05-21 19:59:01]

It does take ownership. I will update the dox to make that clearer.

[epoger, 2012-05-22 17:25:55]

https://codereview.appspot.com/6208089/diff/1/tools/skdiff_main.cpp
File tools/skdiff_main.cpp (right):

https://codereview.appspot.com/6208089/diff/1/tools/skdiff_main.cpp#newcode319
tools/skdiff_main.cpp:319: return SkData::NewFromMalloc(bufferStart,
bytesInFile);
On 2012/05/21 19:37:10, TomH wrote:
> Does this take care of ownership of the malloc'd buffer?

Yes... I had wondered also, so I have already confirmed that the unref() calls
around line 695 actually do call sk_free() on the buffer.

Thanks for noting this!