Configure scripts: a few ideas
Hi,
I have some ideas about the Configuration script for Debian
(semi-automatic configuration of software has been a pet project of
mine for a few years now).
We could do a lot worse than look at some of the mechanisms
provided by metaconfig (in dist-3.60-7), which generates the Perl
Configure scripts. It has the following points going for it:
a) There already is a mechanism in place that stores and retrieves
configuration information (use Config; in any Perl script gives
access), and to maintain and carry over Config info between
machines.
b) Individual quanta of configurable information are collected in
``units'', which state the units they depend on, and the variables
that they determine. (I can see a package providing a unit to ask
about stuff that it needs to know)
c) There is a way of providing defaults (and hints depending on,
amongst other things, architecture of the machine)
d) There is a mechanism for local overrides being provided easily.
e) Each question need be asked only once.
f) There are tools to derive dependency graphs, and ensure that there
are no dependency loops between the units.
g) There are methods to handle ``Special'' units if needed (like
units that are always included, units that are used first, units
that are used last)
h) There are tools to do lint style sanity checks on units
i) The scripts produced is a shell script.
j) It would be trivial to write units to determine, say, the local
organization, needed by, for example, news and mail packages.
k) There are mechanism to provide quiet switch, where little output
occurs.
l) There is a way to say take all default answers and don't ask me
anything method.
m) There are safe primitives to ask questions, test if things are
present in header files/libraries, and these maybe used to create
new units.
n) Help is available on the questions that are asked, and the
variables created, in a Glossary file.
o) The Configure file is just large enough to contain the units
_Currently_ required, not all the units there are, in the correct
order.
The cons are:
a) One has to write units, and learn the syntax, etc. (IMHO, not too
bas, and anything we come up with that provides dependency
ordering, loop detection, and lint checks, will need a syntax
too).
b) metaconfig is written in Perl, 74025 bytes long, plus whatever
units we write.
c) The units are only as good as the people who write them.
d) probably lots of other points I'm blind to, since I am a major fan
of dist.
Hey, I'll even volunteer to write a few units if people choose
metaconfig and Configure. (I'll post man metaconfig if asked)
manoj
--
"The real problem with SDI is that it doesn't kill anybody." Tom Neff
Manoj Srivastava <srivasta@datasync.com>
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-REQUEST@lists.debian.org . Trouble? e-mail to Bruce@Pixar.com
Reply to: