On 2017-03-28 14:50, Marco Atzeri wrote:
On 28/03/2017 18:01, Yaakov Selkowitz wrote:
On 2017-03-28 02:11, szgyg wrote:
FWIW guile 2.2.0 was released two weeks ago [0]. Mostly works, but it
still has failing tests [1].
Oh, great...
So we probably should take this into consideration *now*, even though
most guile consumers aren't ready for the changes in 2.2 yet.
Fortunately, it seems guile itself has taken their own instability into
account:
https://www.gnu.org/software/guile/manual/html_node/Parallel-Installations.html
Although looking at packages which use guile, most haven't fully adapted
to this yet (guile-config is still used by some, and it doesn't seem
anyone is using pkg-config to find guile and guild), so we still need to
provide unversioned binaries/scripts in /usr/bin for now.
It's way too early to make 2.2 "the" guile, so I think we stick with a
versioned guile1.8 for the stragglers, and 'guile' being 2.0 for now,
but eventually transitioning to a versioned 'guile2.0'. We'll have to
take another look at this once the real world has adapted to 2.2.
Considering some are still straggling on the 1.8 to 2.0 conversion,
2.2 is for far future.
Far enough that we need 2.0 right now.
Do you think we need to maintain 1.8 around ?
Based on Fedora's use of each, of the Cygwin packages currently
requiring libguile17, at least lilypond and TeXmacs seem not to be
compatible with 2.0. There are also a few others in Fedora but not in
Cygwin (coot, drgeo, gnurobots, trackballs). So I think it needs to
stay, at least for now.
Debian seems to have drop it.
Indeed, although they did so only by bundling guile-1.8 with the
lilypond source package, attempting to patch some for 2.0 (which doesn't
always work well), and dropping the rest. I much prefer Fedora's
approach of parallel-installing both.
--
Yaakov
