Le dimanche 09 novembre 2008 à 11:04 -0500, Matthias Clasen a écrit : > On Sun, 2008-11-09 at 12:09 +0100, Nicolas Mailhot wrote: > > > > ▶ package splits, to offer more flexibility to spin groups and fedora > > users > > Can I voice some doubt about the usefulness of this ? Sure, the whole discussion is open. > Going to your wiki > page, I read that dejavu has been split into ~10 subpackages. Actually, it has been split from 3 packages (2 full + 1 lgc) to 6 font packages (3 full + 3 lgc) + 1 common package and 2 compat packages which are only used for upgrades (and will be killed in F12). The font packages still weight more than your average package. And I'd be happy to get rid of the lgc packages, except that's pitchfork land, so if I had to maintain them they should follow the same rules as dejavu-full. > If this > happens to all font packages, it will blow up metadata and package lists > and make it harder for users to install a reasonable set of fonts. I don't think so, packages with a clear content are more user-friendly than packages that mix good and bad stuff. And the item users recognize is the font family name they get in font lists. I've seen all too many times users ask what package provides a particular font, because it was hidden in a big bundle. One package per font family means we can get packagekit font auto-install to work (the main problem when it was last discussed was how to handle fonts with different capabilities in a single package) instead of having @font-face install proprietary blobs on use systems. Also it means that when we add support to a new script spins only need to add *one* small package to their default package list instead of big packages that weight 10s of megs. (for example there is enough size pressure on defaults we're seriously considering to pass on korean bold because the space is already taken by other packages) As written in the wiki page we've tried to let maintainers find the "best" split and it ended up a mess, one package per font family is a clear rule which is easy to understand by everyone, and will even result in subpackage consolidation in a few cases. > There > are real costs associated with overly fine-grained package splits. Have > you really weighted to pros and cons of this idea ? The use case you > cite > > Wanting serif from dejavu, mono from liberation, and sans from > tiresias, without dragging in all the other dejavu/liberation/tiresias > fonts is a valid setup. > > Doesn't really strike me as worth supporting... Another use-case is indic fonts. If we had a big monolithic lohit package malayam users would complain. Because it is split we can have one set of defaults taken from lohit, and another from smc, without requiring full install of both of them. And which font is default at any time is a policy decision, having to rework package split each time one font gets better than another is not a good use of resources. Lastly, multi-font packages have all too often turned into a licensing mess, because fonts are usually not created together and bundling fonts often means bundling licenses. tetex is a sorry example of what happens when you start creating font collections. And once you've decided you want to split collections the only simple splitting rule everyone understands without running circles in package reviews is splitting along font family lines. (I'm sorry if I need to insist on clear rules for new packagers. I've just had too many painful font reviews lately) -- Nicolas Mailhot
Attachment:
signature.asc
Description: Ceci est une partie de message =?ISO-8859-1?Q?num=E9riquement?= =?ISO-8859-1?Q?_sign=E9e?=
-- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
