Overlapping functionality: IFTI, templates, is-expressions

[BCS]

Russell Lewis wrote:
> BCS wrote:
> 
>>> So: can you handle arbitrary lookahead?  Or do you, like Bison, make 
>>> a one-time choice of "reduce or shift" and then you're stuck with it?
>>
>>
>> Mine is a recursive decent LL parser that works by way of "assume this 
>> is one of those and if not back-out and try again". So it is LL(*).
> 
> 
> Cool!  Though that leads me to another question: when do the actions 
> run?  Inline, while you're attempting paths that might fail, or only 
> after you know they can't fail?

In-line. That plays hell with side effects as you can't trigger on 
dropped paths for back out. Also there's a time penalty, but given that 
efficiency is so bad to begin with it's not much of an issue. If a user 
needed to have side effects then building a lambda tree might work. In 
2.0 this would be easy with the non-broken closers.

One thing I've bean thinking of is trying to build in a memoizer: "we 
tried to parse a foo at location 294 before and got this, lets use it" 
I'm not sure how or how well it will work but I've got some ideas.