Re: [PATCH v2 03/13] libdpkg: factor out common gzip, bzip2 code

[Jonathan Nieder <jrnieder@gmail.com>]

Hi,

Guillem Jover wrote:
> On Sat, 2009-10-24 at 17:41:02 -0500, Jonathan Nieder wrote:

>> This patch refactors the callers for zlib and libbz2 library
>> functions to share some code, in preparation for some small
>> changes that apply to both.
>
> This one I didn't apply, as the code just seems similar, but it would
> make it more difficult to handle once the need to diverge.

Looking at the resulting compress.c, I agree...

> Also given
> the signal handling brokeness and possible refactoring and combination
> with the buffer API I might want to rewrite those w/o using the
> compression library IO interfaces.

... especially considering that something generic like this ought to
happen anyway.

As long as these are being hooked into the buffer API, using the
lower-level deflate()/inflate()-style interfaces for now does seem
like the right thing to do.  I am imagining a new
BUFFER_WRITE_COMPRESSED where the associated data is a pointer to a
struct with a pointer to a compressor and whatever state the
compressor needs.

Do you think the lower-level deflate()/inflate() interface is the
right one to use?  If it is not hard to do, I think it would be
better to teach zlib and libbz2 to behave better.  Whichever way
we go now, it is easy to switch later.

I’ve filed the zlib bug as <http://bugs.debian.org/568149>.  Thanks
for reporting the libbz2 suggestions before, by the way.

Regards,
Jonathan