Re: chomsky and compiler development

Marco van de Voort <marcov@toad.stack.nl> writes:
> Formalism and unambiguous syntax are very useful principles for
> certain disciplines, and math and beta-science in general have evolved
> to use these formal notations way before computers were even invented.
>
> I can see a place for formal languages on the input side of programs.
> I can't really see much advantage in using them in programming though.




There are two issues here. First, what impact can the study of human
language have on formal languages, and second, how formal should
programming languages be.


On the first, consider just one example: should control flow downwards
or upwards in a file? Logically there is no difference at all, but I
doubt any European would be comfortable reading code that flows
upwards. Consider that this is a *simple* issue; surely there is more
to be found with more careful investigation?


Furthermore, even math isn't as asbtract as it seems. George Lakov
has done some relevant investagations that suggest set theory and
number theory -- surely pretty abstract, right? -- are actually based
on spatio-kinetic metaphors. A computer might prefer a very different
number theory from what a human likes.




On the second issue, it seems to depend on the program being written.
Sometimes you understand the program you are writing (e.g., your
eighth C compiler), and so formalism can only help. Sometimes,
however, you are figuring out the program as you go. In that case,
the formalism will often be a distraction, helping you with things you
don't care about. You'd rather worry about the gestalt of your
system, but the formal system is helping you fix up tricky corner
cases. This is backwards when large swaths of code are being
routinely discarded or rewritten as you play around.




Sometimes you want sloppy running code, as quickly as possible.




-Lex

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