svndiff.py: add support for git checkouts

svndiff.py: add support for git checkouts
(and yes, we should probably rename the tool soon)

R=senorblanco@chromium.org

Committed: https://code.google.com/p/skia/source/detail?r=10153

[epoger, 2013-07-17 04:38:25]

It's a little rough around the edges, but I wanted to put together something
that works to view pixel diffs in JSON expectations within a git checkout, in
time for show-and-tell on Wednesday.

Ready for review at patchset 1.

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode1
tools/svndiff.py:1: #!/usr/bin/python
Tested with JSON expectations as follows:

# Revert any pending changes in my json-expectations directory
EXPECTATIONS_DIR=~/src/skia-git-trunk/trunk/experimental/json-expectations
pushd $EXPECTATIONS_DIR ; git checkout -- */*.json ; popd
pushd $EXPECTATIONS_DIR ; git status ; popd

# Update my json-expectations directory
tools/rebaseline.py --expectations-root $EXPECTATIONS_DIR
pushd $EXPECTATIONS_DIR ; git status ; popd

# Run the modified svndiff script, and view the HTML diffs
TMPDIR=/tmp/jsondiffs
rm -rf $TMPDIR && \
  mkdir -p $TMPDIR && \
  tools/svndiff.py --dest-dir $TMPDIR --source-dir $EXPECTATIONS_DIR

[Stephen White, 2013-07-17 10:44:39]

LGTM, but consider switching to plumbing commands.

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode186
tools/svndiff.py:186: stdout = _RunCommand(['git', 'status', '-s'])
GitNit(tm): git status is a "porcelain" command, and isn't intended to be called
from scripts. I think the equivalent plumbing command for your purposes would be
something like "git ls-files -m".

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode199
tools/svndiff.py:199: args = ['git', 'show', os.path.join('HEAD:.',
file_within_repo)]
Same here. I think the plumbing equivalent is "git cat-file -p", but I haven't
messed with it much.

[epoger, 2013-07-18 15:00:04]

Thanks, Stephen!  I fixed 1 of 2 as you recommended.  Please take a look.

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode186
tools/svndiff.py:186: stdout = _RunCommand(['git', 'status', '-s'])
On 2013/07/17 10:44:40, Stephen White wrote:
> GitNit(tm): git status is a "porcelain" command, and isn't intended to be
called
> from scripts.

Ah, thanks for teaching this git n00b something.  I found this handy overview:
http://git-scm.com/book/en/Git-Internals-Plumbing-and-Porcelain

> I think the equivalent plumbing command for your purposes would be
> something like "git ls-files -m".

Done, and confirmed that it works.

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode199
tools/svndiff.py:199: args = ['git', 'show', os.path.join('HEAD:.',
file_within_repo)]
On 2013/07/17 10:44:40, Stephen White wrote:
> Same here. I think the plumbing equivalent is "git cat-file -p", but I haven't
> messed with it much.

I tried "git cat-file" in various permutations, but I couldn't get it to work.

Looking at
http://stackoverflow.com/questions/610208/how-to-retrieve-a-single-file-from-...
, I think I have two problems:
1. I need the full path to the file from git root, NOT the path from current
working dir
2. I need an object ID for the file in revision HEAD, which it seems I need to
get from another command

Given that "git show HEAD:./$RELATIVEPATH" works fine and keeps it simple, I'd
rather stick with that if it's OK with you.

[Stephen White, 2013-07-18 17:32:38]

LGTM

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode199
tools/svndiff.py:199: args = ['git', 'show', os.path.join('HEAD:.',
file_within_repo)]
On 2013/07/18 15:00:04, epoger wrote:
> On 2013/07/17 10:44:40, Stephen White wrote:
> > Same here. I think the plumbing equivalent is "git cat-file -p", but I
haven't
> > messed with it much.
> 
> I tried "git cat-file" in various permutations, but I couldn't get it to work.
> 
> Looking at
>
http://stackoverflow.com/questions/610208/how-to-retrieve-a-single-file-from-...
> , I think I have two problems:
> 1. I need the full path to the file from git root, NOT the path from current
> working dir
> 2. I need an object ID for the file in revision HEAD, which it seems I need to
> get from another command
> 
> Given that "git show HEAD:./$RELATIVEPATH" works fine and keeps it simple, I'd
> rather stick with that if it's OK with you.

Sure. Maybe just add a FIXME that we should try to avoid this porcelain command
somehow in the future.

https://codereview.chromium.org/19512002/diff/6001/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/6001/tools/svndiff.py#newcode185
tools/svndiff.py:185: return _RunCommand(['git', 'ls-files', '-m']).splitlines()
Much nicer, thanks.

[epoger, 2013-07-18 17:40:35]

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py
File tools/svndiff.py (right):

https://codereview.chromium.org/19512002/diff/1/tools/svndiff.py#newcode199
tools/svndiff.py:199: args = ['git', 'show', os.path.join('HEAD:.',
file_within_repo)]
On 2013/07/18 17:32:38, Stephen White wrote:
> On 2013/07/18 15:00:04, epoger wrote:
> > On 2013/07/17 10:44:40, Stephen White wrote:
> > > Same here. I think the plumbing equivalent is "git cat-file -p", but I
> haven't
> > > messed with it much.
> > 
> > I tried "git cat-file" in various permutations, but I couldn't get it to
work.
> > 
> > Looking at
> >
>
http://stackoverflow.com/questions/610208/how-to-retrieve-a-single-file-from-...
> > , I think I have two problems:
> > 1. I need the full path to the file from git root, NOT the path from current
> > working dir
> > 2. I need an object ID for the file in revision HEAD, which it seems I need
to
> > get from another command
> > 
> > Given that "git show HEAD:./$RELATIVEPATH" works fine and keeps it simple,
I'd
> > rather stick with that if it's OK with you.
> 
> Sure. Maybe just add a FIXME that we should try to avoid this porcelain
command
> somehow in the future.

Good idea... done.

[epoger, 2013-07-18 18:45:23]

Committed patchset #3 manually as r10153.