> So that such an analysis must be done depends only on the tags themselves?
Yes, there is a semantic value to the feature tags, as opposed to the lookups
that implement them (otherwise there would be little point in having two
different levels). This is why there can be "typographic" and "linguistic"
features (smcp would be of the former type, init of the latter).
> I.e. you have to know that even though init and e.g. smcp point to similar
> lookups (simple substitution), they shouldn't be treated similarly: in the
> case of smcp, the lookup suffices, while in the case of init it doesn't. It
> is surprising that nothing in the font signals such a difference...
The features an OpenType font contains are only half the story. The complete
implementation of an OpenType-compliant system needs to take into account the
specification of the layout engine (that commands to apply some features in a
specific order, to do contextual analysis, etc.). There is indeed an
implementation choice here, but it has been made at the very beginning of
OpenType. In my opinion, it provides for a more balanced workload between the
font designer and the implementer of the typesetting system.
Arthur
