Le lun. 2 mai 2022 à 12:24, Jonas Smedegaard a écrit :
> Quoting Jérémy Lal (2022-05-02 11:59:15)
> > Le lun. 2 mai 2022 à 11:54, Jonas Smedegaard a écrit :
> >
> > > Quoting Jérémy Lal (2022-05-02 11:45:36)
> > > > I stand by saying as it is, putting those packages:
> > > > node-chalk
> > > > node-strip-ansi
> > > > node-text-table
> > > > in Recommends just breaks the default functionality of eslint.
> > >
> > > Ignoring recommends breaks systems.
> > >
> >
> > eslint is a build-dependency of enigmail (nothing odd about that).
> > When sbuild builds a package, it doesn't install recommended packages
> > of build-dependencies ?
>
> Correct: Build environments need recommendations explicitly declared as
> build-dependencies, when needed.
>
I don't discuss the usefulness of Recommends.
I'm trying to argue why in the particular case of eslint and of those three
packages
(mentioned above) you put in Recommends, it breaks eslint for other packages
Build-Depending on eslint.
Adding those Recommends in Build-Depends is not a good practice, IMO,
for this reasons:
- separation of concerns
- Build-Depends eslint does not install Recommends resulting in a broken
eslint
- when some eslint recommends change, build-dependent packages will just
break again !
- worse (because undetectable) build-dependent packages will uselessly
build-depend on
no longer required packages.
Is there a strong reason to put those three packages in Recommends ?
The economy of installing three additional small packages is outweighed by
the
over-engineering needed to cope with it.
Jérémy
