On Wed, 10 Oct 2001 mcmahill@mtl.mit.edu wrote:

> But how was it that glib was already installed?  Sorry for all the
> questions, but I'm still missing something.

No, it's OK. If it's not clear to you, there are certainly others that
have no idea what I'm talking about either, which is why I'm taking
the liberty of responding back to the list...

To build the ${DEPENDSFILE} while under "gnome", you first do
something like

  make -V DEPENDS | xargs -n1 echo > ${WRKDIR}/${DEPENDS.tmp}

Now you run the file repeatedly. This almost works,

  awk -v FS=: '{ print $2 }' ${DEPENDS.tmp} | \
  xargs -n1 sh -c 'cd $0; make -V DEPENDS | xargs -n1 echo' \
  >> ${DEPENDS.tmp2} && mv ${DEPENDS.tmp2} ${DEPENDS.tmp}

except it doesn't expand the variables in the DEPENDS. Now you can
trim duplicates, resolve variables, whatever, but don't sort! Finally,
you reverse the file into ${DEPENDSFILE}. [Mere details ommitted.]

Now you're all done with the indefinite recursion hassle. For whatever
top-level target you're doing (fetch, install, package), you simply
run the list, cd'ing to each directory, in turn, and executing it's
non-recursing little brother (e.g. checksum, no-recurse-check-install,
no-recurse-package).

One other immediate advantage is that "make fetch-list" becomes simply
"make -n fetch", and no bogus installing of packages when you only
want to fetch distfiles. You could do "make fetch" in one window, give
it a little head start, and then "make package" in another.

Most of this would be transparent, but one thing: the implied
dependencies must no longer be recorded in ${PKGDBDIR}, as doing so
for each package would defeat the whole purpose of the scheme. We have
other reasons to make this change (the perl-mk fiasco), so it's still
overall a good thing.

Frederick