Re: bsearch utility

[James Youngman]

On Thu, Jul 14, 2005 at 09:12:45AM -0700, Paul Eggert wrote:
> James Youngman <address@hidden> writes:
> 
> > If there is a consensus to add this, I'd have a preference for a usage
> > pattern which is consistent with look(1).
> 
> I like the idea of implementing "look" in coreutils, and doing it
> "right".  "look" was in Unix Version 7, and it is a handy utility.

Agree.

> The flags for look(1) (as specified on
> <http://www.freebsd.org/cgi/man.cgi>) are subtly different from those
> of "sort".  One other problem worth noting: my Debian GNU/Linux 3.1
> r0a look(1) has a flag "-b" whose meaning is not clear to me (and it's
> not in FreeBSD).

I've just read the Debian code.  The program mmap()s the entire file
as MAP_SHARED either way.  If -b is not specified, every line of the
file is examined for a potential match.  On the other hand, if -b was
specified a binary search is performed.  Hence there is a time
difference for sorted files (x3 for /usr/share/dict/words on my system
when the data is cached), but for unsorted files, "look" will still
work while "look -b" probably won't.

> I'd prefer that coreutils "look" have long options that are consistent
> with "sort", even if its short options are different for historical
> reasons.

A small number (--check, --unique, --reverse, --merge) probably
wouldn't make sence, but I see the sense of the idea.