Radical generators suggestion

From: Vladimir Prus (ghost_at_[hidden])
Date: 2003-05-14 09:11:32

I was thinking about reply to last Dave's email about generators. I'm still 
going to write it, but in essense in will says that I don't understand 
something, has some possible example where Dave's suggestion fail and that 
I'm not sure his proposal will be really simpler/better.

But then I stopped to think if there could be any suggestion about generators 
that I can like. The answer is that it's possible. But the current scheme is 
the result of much experimentation and represent, IMO, an optimum over some 
not-so-narrow search space, and one a big change can bring us to another 
optimum. 

The idea for a change is simple. We're actively discussing algorithms, arguing 
if they can find 'optimal' tranformation in some contrived cases. Do we need 
to? Consider the WD example. We're trying to detect that there's no real 
ambiguity. But the person who implements support for this tool 
knows that there's no real ambiguity. Probably, it's better to drop all the 
disambiguation logic and allow users to decide?

Now, the choice between greedily consuming extra targets and passing them up 
is determined by disambiguation. The greedy consumption immediately leads to 
'multiple'. If we switch to passing-up extra targets, 'multiple' will be 
unneded.

Of course, this is just an idea... I'm not sure if all the transformation 
sequences tested now will work and for exact semantic of explicit 
disambiguation. But probably worth exploring. 

To make idea more concrete, the WD example can look

{declare all generators}
generators.equvivalent CPP : WD : whale.whale whale.dolphin ;

This would say that when we're making CPP target from WD source, we can choose 
either whale.whale or whale.dolphin generators, and there's no difference 
between them.

Thoughts?

- Volodya