MSDOS filenames

[Ian Jackson <ian@chiark.chu.cam.ac.uk>]

I'm working on a script to generate the filenames in the `ms-dos'
directories of the FTP site automatically (there used to be a list of
mappings generated by hand).

It seems to do quite well on generating unique names that are
meaningful to humans, but it will occasionally have to `increment' a
name Perl style (ie, increment the final letter or digit) on one file
out of a pair that would otherwise come out the same.  In a little
while I'll post a list of the mapping it produces.

Two things come to mind:

1. Does it matter if I completely replace the existing links under
{stable,development,non-free,contrib,ms-dos} ?  There might be some
mirrors which are mirroring these as files rather than links, and
they'd find they had to reload everything.  Most sites will just have
to update some links, which is cheap.

2. I'd like to organise the ms-dos directory into floppy disks.
Something like
  stable/binary/...
         source/...
         msdos-1200/admin.1/
         msdos-1200/admin.2/
         msdos-1200/admin.3/
         msdos-1200/base.1/
         ...
         msdos-1440/admin.1/
         msdos-1440/admin.2/
         msdos-1440/base.1/
         ...
perhaps, or
         ...
         msdos-1440/devel.imp-std1
         msdos-1440/devel.std2
         msdos-1440/devel.std3
         msdos-1440/devel.std4-opt1
         msdos-1440/devel.opt2
         msdos-1440/devel.opt3-xtr
         msdos-1440/mail.std-opt
         msdos-1440/mail.std-opt
         msdos-1440/mail.std-opt
         ...
(sorted by priorities).

Doing this will obviously mean that everything gets changed round
anyway.

I can do this, but if anyone has some code that can do the knapsack
problem in a reasonable way I'd like to see it.  Perl or sh would be
best :-).

Doing 2. will require the splitting of overly-large packages into
parts, using dpkg-split, on master.debian.org.  Should we split all
the packages even in `binary', or make second copies of all the
`too-large' packages to put in the msdos trees ?  The former has the
advantage of using less disk space and making sure that msdos-* only
contains symlinks; the latter means that binary only contains real
package .deb files and not fragmented package parts which might
disturb dpkg-ftp, dpkg-scanpackages (which makes the Packages files)
and similar programs.

There are currently 23 packages over 1Mb in the development tree,
totalling 43Mb, and 60 packages over 500K totalling 71Mb.  This is out
of a total size oof 108Mb.

Ian.