Re: RFS: lbzip2
On Fri, Feb 13, 2009 at 11:51 AM, Érsek László <firstname.lastname@example.org> wrote:
>> How is this different to or better than pbzip2?
> As I've written in the "hot spice" paragraph of my RFS, lbzip2 can
> - with multiple threads
> - a bz2 file consisting of a single bzip2 stream (eg. created by
> standard bzip2 or 7za)
> - reading it from a pipe
pbzip2 (in 1.0.4 and later)
> - using an input-bound splitter (ie. it's not the splitter thread that
> looks for bzip2 block headers at any bit positions, because that would
> become a bottleneck real fast as the number of cores (worker threads)
Unsure about that as I haven't read the pbzip2 code, perhaps you could
comment on that and if it is not available, please contact the pbzip2
author about adding it:
Based on your list, it looks like lbzip2 has no advantage over pbzip2?
Seems pbzip2 has some more features than lbzip2, like using the load
average to determine the number of threads.
> Please see the reports available at http://phptest11.atw.hu/ for
> comparisons. If you have a SUSv2 conformant system (and you can enable
> one if you run Debian), you're welcome to test it yourself and get a
> similar report; please see the README.
The report URLs give errors when requested. Same with the download link.
>> Perhaps you could get the two upstreams to collaborate and merge the
>> two implementations of the same thing?
> From the wording of this question I reckon you didn't notice the
> following in my RFS:
> Upstream Author : Laszlo Ersek (myself)
Correct, I didn't notice.
> I'm upstream for lbzip2. Its licence is GPLv2+, so anybody can
> incorporate it, according to its licence.
> Pbzip2 is distributed under a BSD-style license.
> I won't release lbzip2 under a non-GPL license. Furthermore, I'll only
> put code into upstream lbzip2 that I've written myself. I'm open to bug
> reports, but I would probably reject feature requests.
> If this precludes lbzip2 getting into Debian (under my maintenance),
> please tell me upfront so I can stop bugging people to sponsor me.
My question was to determine if lbzip2 adds any functionality to
Debian that Debian doesn't already have. Based on what you've said
about lbzip2 it seems that pbzip2 does everything lbzip2 does.
Personally, I'm not going to sponsor this you can think of some
advantage it has over pbzip2, but that doesn't mean you won't be able
to find a sponsor, there are lots of DDs in Debian with widely varying
opinions about what Debian is and what it is not.
Personally I think pbzip2/lbzip2/bzip2smp/smpbzip2/dbzip2/etc
shouldn't exist and parallelism should simply be added to bzip2. If
you could talk to the bzip2 upstream and all the people who rewrote it
to add parallelism and convince them to all work on adding parallelism
to bzip2, that would be a really great contribution to the world of
free software. Same for gzip/pigz.