Re: suggestion for distribution: mega archive

[Havoc Pennington <hp redhat com>]

Philip Brown <phil bolthole com> writes: 
> a) The quick hack method:
>    have a wrapper script, rather than a global configure, that
>    sequentially compiles and installs glib, then configures and
>    installs all the other stuff as needed.
>    Not entirely "clean", but still makes life a LOT easier for folks.
>    (If you were really fancy, you might even have the wrapper check for
>      "if [ -x $prefix/lib/libglib-samerev ] ; then
>          skip-glib-build
>       fi"
>    :-)
> 

This is in fact what most people use; there are many options for this
out there, including "garnome" (uses tarballs), "cvsgnome", "jhbuild",
"vicious-build-scripts", etc. 
 
> b) The way I think might make more sense:
> 
>    Most of the reason you "need" the stuff installed, is so that
>    you can run xyz-config to find out what the include paths are
>    for your dependancies, so that you can compile the "current"
>    library. But the way I understand it, once you have compiled
>    pango, it doesnt need to know where glib is any more, and
>    once you have compiled gtk, it doesnt need to know where pango is any
>    more, etc.
> 
>    SO: for the mega-archive, use some special --with-xyz-path=$PWD/xxx
>    to configure, so that it knows where to -Ixxx -L... without having to
>    worry about the "installed" problem.
> 
>    (but do a double path at link time. eg
>      $(CC) -L$builddir/xyz -R$builddir/xyz \
>        -L$installprefix/lib -R $installprefix/lib 
>    )
> 
>    Or some variation on that theme. Like set LD_LIBRARY_PATH to $BUILD/lib
> 

Well right, that's the kind of thing the mega-archive tries to
address. Trust me it's hard. ;-)

Havoc