[Ecls-list] -dir doesn't really work & build musing

[Hannu Koivisto]

Juan Jose Garcia-Ripoll <Juan.Ripoll at mpq.mpg.de> writes:

> Of all headers, only GMP's ones are needed at build time. We can avoid

Right.  I guess I got confused by the fact that Boehm's headers are
not copied to the build tree but they are installed.  Perhaps they
shouldn't be installed if they are not used?

> the need for this header if we hide the details of the "struct bignum"
> datatype. That will prevent us from inlining stuff for bignums, but I do
> not thing that there are any inliners for bignums at all right now.

Hm.  Still, I guess it would be best to keep this possibility open.

> "build" the same directory layout that it will find in $libdir when
> installed.

Right, this is what I assumed too and then wondered why GC's
headers are not there.

> In any case that is something that needs to be fixed. I take note. The
> most probable solution will be to link headers at configuration times.
> Then configure will use "cp" instead of "ln" in platforms that do not
> support the later.

I sent some fixes to the build process in another mail but I didn't
do this since I'm not particularly fluent in autoconf.  You better
add this.

> There are no specific changes. That is the reason why I did not use
> vendor branches. Regarding the Boehm GC, the version used is rather new,
> and at the time I imported it, it was not used in any platform thus I
> did not use --enable-local-boehm. No other reason, and perhaps it is a
> good idea to add --enable-local-boehm so that ECL takes advantage of the
> fact that the collector is installed as a shared library under Debian.

Even if it couldn't be installed as a shared library from a Debian
package, I would rather install it separately and then have ECL use
that installation.  I kind of wonder why Boehm GC and GMP are in
ECL CVS to begin with if there are no local changes to them.

-- 
Hannu
Please don't send copies of list mail