tweaks used to make test-cycle-2 images

[Philip Hands <phil@hands.com>]

Hi,

Here's the diff between what's in CVS and what I used to make the
test-cycle-2 images:

  http://www.hands.com/~phil/debian/debian-cd/test-cycle-2.diff

The main points to note are:

 o  in CONF.sh the addition of:

     export MBLIMIT=645

 o  in build_all.sh:

+    if [ "$ARCH" = powerpc ] ; then
+       mblimit=$((${MBLIMIT:-630}-45))
+    elif [ "$ARCH" = m68k ] ; then
+       mblimit=$((${MBLIMIT:-630}-54))
+    else
+       mblimit=${MBLIMIT:-630}
+    fi
+    export SIZELIMIT=$(((${mblimit} - 11) * 1024 * 1024))
        disks=`du -sm ${MIRROR}/dists/${CODENAME}/main/disks-${ARCH}/current/. | \
                awk '{print $1}'`
-       make list COMPLETE=1 SIZELIMIT1=$(((630 - ${disks}) * 1024 *1024)) \
-               SRCSIZELIMIT=$((635 * 1024 * 1024))
+       make list COMPLETE=1 SIZELIMIT1=$(((${mblimit} - ${disks} - 27)*1024*1024))\
+               SRCSIZELIMIT=$((${mblimit} * 1024 * 1024))

     The main point being that there seems to be about 27MB of
     overhead in making a 650MB image, with another 50MB-ish on top of
     that if you are creating a HFS hybrid disk.  Does anyone have any
     ideas where these numbers come from?  It would be nice to have
     cds2src & list2cds work out the numbers on a real basis, rather
     than using these guesses.

     There's also a conditional [ "$1" = retry ] that allows ``make list''
     onwards to be rerun without redoing all the mirror checks etc.

 o  The latest README.html.in has been added.

 o  I added kernel-source-2.2.15 to forcd1, so at least one set of
    source ends up there.

 o  I've tweaked the size calculation in cds2src & list2cds to make
    them what I think is a little more accurate, but it's a drop in
    the ocean compared to the 27MB/80MB error.

Cheers, Phil.