On Sep 6, 2000, Richard Stallman <[EMAIL PROTECTED]> wrote:
> But then, what if someone
> starts a build with `prefix=/foo', modifies some files, then re-starts
> with `prefix=/bar'?
> Indeed, that would lead to bad results.
> I see the point you're making. If prefix were only a configure-time option,
> that problem could not happen. That would be an advantage.
Not only prefix, but all directory options. However, we can't prevent
these directories from being redefined in the `make' invocation. I'd
just recommend against this practice, and leave developers free to
support it as much as possible, without imposing an unsurmountable
burden on them. I.e., change the specs so that properly supporting
the assignment of directory variables at make time is downgraded from
mandatory to desirable, but a warning be added so that users become
aware of the technical difficulties (impossibilities?) of implementing
it correctly and safely, and don't demand it from maintainers.
> * Is the advantage big enough to justify the incompatible change
> of saying "You can't specify the prefix when you run Make"?
I've always considered it impossible to implement correctly for
certain programs, even though it is certainly possible for many
programs.
> * What about other related Make variables, such as *dir?
> Should they all be configure time options too?
Yes, and, in fact, they already are. Of course, the user still has
the option of overriding them when running `make', but the results
have never been predictable.
> If yes, that is a much bigger incompatible change.
True, but it's just the realization that the demands of the current
spec cannot be satisfied without an unreasonable effort using only the
available tools. We'd better warn users of the difficulties and
potential for trouble than giving them the hope that this will work,
because there are cases in which it is just too hard, or even
impossible, to do it correctly.
> If not, then prefix would be treated differently from the rest;
> that lack of consistency is also not a good thing.
It shouldn't be different.
> However, it *might* be a good idea to say that all configure scripts
> should accept --... args to specify all these directories and
> encourage people to use them that way. (It looks like
> Autoconf-generated scripts already support them.)
Yep.
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist *Please* write to mailing lists, not to me
