Currently the code caches all source that is read from .pike files (not .pmod), but I guess that's a bug then.
Anyway, it's not anywhere near worth the cost to store the source for pure compiled strings for this. Afterall, at least 99.995% of all pike processes won't use any doc extraction at all, and of those that do I'd say at least 80% is perfectly fine with that it only works for normal files.
If you want to cope with code that comes from odd places I think you should provide an API to let other programmers explicitly hook in their class loaders into your doc extraction system. (And such an API should be that the doc extractor can ask the loader for the source when needed, not the other way around.)
In specific projects one can also extend the master to do this all the time, if it seems like a good idea.
i did not find a way to control whether docs should be available or not, early enough in the process to not lose some docs.
I don't understand this. All pike sources that are used during the bootstrap comes from normal .pike and .pmod files in the lib directory. It's all there, even the master itself.
And btw, I do think it's a very neat feature to get online docs in hilfe. (It's not all gripes. One shouldn't forget the kudos.)