Re: also copy install-sh

[Ralf Wildenhues]

* Alexandre Duret-Lutz wrote on Mon, Dec 20, 2004 at 11:07:49PM CET:
> >>> "Ralf" == Ralf Wildenhues <address@hidden> writes:
> 
>  Ralf> Only if Libtool and Automake would carry different,
>  Ralf> incompatible versions of the script.
> 
> The difference will easily happen since install-sh still changes
> (8 changes just this year).  A similar problem occurred when
> gettextize started to install `mkinstalldir': people who kept
> `mkinstalldir' in CVS started to complain about the spurious
> changes to that file, caused by running one tool or the other.  
> 
> It is also possible that the HEAD version of install-sh (the one
> you are wgetting) is different from that from the stable branch
> (the one that is released).  This is not the case right now, but
> it has been the case in the past.
> 
> The incompatibilities could also happen.  For instance
> Makefile.ins produced by Automake >= 1.8 rely on the way the -d
> option has been changed to work in install-sh.

Bummer.

>  Ralf> Fortunately, it is pretty stable and mature, so we won't
>  Ralf> have to worry much about this.  With config.guess and
>  Ralf> config.sub, things have worked fine so far as well
>  Ralf> (AFAIK), and they tend to change much more often.
> 
> Would it make sense if libtoolize did not install any of these
> scripts when Automake is used?  That way there is always only
> one tool installing them in a given setup.

Probably.  But let's ask the question the other way round:  What could
be broken if libtoolize did not install any of these?
Next question:  How does libtoolize reliably detect whether Automake is
used (I can guess the answer, but am not certain)?
Another one:  Can we expect forward-compatibility (in the sense that
the newest available script will always be the best)?

Regards,
Ralf