Re: Why context-free?

Tony Finch <dot@dotat.at> writes:
|>
|> The impression I got from reading the revised report on Algol 68 was that
|> the grammar involved a lot of fiddly programming at the wrong level of
|> abstraction - it wasted effort on recursive definitions of numerals, for
|> example. A good formalism should make things clearer, but the vW grammaer
|> is just obscure. Furthermore it has a reputation for being brittle - the
|> grammar was difficult to change as the language was refined.


As someone who used it a fair amount, in several different variants, I
agree with that. Given my complete failure to find a comprehensible
description of that type of grammar, my suspicion is that it should be
regarded as a blind alley.


[ And to Chris Clark ]


Yes, attributes are good, but I react against making them part of the
'active' syntax - i.e. I feel that they are better used to impose
constraints and improve diagnostics than allow alternative syntax.
Others may disagree.


> Trying to put types directly into a grammar is a disaster. It makes
> the grammar grow exponentially. That's the general failure of VW
> grammars--such grammars are a way of succinctly describing the
> explosion. Unfortunately, when one is actually parsing with them, one
> has to explode them.


But, if you start to use attributes for 'active' syntax, you have to
explode THEM (directly or indirectly)! TANSTAAFL.


However, I fully agree that you need some way of controlling the
exponential explosion. You can't JUST add types to a traditional
design and expect things to work.


Regards,
Nick Maclaren.

Return to the comp.compilers page.
Search the comp.compilers archives again.