Improve sorting of skdiff output, and make it consistent across platforms

Improve sorting of skdiff output, and make it consistent across platforms
BUG=https://code.google.com/p/skia/issues/detail?id=677
Committed: https://code.google.com/p/skia/source/detail?r=4388

[epoger, 2012-06-27 10:59:45]

Ready to review at patchset 3... I updated the "expected" self-test results to
match the new sort.

While reviewing, please see my attempt at automatically wrapping each sortProc
rather than requiring each sortProc to call the presort and tiebreaker.  Can you
think of a clever way to make this work?

[TomH, 2012-06-27 13:23:16]

https://codereview.appspot.com/6351045/diff/3004/tools/skdiff_main.cpp
File tools/skdiff_main.cpp (right):

https://codereview.appspot.com/6351045/diff/3004/tools/skdiff_main.cpp#newcode20
tools/skdiff_main.cpp:20: // TODO(EPOGER): see if we can find a way to
automatically wrap sortProc
Why are we wrapping it? It looks like the intent is to always sort by kResult
first, and only then sort by whatever metric the sort proc is trying to use?

I'm afraid qsort() won't let us do the automatic wrapping that would be nice
here - but if we give up on trying, we can get rid of all the #defines and
alternate implementations, and have cleaner code.

[epoger, 2012-06-27 13:34:21]

Thanks, Tom.  Please grab me when you have a minute to discuss live.

https://codereview.appspot.com/6351045/diff/3004/tools/skdiff_main.cpp
File tools/skdiff_main.cpp (right):

https://codereview.appspot.com/6351045/diff/3004/tools/skdiff_main.cpp#newcode20
tools/skdiff_main.cpp:20: // TODO(EPOGER): see if we can find a way to
automatically wrap sortProc
On 2012/06/27 13:23:16, TomH wrote:
> Why are we wrapping it? It looks like the intent is to always sort by kResult
> first, and only then sort by whatever metric the sort proc is trying to use?

The idea is that, regardless of how we want to sort within the
file-pairs-that-differ-only-by-some-pixels, we have a top-level sort we want to
apply first and a tiebreaker to apply afterwards...
- top-level: sort by fResult
- tiebreaker: a final sort so that otherwise-equivalent file pairs have some
consistent ordering

> 
> I'm afraid qsort() won't let us do the automatic wrapping that would be nice
> here - but if we give up on trying, we can get rid of all the #defines and
> alternate implementations, and have cleaner code.

Sure, but first I wanted to see if I was missing some relatively straightforward
way of adding a level of indirection to sortProc so that it would be automatic.

Could I use C++ classes for this somehow, and pass a pointer to a method whose
object contained information about which pixel comparison sort to use if the
top-level sort returns 0?

If we can't come up with a clever/simple solution, I'll just remove everything
except the MANUALLY_WRAP_SORTPROC implementation and update the patch.

[TomH, 2012-06-27 14:15:19]

Verbal discussion: try templates, otherwise rip out everything except
MANUALLY_WRAP_SORTPROC.

[epoger, 2012-06-27 16:12:09]

Wow.

Between templates and cast operations, it took some doing, but it works!

[epoger, 2012-06-28 16:22:15]

Hey Tom- any thoughts on patchset 4?

[TomH, 2012-06-28 16:24:30]

Sorry - LGTM!