On Saturday 11 March 2006 19:15, Davide Viti wrote: > I would push for having the udebs without stripping. > We can then play with those eventually turning them into tarballs > and once we've found the right combination of font files / ranges we > can file bugs asking to perform some stripping. You don't need udebs for that: you can do it straight from the regular debs. > I expanded the spellchecker as to compute the list of codepoints needed > by every language (see [1], "codepoints" column); from what I've seen > in the few days it's been running, the list of codepoints is quite > stable. Hmm. It would very much surprise me if that wasn't the case. After all we've been effectively in a string freeze for the release with only some changes for partman-crypto happening. In general d-i is now mature enough that you get only very few string changes, so the only new glyphs would come from new translations or from occasional string changes. > My idea for finding out a "safe" list of codepoints is to use a list > collected over a long period of time; it can be done using the > following: > > cp collected_list collected_list.tmp > cat daily_list collected_list.tmp | sort | unique > collected_list Because of what I've outlined above, that has limited value. The only thing you would need to watch out for really is the occasional new glyph. > and use collected_list as a reference for stripping the font file; I > expect an overhead smaller let's say than 5% if compared to the daily > list of codepoints. Let me know if I was clear enough and if you think > it does make sense. I'm still not convinced that stripping further than overlapping ranges and making sure that only ranges needed for the script that a font is supposed to support makes sense. This is where the big savings against the current font use are. That means we can just use the ranges listed in [1] for stripping: Identify/decide which font we want to cover which scripts/languages; identify the gpyph ranges needed for that; strip the fonts accordingly. The codepoints collected in the spellchecking scripts can be used to check we're not missing anything. Let's start basic but solid and refine later... [1] http://rudhar.com/lingtics/uniclnks.htm
Attachment:
pgpGefxX2q8d4.pgp
Description: PGP signature