Re: GOTO considered necessary (reworked)

[gelleric@kafka.informatik.uni-stuttgart.de (Wolfgang Gellerich)]

In article <EBxrpn.67C@world.std.com>, bobduff@world.std.com (Robert A Duff) writes:

(...)

|> In my experience, a hand-coded lexical analyzer is always better done
|> with loops and ifs and case statements, rather than as above.  That is,
|> forget that it's a FSM, and just write it in the natural way.  ("An
|> identifier is a letter, followed by a sequence of zero or more letters
|> and digits" translates quite nicely into loops, not gotos.)  I know you
|> agree, since I've seen the GNAT lexer.  (A tool-generated lexer is a
|> whole 'nother story, of course -- throw readability to the winds, and go
|> for efficiency.  Funny thing is, tool-generated lexers are less
|> efficient, in practise, in my experience.)

Definitely. We recently made a study comparing different methods how to 
implement an Ada 95 scanner and then measured the time they needed to lexical
analyse some large Ada sources. The slowest and the fastest scanners
differed by factor 74 (!) and the fastest one was hand-coded scanner, using
no GOTOs and basically following the strategy described above.

------

  Wolfgang