Re: echo vs. printf regression (darwin8)

[David Fang]

> > > On 08/15/2010 06:59 PM, David Fang wrote:
> > > > The ECHO that is chosen by autoconf on darwin8 (printf %s\n) is buggy:
> > > >
> > > > fangism % printf %s\n foo
> > > > foonfangism% printf '%s\n' foo
> > > > foo
> > > >
> > > > Must be due to the version of the Bourne shell (2.05b).
> > > > Failure is similar with sh-3.2.
> > > > The 'correct' value of ECHO should be printf '%s\n' (note the extra
> > > > single quote protection).
> > > > I don't know yet whether that is forward-compatible with newer versions
> > > > of Bourne shell.
> > > >
> > > > Don't recall which version of autoconf broke this, betweem 2.63 and
> > > > latest.
> > >
> > > The problem seems to be in the way that "printf '%s\n'" is assigned to
> > > as_echo.
> >
> > I've tried to hack config.status with quote-protections and escapes:
> >
> > ECHO='printf "%s\n"'
> > ECHO='printf %s\\n'
> >
> > (and tried hacking the generated 'configure')
> >
> > But the substitution process (uses some creative sed magic) during
> > AC_OUTPUT keeps destroying the result.
>
> Yes, the problem is in how $ECHO is then "unquoted" by the shell.  We should 
> really use a shell function for echo (and printf for echo -n) instead.

My autotoolized projects all use $(ECHO) in Makefile{.am} inside the 
target scripts.  Everything that would normally use 'echo' is dying 
horribly, especially constructs like: var=`$(ECHO) ...`. 
Can you suggest a temporary work-around until that is fixed?
either at configure-time or post-configure-time?

I'm desperate enough to use sed to patch config.status after it's been 
generated:
        ECHO='printf %s\n'
with
        ECHO='echo'
to get the old behavior.

Fang

David Fang
http://www.csl.cornell.edu/~fang/
http://www.achronix.com/