Re: Working with zip files

From: Jakub Narębski <jnareb@gmail.com>
Date: Aug 18, 2016, 12:16 UTC
Message-ID: <12866c04-f910-2a83-b445-6eada3d2efc9@gmail.com>
In-Reply-To: <xmqq8tvwcvrc.fsf@gitster.mtv.corp.google.com>

W dniu 16.08.2016 o 22:19, Junio C Hamano pisze:

Show 8 quoted lines

> Jakub Narębski <jnareb@gmail.com> writes:
> 
>> There is also `textconv` filter that can be used instead; it might
>> be 'unzip -c' (extract files to stdout, with filenames), or 'unzip -p'
>> (same, without filenames).
> 
> That assumes that the in-repository data is zipped binary blob; the
> result won't delta well, will it?

Full solution would involve `clean` filter to rezip with no compression (which should delta well) and optional `smudge` filter to recompress; if round-trip bit-for-bit equality is needed, the original zip parameters must be saved somewhere, e.g. as ZIP archive comments. This was mentioned in the earlier part of my email (which might have been not clear):

JN>> You can find rezip clean/smudge filter (originally intended for
JN>> OpenDocument Format (ODF), that is OpenOffice.org etc.) that stores
JN>> zip or zip-archive (like ODT, jar, etc.) uncompressed.  I think
JN>> you can find it on GitWiki, but I might be mistaken.
 
Using 'unzip -c' as separate / additional `textconv` filter for diff
generation allows to separate the problem of deltifiable storage format
from textual representation for diff-ing.

Though best results could be had with `diff` and `merge` drivers...

-- 
Jakub Narębski