Haskell packages: size and number of source packages

[Ansgar Burchardt <ansgar@debian.org>]

Hi,

[ Please CC me on replies. ]

as I spotted some tiny haskell-* packages in NEW that still build three
binary packages recently, I looked at the size of haskell packages in
Debian:

Size of *.orig.tar.gz for source packages (haskell-*):
  Median:   11k
  Average:  40k
  Stddev:  140k
  #:       597

Size of libghc-* entries in Packages for amd64: 1701381
Size of libghc-* entries in Translation-en: 957597

Assuming the size of entries in Packages is about the same for other
architectures as well, the total size of libghc-* entries in Packages
and Translation-en is: 12*1701381 + 957597 = 21374169. That is an
average of 35k per haskell-* source package.

So the size of entries in Packages indices is larger than the median
source package size and almost as large as the average size...

I noticed that haskell-* packages built three binary packages:
libghc-*-dev, libghc-*-prof and libghc-*-doc. Given only few users
install these packages, I'm wondering if it makes sense to merge them?

It seems to make sense to merge -dev and -prof: from looking at a few
random packages, haskell-* source packages already build-depend on both
variants. Also both are architecture dependant packages.

The *-doc packages seem to be quite a bit smaller than -dev packages.
For example:
  libghc-irc-dev  140696
  libghc-irc-prof 135360
                  ------
                  276056
  libghc-irc-doc   39960, ~14% the size of -dev + -prof
Including the documentation in the arch-dependant -dev packages would
mean duplication between architectures, but reduce the size of entries
in Packages (which every user downloads). This is probably a more
arguable point ;)

Ansgar