From: Robert Klemme <bob.news@...>
Date: 2003-09-19T00:14:34+09:00
Subject: Re: Why IO#readlines does'nt accept a Regexp?


"Peter" <Peter.Vanbroekhoven@cs.kuleuven.ac.be> schrieb im Newsbeitrag
news:Pine.GSO.4.10.10309181322340.2343-100000@iris.cs.kuleuven.ac.be...
> > Just a guess: normally it's not necessary and another reason might be
> > performance, since the overhead of a regexp might be significant for
large
> > files.
>
> Another guess... For what I know of regexps, it's that the are compiled
to
> a finite state machine. This compilation is more expensive than the
actual
> use since an FSM only considers each input character once. The only
point
> in splitting at newlines is that newlines tend to occur more frequently
> than a random regexp match. BTW, what I meant by more expensive is in
> terms of complexity, i.e., execution time as a function of the size of
the
> input. If your file is large enough, the compilation of the regexp will
> become negligable since the regexp is small and the file is big.

A totally different reason might be that ruby regexp interface is not
suitable for streaming input.  I don't know about the C internals but from
Ruby you have to provide a String, i.e. a sequence with known length.
This is different from providing an interator that just hands out a char
at a time.

    robert