Re: Again: Bootstrap failure for objdir containing colon

[Alexandre Oliva <aoliva at redhat dot com>]

On Dec 15, 2000, Gerald Pfeifer <pfeifer@dbai.tuwien.ac.at> wrote:

> Geoffrey, what do you think about having the regression tester
> use an objdir whose name contains a colon?

One of the big problems of `make' is that there's no good way to quote
colons.  If we start actively supporting build directories containing
colons, some day someone may demand supporting source directories
containing colons too, and then there won't be anything we can do.

That said, I'm not against fixing this particular program.  The
problem is that it's relatively tricky to do.

Basically, what we need is some way to create libstdc++.INC such that
it contains full pathnames into the source and the build tree.  The
only way I could think of to do that is to run `make' recursively
passing it builddir and srcdir as full pathnames instead of the
potentially-relative names, so that the INCLUDES definition in the
Makefile gets expanded by the sub-make using the full pathnames.

Maybe we can do it with a separate, auxiliary Makefile?  The problem
still remains that we'll want to have some macros defined in both the
main Makefile and this auxiliary one.  We can't just pass INCLUDE down
to the other Makefile, because we really want INCLUDE to be expanded
in the auxiliary Makefile, with the full pathnames of srcdir and
builddir.

Automake has support for `include'ing Makefile fragments.  Maybe
that's what we want to use?

-- 
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