dep3 parser/generator for git-buildpackage

[John Wright]

On Mon, Jan 21, 2013 at 12:01:36PM +0100, Thomas Koch wrote:
> Hi,
> 
> I wrote a dep3 parser / generator to enhance the patch queue handling of git-
> buildpackage. The module (and a test module) currently lives here:
> 
> https://github.com/thkoch2001/python-
> debian/blob/thkoch/dep3_patch/lib/debian/dep3_patch.py
> 
> I intend to prepare some commits to git-buildpackage including the module 
> there first to let it mature and later include it in python-debian. In git-
> buildpackage I'd rewrite / delete some code related to git-pq to let it use 
> the new module.
> 
> My motivation for writing this was that the git-buildpackage version in 
> unstable did not preserve the dep3 header fields during an import-export 
> cycle.
> 
> The dep3 module does not intend to retain a patch header verbatim. It would be 
> a huge effort to do so with little benefit. After a parse-format cycle the 
> following will have changed:

I haven't read the dep3_patch module, but it's worth noting that the
main motivation for deb822 was to maintain case and order of fields...

> - The formatter outputs the fields in a fixed order regardless of the original 
> order.
> - The formatter outputs the description fields always as continuation fields 
> with one space indentation. It does parse but not generate unindented 
> description lines.
> - The formatter outputs field names in lowercase regardless of the original 
> case.

I appreciate the idea of case normalization, but I think it would be
best if it normalized like the examples in DEP3, e.g.  'Description',
'Reviewed-By', etc.

> The module does not support dpatch, only quilt.
> I'm aware of the deb822 module in python-debian but it is rather hard to 
> understand, provides more than needed and might fail to parse unindented 
> description lines.

It's actually pretty easy to use, but it certainly could use better
documentation.  Basically you want

    d = deb822.Deb822(lines)
    d['Some-Field'] = 'value'
    # ...
    with open(filename, 'w') as f:
        d.dump(fd=f)

All that said, you're right about it failing to parse free-form
unindented description lines.

> I intend that gbp-pq would create commit messages that conform to the dep3 
> specification i.e. I can use the parser to parse the patch header as well as 
> the commit message.

Can you help me understand what gbp-pq does with the headers?  Does it
need to twiddle them after having parsed them?

> Is this OK with you, Guido, Markus? I'm relatively new to python and 
> appreciate any review.

I'll try to take a closer look in the next few days.

-- 
John Wright <jsw at debian.org>