Home Original page

Stackless & String-processing

Fernando Pereira pereira at research.att.com
Tue Jul 20 23:48:21 EDT 1999 
In article <7n27r8$mm4$1 at cronkite.cc.uga.edu>, Graham Matthews
<graham at sloth.math.uga.edu> wrote:
> Could you please point out to me where I claimed that "a program could
> optimise your combinator program". If you can't find such a quote then
> what is your point?
You didn't. The point, which is obvious to anyone who has actually done
serious pattern matching and parsing applications operating on
gigabytes of data, is that pattern matchers and parsers *must* be
optimized automatically (in the same way that programs in high-level
languages *must* be optimized automatically) because humans are just
not good enough at doing hand optimization of complex programs. Insofar
as a style of writing pattern-matchers/parsers is incompatible with
automatic optimization, it is only useful for toy examples, or, at,
best, for proofs of concept (although in my experience a proof of
concept that doesn't deal with efficiency issues is not much of a
proof). 
> 
> Fernando Pereira (pereira at research.att.com) wrote:
> : But from my readings (those mentioned in the thread
> : and others) as well as your comments, the whole point of monadic
> : combinators for parsing is to interweave the grammatical structure with
> : other computations, thus making the underlying grammatical structure
> : less accessible to optimization. As I suggested above, it would be
> : interesting to define classes of monadic programs that still allow (an
> : appropriate extension of) those optimizations. But AFAIK that has not
> : been done.
> 
> It hasn't been done, it won't likely be done, and I never said it
> could be done. So what is your point?
That it might be interesting to understand under what conditions it
could be done, building on the work I mentioned in my earlier posting
as well as on the work on attribute grammars, which is directly
relevant. Of course, if what is at stake is religion rather than
science, such curiousity may be frowned upon. Oh well... Enough said.

-- F
More information about the Python-list mailing list