Future Fedora Development

[Tomasz Kłoczko]

Dnia 24-05-2006, śro o godzinie 10:00 +0200, Bernardo Innocenti
napisał(a):
> Tomasz Kłoczko wrote:
> 
> > It will be good remove multiple implementation of:
> > - term toolkit (lang, libtermcap and ncurses). Use only ncurses will fit
> >   all what is neccessary,. Patching all program for use libncursesw will
> >   allow remove distribute libncurses library (now are distributed
> >   libncurses and libncursesw) and move libncurses to kind
> >   compat-ncurses package.
> 
> I also want to see libtermcap die, but then libncurses and its
> terminfo database would have to move from /usr to / to support
> applications such as vim.
> 
> Maybe all the exotic terminals could remain in /usr/share/terminfo
> while the root partition would only have minimal support for the
> most common ones (linux, vt100, xterm, etc).

Best will be generate ncurses with --with-termlib and move only libterm
to /%{_lib}. This library can bring some most offen used teminals
definition and all other shoud be moved for example to compat-terminfo
package.

> > - multiple digests operation (openssl, gnutls, nss .. some programs are
> >   now linked with *all* avalaible implemntations 8-0 ),
> 
> Yeah, I agree.  OpenSSL seems to be the most comprehensive, but
> its licence stinks and this is why gnutls was written in the first
> place.  Now that Mozilla adopted the triple MPL/GPL/LGPL license,
> there's no reason to keep gnutls any more.
> 
> OpenSSL also tends to break its ABI very often and this is very
> annoying when upgrading to a new version of Fedora. I'd ditch
> that one too if wasn't used by so many packages.

I can't understand this situation. It is sick ..
Seems now strict license compliance is used only in libsoup and all
other cases current using gnutls for example in cups can be switched
between openssl and gnutls.
All this cases (current using gnutls) this very small subset of all
using SSL in distro. If in all other cases seems noone cares why now
some "for make this betetter" someone (temporary ?) decide for "make
this wors" ? Next ideology ?
For me preparing distribution it dominion of practise .. not ideology :>

If now is possible freely distribute Fedora and other distros with
openssl without lawyers actions why this is moved to ideology groud ?
Openssl not only have most offen used SLL suport. It also makes possible
using some hardware SSL accelerators. Now nss and gnutls even do not
have this on TODO lists ..

> - multipme XML parsers. Now some programs are linked with more than
> >   one XML parser library. For example fontconfig now uses expat but can
> >   be on build stage compiled as using libxml2 .. but most GNOME programs
> >   uses libxml2.
> 
> I used libxml2 once and I really don't like its API (very error prone,
> like glib).
> 
> libexpat looks much better, has no other dependencies and it's very
> small, but it doesn't have an equivalent of libxslt and xsltproc.

Look at some benchmarking materials. libxml2 is faster than expat.
Also make some programs two the same class libraries dependent makes
them additionaly slower.

> > This is only few points from much more larger list.
> > For example now in Fedora is avalaible libdbi but there is no other
> > programs which uses this library .. except dictd which unondiotionaly if
> > will find libdbi on autoconf level causes compile dbi dictd pliugin
> > (anyone uses this ? one persone or more ?). Why this unused piece of
> > <censored> was added to distributed packages list ?
> >
> > More .. why many packages do not have separated devel resources ? Why so
> > many packages still have static libraries and .la files ? Why so many
> > packages still build static libraries and and removes this on %install
> > sttage if on %build can be added --disable-static to autoconf options ?
> 
> I 100% agree with you here.  Fedora could use a good diet.

This thread ("Future Fedora Development") is for me formed on some
"wishes" level completly not understendable on development level.
Instead defining this in this way better will be take some real study on
how technically this must be gained. For example it will be good to show
which resources are most offen used, which sometimes and which completly
not used (shooting .. now more than 30% current files are from this
class). After this will be possible describe how to stip down amount of
distributed resources. On current Fedora I see few hundriets MBs which
can be removed without hurting any functionality. Examples:
- not installing doubled documentations (some packages provides the same
  documentation texts in more than one form),
- remove all Changelog files (if anyone want to move to datailed list of
  changes better is review this in SVN/CVS repo),
- remove all automake copy INSTALL files from %doc,
- instead distributing hudriets copies of GPL and simillar common
  licenses better will be make licenses package which will provide all
  this texts in roff man pages (distribute license text as %doc will be
  allowed only if it is not common license),
- remove most of .la (only fiew are neccessaruy if program/DSO uses
  libltdl) and all static libraries and also drop all static linking
  (now initrd formed from neccessary shared libraries and dynamically
  linked utilities can be smaller than formed from statically linked
  tools).
 
kloczek