BSE: bse_cast - an API suggestion of how to handle bse object casts

[swesterfeld]

During fixes for #32 #24 you introduced manual null pointer checks for as<TrackIfaceP> and so on. I don't believe that this is the best approach; so I here is a suggestion of how I think it should be done.

I would recommend replacing each invocation of as<> with bse_cast<>, even if in some cases this will be a little longer:

casting track object
old: track ? track->as<TrackIfaceP>() : NULL;
new: bse_cast<TrackIfaceP> (track);

casting this object
old: as<TrackIfaceP>();
new: bse_cast<TrackIfaceP> (this);

However, I still think my version is the better API, because

• you cannot forget the null pointer check by accident
• it is a little more intuitive to see that we're just casting here

Its a bit of work to do the details and replace all cases where this is used, but if you want to go this way, I can provide a complete patch which eliminates as<>() completely.

[tim-janik]

I've added some review comments inline (trying this out in github to see if that's a helpful feature). The most important point is the fact that as<> (or a new bse_cast<>) will be temporary measures that we ultimately want to get rid of in libbse.

[tim-janik]

woot?

[tim-janik]

1. return values should always go on their own line, so here, you'd write:

   template ObjectImplP
   bse_cast (BseObject *object)

2. There're a lot more back and forth conversions going on between C and C++ objects. E.g. we have as<> variants for BseItem* -> Bse::ItemImpl, -> Bse::ItemIface, -> Bse::ItemImplP, Bse::ItemIfaceP, and also for going from each of those to back to BseItem*. A bse_cast<> macro that wants to handle those conversions should use SFINAE measures (or static_assert - easier but not always possible) to a) only convert BseObject* or derived into C++ types, b) convert only Bse::ObjectIface, Bse::ObjectImpl or shared_ptr variants thereof into the corresponding BseObject* C type.

I honestly don't know how much work that involves to get right. You might want to give it a shot, but please keep in mind that after we've completed the SfiProxy property -> C++ accessor and signal -> Aisa::Event migrations, there's a huge BSE-internal cleanup phase pending where we'll attempt to merge the existing C structs with the Aida C++ classes. So ultimately, the as<> methods or any bse_cast<> variant thereof will be a temporary measure.

[tim-janik]

Ultimately, the Bse* C-Objects are scheduled for removal now, so I guess this isn't needed anymore.