[antlr-interest] postmortem

[Andy Tripp]

Thomas Brandon wrote:
> On Fri, Mar 14, 2008 at 5:38 AM, Andy Tripp <antlr at jazillian.com> wrote:
>>  My guess is that you could use various heuristics to try to figure out what
>> a reasonable AST might be.
>>  But that approach of using heuristics and producing less-than-perfect
>> output doesn't go over well with the compiler crowd.
>>
> Oh come on Andy. No one said that. A feature that produces imperfect
> but usable output than can be tweaked to fit needs is fine. 

There are plenty of examples in this thread and others in which the
response is essentially "that won't work in general, so it shouldn't be done".

> A tool
> that produces output that is unusable 

I think I've justified how a parse tree is not "unusable",
even when returned as an AST.


> and where the effort required
> and end-product quality is worse than just doing it by hand to begin
> with is just stupid whether you're a member of the compiler
> intelligentsia or a practical layman.

Right.
So we should have this discussion, figure out what effort is required,
and whether the end-product quality is better or worse. Not just
dismiss it with "a parse tree is not an AST" or "if you want a parse
tree, just ask for it" or "an flat tree is exactly what you've asked for".

I agree with that.

Andy
> 
> Tom.
>