Add option to gm: write out images into a hierarchy, rather than a flat set of files

Add option to gm: write out images into a hierarchy, rather than a flat set of files
BUG=https://code.google.com/p/skia/issues/detail?id=743
Committed: https://code.google.com/p/skia/source/detail?r=6167

[epoger, 2012-10-29 14:53:04]

Verified that this compiles on both Linux and Windows.

Verified the behavior on Linux as follows:

$ OPTION=nohierarchy && OUTDIR=out/gm/$OPTION && rm -rf $OUTDIR && mkdir -p
$OUTDIR && \
>   out/Debug/gm -w $OUTDIR --match complexclip --config 8888 --$OPTION
&>/dev/null && \
>   ls -R $OUTDIR
out/gm/nohierarchy:
complexclip2_8888.png	       complexclip2_rect_aa_8888.png 
complexclip_bw_8888.png
complexclip2_path_aa_8888.png  complexclip_aa_8888.png	     
complexclip_bw_layer_8888.png
complexclip2_path_bw_8888.png  complexclip_aa_layer_8888.png

$ OPTION=hierarchy && OUTDIR=out/gm/$OPTION && rm -rf $OUTDIR && mkdir -p
$OUTDIR &&   out/Debug/gm -w $OUTDIR --match complexclip --config 8888 --$OPTION
&>/dev/null &&   ls -R $OUTDIR
out/gm/hierarchy:
8888

out/gm/hierarchy/8888:
complexclip2_path_aa.png  complexclip2.png	    complexclip_aa_layer.png 
complexclip_bw_layer.png
complexclip2_path_bw.png  complexclip2_rect_aa.png  complexclip_aa.png	     
complexclip_bw.png

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp
File gm/gmmain.cpp (left):

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp#oldcode69
gm/gmmain.cpp:69: const static char* PATH_SEPARATOR = "/";
I moved this into SkOSFile.h, and made it a character instead of a string.

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp#oldcode72
gm/gmmain.cpp:72: static bool gNotifyMissingReadReference;
I moved this flag into the GMMain object while I was at it.

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp#newcode619
gm/gmmain.cpp:619: ErrorBitfield test_deferred_drawing(GM* gm,
I had to remove the "static" qualifier on a number of these methods in order to
access the new fUseFileHierarchy and fNotifyMissingReadReference fields of the
GMMain object.

[bsalomon, 2012-10-29 15:04:40]

LGTM.

Should we remove the flag and always use hierarchy (once all automated systems
use hierarchy)?

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp#newcode703
gm/gmmain.cpp:703: // if true, emit a messange when we can't find a reference
image to compare
messange->message

[epoger, 2012-10-29 15:10:50]

On 2012/10/29 15:04:40, bsalomon wrote:
> LGTM.

I'll leave this sitting here for a bit just in case someone in CC has something
to say.

> Should we remove the flag and always use hierarchy (once all automated systems
> use hierarchy)?

Probably.  My plan, if it sounds right to you: 
1. wait until we get the buildbots running it with --hierarchy
2. change default from --nohierarchy to --hierarchy
3. barring complaints, remove the option entirely

[reed1, 2012-10-29 15:10:51]

Lets keep the option for the old flat structure, at least for a while, until we
can all try out the hierarchy and are sure there is no real value loss from
today (at least for manual testing).

[epoger, 2012-10-29 15:11:37]

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.appspot.com/6810047/diff/1/gm/gmmain.cpp#newcode703
gm/gmmain.cpp:703: // if true, emit a messange when we can't find a reference
image to compare
On 2012/10/29 15:04:40, bsalomon wrote:
> messange->message

Done.

[epoger, 2012-10-29 15:19:05]

Added tiny patchset 3 to create the writePath directory also (not just the
subdirectories inside it).  If anyone objects, let me know...