GM self-test: add --writePath and --mismatchPath , but in a way that works cross-platform

GM self-test: add --writePath and --mismatchPath , but in a way that works cross-platform

R=scroggo@google.com

Committed: https://code.google.com/p/skia/source/detail?r=9077

[epoger, 2013-05-09 17:37:29]

Something to read while you await my review of your code. :-)

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
File
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/selftest1.png
(right):

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/selftest1.png:1:
[contents of
gm/tests/outputs/compared-against-different-pixels-images/output-actual/mismatchPath/565/selftest1.png]
Ideally, each PNG file would instead be rewritten like this:

[image with bitmap-64bitMD5 12927999507540085554]

Because that would validate the actual pixel values.  But I don't have a binary
yet that generates the bitmap-64bitMD5 of a PNG file, so this is better than
nothing.  (Currently, the GM self-test doesn't test --writePath or
--mismatchPath behavior at all!)

https://codereview.chromium.org/15085007/diff/1/gm/tests/run.sh
File gm/tests/run.sh (right):

https://codereview.chromium.org/15085007/diff/1/gm/tests/run.sh#newcode66
gm/tests/run.sh:66: COMMAND="$GM_BINARY $GM_ARGS --writeJsonSummaryPath
$JSON_SUMMARY_FILE --writePath $ACTUAL_OUTPUT_DIR/writePath --mismatchPath
$ACTUAL_OUTPUT_DIR/mismatchPath"
The actual scripting change (as opposed to all the rebaselining) is in this file
only.

[scroggo, 2013-05-09 17:58:34]

lgtm

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
File
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/bogusfile
(right):

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/bogusfile:1:
Created additional file to make sure directory isn't empty, because self-test
cannot handle empty directories.
Hahaha. We may need to get used to this, because git doesn't like empty
directories either (although there are some solutions:
http://stackoverflow.com/questions/115983/how-do-i-add-an-empty-directory-to-...).

[epoger, 2013-05-09 18:08:39]

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
File
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/bogusfile
(right):

https://codereview.chromium.org/15085007/diff/1/gm/tests/outputs/compared-aga...
gm/tests/outputs/compared-against-different-pixels-images/output-expected/mismatchPath/565/bogusfile:1:
Created additional file to make sure directory isn't empty, because self-test
cannot handle empty directories.
On 2013/05/09 17:58:34, scroggo wrote:
> Hahaha. We may need to get used to this, because git doesn't like empty
> directories either

Yeah.  I could probably wrap my call to diff in the self-test, and treat
nonexistent and empty directories as equal.  But this was easier.

> (although there are some solutions:
>
http://stackoverflow.com/questions/115983/how-do-i-add-an-empty-directory-to-...).

Thanks for the pointer.  That's a great stackoverflow, lots of
you-shouldnt-be-doing-this comments.

[epoger, 2013-05-09 18:09:37]

Committed patchset #1 manually as r9077 (presubmit successful).