Re: packaging questions
On Thu, 28 Aug 1997, Sue Campbell wrote:
> As you can see, it is a bit of a mess and makes me want
> to totally ignore Debian policy and simply make packages as
> if they are original Debian packages, i.e. no .diff files.
> What do others think about this?
> Note that this software is very mature (most written in the
> 1970s) and doesn't change very often.
I'm not sure what the right thing to do in this case is, but in
creating a patched version of povray (pvmpov + isopov), I had multiple,
overlapping distributions of modified files and patches off the current
(and changing) base. I used diff to turn each tar-patch (replaced files)
into genuine diff patches, then put those patches into the source
archive, and added a Debian patch over the top, so that with some effort
any intermediate step can be recreated (if one of the intermediates is
updated). It takes a lot of time and is difficult to maintain this way.
The netlib packages aren't changing as much as the povray patches and
> What do people think of the idea of creating a package with only
> static libraries, i.e. a library package without a -dev version and
> no shared libs? Another issue in favor of this is the size
> of the libraries, e.g. the shared library for lapack is almost
> 3M in size. Only having one version of the library will save
> a lot of space.
In this case, I think static libraries will be better. Most programs
will use only a tiny fraction of the library, a few K out of a 3M library
file. There will be negligible overlap between dissimilar simultaneously
running programs. The few-percent shared-library penalty is not always
trivial. Povray suffers a 25% performance penalty when using shared
libpovray versus statically linked (on pentium, RISCs shouldn't suffer the
same penalty due to greater numbers of registers). People using these
codes are interested in performance. They should be running one CPU
intensive task at a time to optimize cache usage and minimize context
switches anyway. They'll probably also want programs that can be run
anywhere they can get cycles - without having to install huge shared
If you have the time, a shared library might be useful to some, but I
think the static ones are most useful (especially on i386 machines).
Dr. Drake Diedrich, Research Officer - Computing, (06)279-8302
John Curtin School of Medical Research, Australian National University 0200
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
Trouble? e-mail to email@example.com .