On Mon, 2 Jul 2007 11:34:42 +0530
"Kumar Appaiah" <akumar@ee.iitm.ac.in> wrote:
> I have a new package ready at:
> http://mentors.debian.net/debian/pool/main/l/libitpp/libitpp_3.99.2-1.dsc
Minor problems:
1. libittp6 describes itself as:
C++ signal processing and communication library: Debug symbols
which is the same as libittp6-dbg:
C++ signal processing and communication library: Debug symbols
Remove the ": Debug symbols" from the description of libitpp6.
:-)
2. debian/changelog: experimental. I'm just checking - is this version
to be uploaded to experimental or is this a hangover from the previous
version? Change to unstable?
3. You need a debian/libittp-doc.doc-base file, including things like:
Document: ...
Title: ...
Author: ...
Abstract: <short abstract>
<long version>
.
This manual describes ...
Section: Apps/Math
Format: HTML
Index: /usr/share/doc/lib...-doc/path/index.html
Files: /usr/share/doc/lib..-doc/path/html/*
Important:
libittp-doc should be Section: doc and Architecture: all because it
is all HTML. (Architecture: all is v.important in this case - it
prevents the doc package being rebuilt for each architecture which
would be utterly pointless.) The control entry for libittp-doc also
specifies '${shlibs:Depends},' which is meaningless for a -doc package:
remove. Actually, ${misc:Depends} isn't useful for a -doc either. All
you need for a HTML doc package is:
Recommends: www-browser, dwww
> > ? You should always build your own packages with pbuilder at least once
> > for each upstream release - if only to ensure you have the
> > Build-Depends right. Simply watching or reading the pdebuild log will
> > show you that your package brings in more packages than a relatively
> > large GUI application. Please look beyond your own package and try to
> > anticipate problems.
>
> Henceforth I shall be more aware. And the above debs are made using pbuilder.
Hmm. The pbuilder log still shows gcc-3.4 being installed. You appear
to build depend on two different fortran packages:
refblas3-dev Fortran 77 reference implementation
lapack3-dev library of linear algebra routines 3
lapack3 Depends on libg2c0 (>= 1:3.4.4-5) which brings in gcc-3.4
again. lapack3 is also an old package (2005).
Just something to bear in mind. (See later).
> Right now, I've just removed the atlas3 dependency. Please check it
> out, and I'll also talk to upstream about it. However, I don't think
> features are lost by this change. If anything, performance will be
> affected, to the best of my understanding.
I'm not sure it will - atlas is still shown as an alternative to
rfblas3 in the dependency list of the libittp6 package.
> I accept this advice. But note that it is highly unlikely that a
> mathematical simulation type toolkit type library like IT++ will have
> any applications based on it, though the possibility cannot be
> discounted.
Nonetheless, libraries are still more difficult to get right than an
application.
This does, however, raise a different issue - is there some way I can
test the package?
As sponsor, it is important for me to know that the package I built
(which is the one actually uploaded) does work. The package appears to
support a 'make check' command. I did a test build on this amd64 box
and adding a '$(MAKE) check' line in the build-stamp: target after
$(MAKE) took less time than building the docs. I would recommend making
this a default command in the build-stamp rule, wrapped in a build
option:
ifeq (,$(findstring nocheck,$(DEB_BUILD_OPTIONS)))
$(MAKE) check
endif
(See the noopt rule in debian/rules for an example.)
--
Neil Williams
=============
http://www.data-freedom.org/
http://www.nosoftwarepatents.com/
http://www.linux.codehelp.co.uk/
Attachment:
pgpUNWfgsdImp.pgp
Description: PGP signature