Review Request: svn.fs.FileDiff

[Troy Curtis Jr <tr...@gmail.com>]

In parallel to the other discussion on import order of svn.fs.FileDiff, I'd
like some comments on the attached patch.  Mostly, this is to address the
failing tests on Windows [0], since there isn't a 'diff' executable
necessarily available.  The attached patch uses the internal Subversion
diff implementation (which IMHO is actually better all around anyway),
unless the  'diffoptions' constructor argument is provided.  If diffoptions
are given, then the old behavior of shelling out to 'diff' is still used.

I'm asking for comments because the actual diff output given when
'diffoptions' is None, is changed.  Currently a no argument call to 'diff'
is used, which results in a 'normal' diff output.  Using
svn_diff_file_output_unified4(), as I have in the patch, results in a
unified diff output.  This output is more consistent with other subversion
diff output, but it *is* different than what svn.fs.FileDiff.get_pipe()
currently returns.

Another option would be to default diffoptions to [ '--normal' ], so that
it takes an explicit argument of 'None' or '[]' to get the new behavior,
which we could use in the test.

Thoughts?
Troy

0: https://ci.apache.org/builders/svn-windows-ra/builds/1998

verify-before-commit knob [was: Review Request: svn.fs.FileDiff]

[Julian Foad <ju...@apache.org>]

Daniel Shahaf wrote:
> 2. We might include the change in an alpha/beta release to give it more testing.
> 
> (#2 applies to FSFS's new verify-before-commit knob, too.)

I noticed a 'verify-before-commit' option is included in fsfs.conf 
nowadays, "enabled by default in debug builds". (On unix that's 
--enable-maintainer-mode, not --enable-debug.)

- Julian

Re: Review Request: svn.fs.FileDiff

[Troy Curtis Jr <tr...@gmail.com>]

On Sun, Feb 11, 2018 at 4:33 PM Daniel Shahaf <d....@daniel.shahaf.name>
wrote:

> Troy Curtis Jr wrote on Sun, 11 Feb 2018 16:10 +0000:
> > I'm asking for comments because the actual diff output given when
> > 'diffoptions' is None, is changed.
> ...
> > Thoughts?
>
> 1. We have two consumers of FileDiff in tools/.  Check whether they are
> affected?
>
>
tools/examples/svnlook.py calls with diffoptions, so it is unaffected as
using args will call the executable (since that is obviously the desire
having given arguments).

tools/hook-scripts/mailer/mailer.py's output would be changed. There are
also other headers output on various conditions that mailer.py detects,
like adding and removing files.  The more I look at it the more I think
that perhaps the backwards compatible change of defaulting diff options to
'[]' and checking specifically for 'None' to use the internal tool would be
the way to go.  That way you can actually get to the internal subversion
diff implementation if you want, but the default behavior doesn't change.

Troy

> 2. We might include the change in an alpha/beta release to give it more
> testing.
>
> (#2 applies to FSFS's new verify-before-commit knob, too.)
>
> Cheers,
>
> Daniel
>

Re: Review Request: svn.fs.FileDiff

[Daniel Shahaf <d....@daniel.shahaf.name>]

Troy Curtis Jr wrote on Sun, 11 Feb 2018 16:10 +0000:
> I'm asking for comments because the actual diff output given when
> 'diffoptions' is None, is changed.
... 
> Thoughts?

1. We have two consumers of FileDiff in tools/.  Check whether they are affected?

2. We might include the change in an alpha/beta release to give it more testing.

(#2 applies to FSFS's new verify-before-commit knob, too.)

Cheers,

Daniel