Re: [Devel] OpenType futures

[Antoine Leca]

Owen Taylor wrote:
> On Wed, 2004-08-11 at 07:40, Antoine Leca wrote:
> > If we do validation apart from parsing, we will duplicate the effort of
> > conversion, won't we?
>
> Duplicating the effort of conversion twice isn't what concerns me. The
> question is whether duplicating the effort of conversion *every* time
> you apply a particular feature lookup is going to be prohibitive.

Well, I should have been more verbose ;-)

First, my idea is that once some table (or part of) is validated, this fact
is recorded somewhere, and is not redone (at least until the face is open in
other place/thread/client/etc.)
Are we all in agreement here?

Next, it is obvious to me that the process that applies lookups (the third
of your list) should use some form of cache; in fact, probably two levels of
caches: one to cache the lookups themselves (to avoid the problem you are
mentionning above), and another to cache frequently used situations (like
RA+Uvowel when doing Hindi; in fact, a cache that with some input syllable
will directly give the output sequence is probably the good choice here.)
However, this is purely and exclusively the matters of the final client in
the scheme we are talking about, isn't it? (And here I hear your concerns
about serving a non-hacky solution right now, while leaving room for more
efficient solutions in the future.)
The only thing we have to worry about is to make such a scheme possible
efficiently, and I believe that if we digest the outermost structure early
it would be easier to deal at all levels later; but I have no real idea
about if we should, or not, digest the lookup headers, or the coverages (in
other words, all the non-terminal nodes). Since of course there is a
tradeoff between doing it in detail, thus having to record for a long time
(see above) many useless informations, or doing it a number of times.
Which is why I asked for some numbers...  ;-)

Any student in CS out there ?


Antoine