Re: [Libcdio-devel] problems with sed test

[Rocky Bernstein]

I think you are correct, and this is now in git commit  8992c19 with one
addition: set SED if it is not previously set. In other words, this line:
   test -n "$SED" || SED=$ac_path_SED



On Fri, Oct 26, 2012 at 12:34 PM, Robert William Fuller <
address@hidden> wrote:

> Consider the following code in configure.ac:
>
> SED_EXTENDED_RE_FLAG='-r'
> result=$(echo 'define' | $ac_path_SED -r -e 's/^define/foo/' 2>/dev/null)
> if [[ $? -ne 0 ]] ; then
>    result=$(echo 'define' | $ac_path_SED -r -E 's/^define/foo/'
> 2>/dev/null)
>    if [[ $? -ne 0 ]] ; then
>      SED_EXTENDED_RE_FLAG='-E'
>    else
>      AC_MSG_WARN([Don't have SED that understand extended RE's. Some minor
> compilation issues may fail.])
>      SED_EXTENDED_RE_FLAG=''
>    fi
> fi
>
> Now consider that the second test ought to be [[ $? -eq 0 ]] because we're
> checking to see if the -E flag works.  BUT, the second invocation of sed
> will never succeed because the -r and the -E flag are mutually exclusive.
>  One exists on one platform and not the other.  So the second run of sed
> really ought to be "-E -e" not "-r -E".  (I suspect that someone
> accidentally changed the wrong option during a copy and paste of the prior
> invocation when they wrote this code originally.)
>
> Supposing we make all these changes, this still won't work on MacOS
> because it has ac_path_SED set to /usr/X11/bin/gsed which does not exist.
>  We should use the SED variable rather than ac_path_SED.  So when all is
> sed and done, it should look like this:
>
> SED_EXTENDED_RE_FLAG='-r'
> result=$(echo 'define' | $SED -r -e 's/^define/foo/' 2>/dev/null)
> if [[ $? -ne 0 ]] ; then
>     result=$(echo 'define' | $SED -E -e 's/^define/foo/' 2>/dev/null)
>     if [[ $? -eq 0 ]] ; then
>         SED_EXTENDED_RE_FLAG='-E'
>     else
>         AC_MSG_WARN([Don't have SED that understand extended RE's. Some
> minor compilation issues may fail.])
>         SED_EXTENDED_RE_FLAG=''
>     fi
> fi
>
> Thoughts?
>
> Rob
>
>