Re: How can the speed of a scanner be independent of the number of rules?

On Thursday, March 24, 2022 at 6:39:31 PM UTC+1, Roger L Costello wrote:
> Kaz Kylheku wrote:
>
> > suppose we have some existing scanner with 10 rules,
> > which correctly tokenizes an input. Then suppose we
> > add 990 rules to it. None of these rules take precedence
> > over the 10 rules, and so the the input is handled by the
> > same rules.
>
> Ouch!!!
>
> Such a letdown. So the statement "adding rules does not slow down the scanner"
> really isn't remarkable or awesome. Add 990 more irrelevant rules, and the
> scanner operates just as fast. Big deal.


That is, in my opinion, an invalid way of looking at scanning, because it
isn't computer science. A better way (I am not saying "the only right way") is
for example to compare the size of integers, the size of a memory address,
that a typical year-2022 CPU can process/access/read/write/lookup using in a
single CPU instruction. The fact is that the size of those integers and
addresses (such as: 32 bits, 64 bits) is extremely large compared to the
number of rules and complexity of any grammar that a human is able to directly
(by hand, by means of keyboard/mouse/touchscreen) enter into a computer. And,
creating a program just to generate or download an extremely large complex
grammar, containing quantities exceeding the capabilities of a single CPU
instruction and exceeding the size of memory installed in a typical year-2022
computer, is pointless.


-atom

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