The current votes seem to favour the solution based on our own
installation procedure. Although we did not specify who's allowed to
vote, and therefore when voting ends, I'd like to move on a bit with
the proposal. We can always change things later should the wind turn.
(In the end, my impression was that the differences between the
solution based on an "install" script and the one based on
"configure"/"make" were relatively minor anyway [from our point of
view at least, not from the user's one], in the sense that both
solutions relied on some scsh library to do the installation. The main
difference was about how this library should be wrapped. The
organisation of the package directory did not seem to be questioned).
Thus, I plan to do the following now:
- rewrite or complete the original document to incorporate the
various ideas that were proposed during discussion, namely:
* say something about libraries which contain C code (see Martin's
message on the subject),
* take into account the fact that packages will be loaded using
the (hopefully) forthcoming -lel option, and that the "main"
file therefore has to be written in the exec language,
* augment the install library with one or two things I mentioned
in a previous message,
- augment the install library to make it possible to specify two
"root" directories: one which will be the final one, one in which
the files actually have to be installed (this will make it
possible to use staging directories, and is needed if the script
has to be useable by Debian/Fink/... packages),
- write some correct package definitions for sunterlib, sunet,
pgscsh and maybe some other packages (if people would like to
suggest one, I'd be interested; If I have time, I might give a
look at Mike's port of Oleg's SXML stuff, which I find pretty
useful).
Considering the fact that I'm pretty busy these days, I fear that this
will take me about a week.
Michel.
|