[Cubicweb] CW app distribution with a not brand new cw version

Florent Cayré florent at secondweb.fr
Wed Apr 21 09:50:20 CEST 2010


2010/4/21 Sylvain Thénault <sylvain.thenault at logilab.fr>:
> On 20 avril 23:10, Florent Cayré wrote:
>> Well, that said, I am not sure it is a good idea to write a helper
>> script that would compute the real dependencies and force apt-get
>> using its output versions as you suggest. I would rather prefer
>> distributing all needed packages in a customer dedicated repo (as you
>> do yourself I believe), as I already do anyway (but without any cw
>> package for the moment).
>
> That's what we're trying to avoid with this proposal. What problem
> do you see with such script? Isn't it what you were expecting apt
> to acheive at a first glance?
>

This is exactly the problem : it should be apt-get's job ; not more than that.

>> Your document addresses a bunch of other (more or less related) points
>> however, and is of great value for people working with cw (especially
>> public APIs imo). I would add some requirements though :
>>
>> * move a lot of cw debian/control Recommends section content (if not
>> all) into Suggests section : "final" cubes (i.e. top-most cubes) can
>> easily add dependencies if needed ; adding --no-install-recommends to
>> apt-get install command is an ugly way not to install too much
>> packages (imo)
>
> what are you thinking about? I can't see that much cubes using Recommends.
> I can still add that recommendation to the final document though.
>

I was thinking about cubicweb itself : now that debian has changed the
default policy regarding recommanded packages (now installed by
defaut), I think cw has too much unnecessary dependencies : more than
that, these dependencies are only "packaging driven", cw code itself
does not require all this (and I know you just worked avoiding fop as
a cw dependency, which is a step back to the right direction : keep cw
dependencies minimal, unless it is really helpful for every end-user).

>> * make it easy (if not already done, in which case a pointer to the
>> corresponding doc would be highly appreciated in response to this
>> email ;-) ) to setup a development environment (using mercurial
>> forest) for a given cw version (and corresponding cubes dependencies),
>> and give another cw version a try
>
> we could use the same idea described here but using dependency info in
> the __pkginfo__ file. Nico started a tool to do such thing. Though
> currently, it's only easy to test the current development version
> and the latest stable release by doing 'hg fup default' or 'hg fup stable'
> in the forest.
>

Will this tool include logilab.common, yams, rql, indexer, and so on ?
I was rather thinking of forest snapshots or so ; I currently use my
own (too) simple bash scripts to achieve this, and it is really a pain
to maintain for a non cw developper, forest snapshots would be nicer
if they still exist.

> --
> Sylvain Thénault                               LOGILAB, Paris (France)
> Formations Python, Debian, Méth. Agiles: http://www.logilab.fr/formations
> Développement logiciel sur mesure:       http://www.logilab.fr/services
> CubicWeb, the semantic web framework:    http://www.cubicweb.org
>
>

Florent.

-- 
Ce message est la propriété de SecondWeb et peut contenir des
informations confidentielles. Si vous n'êtes pas le destinataire
désigné, nous vous remercions de bien vouloir nous en aviser
immédiatement et de nous retourner ce message ou de le détruire, sans
faire un quelconque usage de son contenu, ni le communiquer ou le
diffuser, ni en prendre copie, électronique ou non.

This message is the property of SecondWeb and may contain confidential
information. If you are not the designated recipient, please notify us
immediately and return the message to us or destroy it, without making
any use whatsoever of the contents thereof. Furthermore you should not
forward or copy the message by electronic or other means.



More information about the Cubicweb mailing list