freebsd stable - Nov 2004 - Either I do something wrong or there is a regexp bug in sed !!

I'm trying to use sed under FreeBSD 5.3-RELEASE in a new 'netqmail'
port I
am currently working on. I want to replace a bunch of digits (in plain 
English: a decimal number) in a text file at the beginning of a line. Here 
is how the original file looks before I do anything (this file is part of 
the netqmail-1.05 package, but it is unimportant):

--- file conf-split begins
23

This is the queue subdirectory split.
--- file conf-split ends

Okay, so I try to replace 23 (or whatever number is there!) at the beginning 
of the first line to let's say 199 in this file using sed. I would expect 
this to work:

sed -e "s/^[0-9]+/199/" conf-split > conf-split.new

But it doesn't change anything in conf-spilt.new!! My regexp ^[0-9]+
doesn't
match anything! After spending like an hour investigating this, I realized 
that the + after my bracket expression ( I'm talking about this part here: 
[0-9]+ ) does not match! If I omit the use of + and use * instead, I can 
make my regexp to match. So this works - but IMHO it's ugly:

sed -e "s/^[0-9][0-9]*/199/" conf-split > conf-split.new

It gives this output, which is what I always wanted:

--- file conf-split.new begins
199

This is the queue subdirectory split.
--- file conf-split.new ends

According to the sed man page, the regexp syntax that is used by sed is 
documented in the re_format man page. And according to the re_format man 
page: "A piece is an atom possibly followed by a single= `*', `+',
`?', or
bound.  An atom followed by `*' matches a sequence of 0 or more matches of 
the atom.  An atom followed by `+' matches a sequence of 1 or more matches 
of the atom. ..."
And the definition of an "atom" is (quoted from the same man page):
"An atom
is a regular expression enclosed in `()' (matching a match for the regular 
expression), an empty set of `()' (matching the null string)=, a bracket 
expression (see below) ..."

So either my bracket expression ( [0-9] ) in my first sed command was not 
recognized as an atom, or if it was recognized as an atom then the + that 
followed it was not interpreted properly... Can anyone please tell me why?

I believe this is a bug in sed or in the regexp library which sed uses. If 
it is a regexp library issue, then there is a chance that it affects other 
programs that use it, as well! At least it can break all programs that use 
sed regexps, especially ports...

My uname -a is:
FreeBSD www.xxxxxxxx.com 5.3-RELEASE FreeBSD 5.3-RELEASE #0: Fri Nov 12 
01:07:41 PST 2004     xxx@www.xxxxxxxx.com:/usr/obj/usr/src/sys/XXXXXXXX 
i386

Zoltan

On Sun, 2004-11-14 at 18:39, Zoltan Frombach wrote:
> match anything! After spending like an hour investigating this, I realized 
> that the + after my bracket expression ( I'm talking about this part
here:
Normal.
> According to the sed man page, the regexp syntax that is used by sed is 
> documented in the re_format man page. And according to the re_format man 
> page: "A piece is an atom possibly followed by a single= `*',
`+', `?', or
You need to read it more carefully.  There are two kinds of regular
expressions, "basic" and "extended".  sed, ed, and grep
speak BRE
syntax, whereas awk and egrep speak ERE syntax.  + is special only in
ERE syntax.

(And then there's GNU, where the difference between BRE and ERE is that
some things use a preceding backslash in BRE and don't in ERE, and vice
versa, so GNU sed does what you want if you use \+ instead of +.)

-- 
brandon s. allbery    [linux,solaris,freebsd,perl]     allbery@kf8nh.com
system administrator      [WAY too many hats]        allbery@ece.cmu.edu
electrical and computer engineering, carnegie mellon univ.         KF8NH

use -E