Bug#599652: openoffice.org-common: Please document reason for libtextcat-data-utf8 recommends, and please consider changing dependency level
On Sun, Oct 10, 2010 at 06:57:45PM +0200, Rene Engelhard wrote:
> On Sat, Oct 09, 2010 at 02:11:38PM -0700, Josh Triplett wrote:
> > openoffice.org-common has a Recommends on libtextcat-data-utf8, but
> > doesn't give any indication of what it needs this package for, making it
>
> $ LANG="C" apt-cache show libtextcat-data-utf8
> Package: libtextcat-data-utf8
> Priority: optional
> Section: text
> Installed-Size: 252
> Maintainer: Daniele Favara <nomed@dsslive.org>
> Architecture: all
> Source: libtextcat
> Version: 2.2-4
> Replaces: libtextcat0 (<< 2.2-1)
> Conflicts: libtextcat-data-utf8
> Breaks: libtextcat0
> Filename: pool/main/libt/libtextcat/libtextcat-data-utf8_2.2-4_all.deb
> Size: 78124
> MD5sum: 99232646336e1b3b02c6eb9ff732a89b
> SHA1: d66a696721c5aadb6e48b68110ae6e9e23278156
> SHA256: 3827ae4aad34766cd4050748f90a298ed895efdc1122011d2202eb3fc5d9facb
> Description: Language detection library - data files
> ^^^^^^^^^^^^^^^^^^
> Libtextcat is a library with functions that implement the classification
> technic described in Cavnar & Trenkle, "N-Gram-Based Text Categorization".
> It was primarily developed for language guessing, a task on which it is known
> ^^^^^^^^^^^^^^^^^
> to perform with near-perfect accuracy.
> .
> This package provides the libtextcat data files.
>
> Really? The marked part is no indication?
>
> > difficult for a user to decide whether to install it.
>
> Maybe.
It tells me that some part of OO.o probably wants to do some kind of
language guessing, sure. But that describes the functionality from the
perspective of libtextcat-data-utf8, not from the OO.o user's
perspective. In particular, from the standpoint of "do I install this
package or not", it only tells me "some part of OO.o having to do with
languages might or might not break if I don't blindly install this
package". That hardly seems like the level of decision I want to make
when installing a pile of packages.
> > If this package supports notable functionality of openoffice.org, please
> > consider simply depending on it; one more (tiny) dependency does not
> > seem unreasonable given the existing size of OO.o, and it seems better
> > to depend on it than to have OO.o functionality missing or
> > non-functional for users that didn't install it.
>
> Needed for OOo guessing the text language. See the description.
I got that far. :)
> > Alternatively, if openoffice.org really can do without this package for
> > most purposes, please consider documenting the need for this package in
>
> It can, but that feature would fail for many languages (and we'd have
> dangling symlinks here) for the languages where we get these patterns
> from libtextcat directly.
So, if OO.o needs this functionality to the point of having dangling
symlinks without it, why not just Depends on it? Hardly seems like an
OO.o user would notice a 250k package, but they'd notice missing
functionality. (Or worse yet, they *wouldn't*.)
This just seems like taking Recommends to an extreme. I'd tend to think
that packages shouldn't use Recommends for something that only takes up
a bit of space (and takes up proportionally far less than the package
itself), as opposed to something that a user seriously might not want on
their system for other reasons (like an extra running daemon).
> > the openoffice.org-common package description, and please consider
> > whether it should use "Recommends" or "Suggests".
>
> package descriptions are not the place to document this (yes, I know
> we have many counter-examples here, but for historic reasons...)
I don't know of anywhere else besides the package description to put
such documentation so that the user will see it when installing packages
in a package manager. Unless you'd propose trying to make OO.o use
packagekit or similar to auto-install it on demand? :)
- Josh Triplett
Reply to: