Re: [PATCH 3/5] configure: Don't use bash-specific string-replacement syntax

[Peter Maydell]

On Wed, 20 Jul 2022 at 17:30, Eric Blake <eblake@redhat.com> wrote:
>
> On Wed, Jul 20, 2022 at 04:26:29PM +0100, Peter Maydell wrote:
> > The variable string-replacement syntax ${var/old/new} is a bashism
> > (though it is also supported by some other shells), and for instance
> > does not work with the NetBSD /bin/sh, which complains:
> >  ../src/configure: 687: Syntax error: Bad substitution
> >
> > Replace it with a more portable sed-based approach, similar to
> > what we already do in quote_sh().
> >
>
> >    for e in $1; do
> > -    e=${e/'\'/'\\'}
> > -    e=${e/\"/'\"'}
> > -    printf '"""%s""",' "$e"
> > +    printf '"""'
> > +    # backslash escape any '\' and '"' characters
> > +    printf "%s" "$e" | sed -e 's/\([\"]\)/\\\1/g'
>
> You've fixed the bashism, but at the expense of a non-POSIX use of
> sed.  POSIX says the input to sed must be a text file (ending in a
> newline; but $e does not), and as a result it always outputs a newline
> (but you don't want a newline before the closing """).  GNU sed
> happens to do what you want for input not ending in a newline, but I
> don't remember off-hand whether BSD sed does, and I know that Solaris
> sed does not.

I just copied the approach we already take in quote_sh:

quote_sh() {
    printf "%s" "$1" | sed "s,','\\\\'',g; s,.*,'&',"
}

Is that also relying on this non-portable sed use?

> If this passes on BSD, then I'm okay with it; but if we want to avoid
> non-POSIX altogether, this should work (using the shell's $() to strip
> the trailing newline we added to keep sed happy):
>
> # backslash escape any '\' and '"' characters
> printf '"""%s""",' "$(printf "%s\n" "$e" | sed -e '/s/\([\"]\)/\\\1/g')"

Mmm.

-- PMM