[Call for testers] "BSD ar"

Arjan van Leeuwen avleeuwen at gmail.com
Thu Jan 17 01:13:58 PST 2008


Hi Kai,

2008/1/11, Kai Wang <kaiwang27 at gmail.com>:
>
> Hi list,
>
> I think it's time for "BSD ar" to get a public review after months of
> bug hunting and cleaning up done by jkoshy@ and me.  The majority of
> the coding work was done during the summer of 2007 and was sponsored
> by Google SoC 2007 programme.
>
> You can find the tarball at:
> http://web.student.chalmers.se/~kaiw/patch/ar-20080111.tgz
>
> BSD ar is mostly compatible with GNU ar and it has advantages like:
>
> * BSD licensed.
> * hopefully cleaner and simpler. (build on libarchive and libelf)
> * inherited compression support from libarchive.
>   (the usefulness of this function is limited though, since ld(1)
>    does not recognize compressed .a files)
> * faster, due to the adoption of libarchive and libelf.
>   (see benchmark below)
>
> Some details on the speedup brought by BSD ar:
>
> Below is a speed comparison of processing 5329 .o files (38 MB in
> total) in create, list and extract mode of both ar. The benchmark was
> done by the acp.sh script (you can find it in the tarball) which was
> adapted from Jan Psota's Tar Comparison Program.
>
>         outcast% sh acp.sh ~/arobj .
>         bsdar:  BSD ar 1.0.0 - libarchive 2.4.10
>         gnuar:  GNU ar 2.15 [FreeBSD] 2004-05-23
>
>         best time of 3 repetitions
>                 src=/home/kaiw/arobj, 38M in     5329 files
>                 archive=./acp.a, extract to ./acptmp
>
>         program operation       real    user    system     speed
>         bsdar   create          2.17    0.26    0.68     15741 KB/s
>         gnuar   create          8.80    5.30    2.89      3881 KB/s
>
>         bsdar   list            0.04    0.00    0.04    854000 KB/s
>         gnuar   list            0.70    0.51    0.18     48800 KB/s
>
>         bsdar   extract         4.10    0.02    1.22      8331 KB/s
>         gnuar   extract         4.95    0.68    1.65      6901 KB/s
>
>
> Any feedback would be appreciated. After more thorough test, I'd
> like to bring it to the tree as an alternative ar.

I've been using this (and its updated versions) to build a large software
project (file created by ar is 405MB) several times a day for the last week.
I didn't encounter any problems during building or debugging. It's indeed
faster :). Thanks a lot!
Arjan


More information about the freebsd-current mailing list