bug#19414: bug in sed

[Buchs, Kevin J.]

Bob,

Thanks for the reply. Attached is a tar archive with the test cases and 
the expected output. The bottom line, I suppose, is I did not expect the 
asterisk cases would not anchor to the beginning of the line matching 
the trivial pattern of zero spaces. I realize that asterisk is for zero 
or more matches of prior regexp, but this seems to be a non-greedy 
(anti-greedy) extreme. If I try to think like a regexp evaluation 
machine, I can see the logic for the way it works. Maybe, because it was 
in a "s" command, and not a line address, it seems like it should not be 
behaving as if it is anchored at the start of the line.

--
Kevin Buchs   Research Computer Services   Phone: 507-538-5459
Mayo Clinic   200 1st. St SW   Rochester, MN 55905   http://mayoclinic.org

On 12/19/2014 09:06 PM, Bob Proulx wrote:
> Buchs, Kevin J. wrote:
> > I am using sed, version 4.2.1 on a few different systems. What I have
> > discovered is that inside a substitute command, a space alone is magically
> > anchored at the start of the line in an anti-greedy match.
> First, thank you for your bug report.  Efforts to find and fix bugs
> are appreciated.  However...
>
> Could you provide a test case *along with the output* that you are
> expecting to see?  Otherwise any of us that look are just going to see
> what we expect and go, yep, looks okay to me.
>
> > As an example, consider this input stream:
> >
> >   a                     <-- leading space
> > b c
> > d  e
> > f    g
> Sure.
>
> > Along with these one-liner invocations:
> >
> > sed -e 's/ *//'
> > sed -e 's/[ ]*//'
> Sure.
>
> > sed -e 's/[[:space::]]*//'
> Syntax error.  You have two ':' characters where you almost certainly
> wanted only one.
>
> > sed -e 's/ ?//'
> > sed -e 's/ +//'
> Those look like you want extended regular expressions in those two
> because of the use of ? and + but you are using the basic regular
> expressions by the syntax.  So those probably won't do what you want
> but will work correctly.
>
> > What seems to be happening is that the wildcard match gets anchored to the
> > beginning of the line with a zero character hit.
> >
> > This case seems to apply not only to spaces but other characters.
> Please provide the output you are expecting.  This is what I get and
> expect:
>
> >   a
> > b c
> > d  e
> > f    g
>    sed -e 's/ *//'
>    a
>    b c
>    d  e
>    f    g
>
>    sed -e 's/[ ]*//'
>    a
>    b c
>    d  e
>    f    g
>
>    sed -e 's/[[:space:]]*//'
>    a
>    b c
>    d  e
>    f    g
>
>    sed -e 's/ \?//'
>    a
>    b c
>    d  e
>    f    g
>
>    sed -e 's/ \+//'
>    a
>    bc
>    de
>    fg
>
> Those are all as expected.  Are you expecting anything different?
>
> Bob

sed-tests-Buchs.tar.gz
Description: GNU Zip compressed data