Re: 01-as-require-shell-fn.patch

[Paolo Bonzini]

> It won't suffice to try the few little tests that have been proposed
> so far. The only real test will be an extensive use of shell
> functions. So, I don't favor having a "spy" in Autoconf: I don't
> think it will suffice

It will if at least as a starter we don't use shell functions to full 
power.  (Hey, 20% reduction in autoconf run-time & configure size seems 
already nice to me).

> (and also I don't like to wade through all the
> useless chatter that such spies produce :-).

The problem with the "Present but cannot be compiled" spy was twofold.  
First, the spy should have told people to tell the maintainers and *CC* 
the autoconf mailing list.  Second, it should have made clear that it's 
not autoconf's bug, so that the need for a response was less.  But spies 
do work.

> I'd prefer to have Autotest use shell functions extensively,

Any ideas?  I did some work on Autotest last May and I do not have a 
clue about how to use shell functions there.  Autoconf has very clear 
duplication problems that are less pressing in Autotest.  (Though Derek 
may have some ideas).

Also, Autotest does convoluted stuff with subshells that make it much 
harder to use shell functions (and requires more cooperation from the 
shell than Autoconf will probably ever need).  And finally, we don't 
have an Autotest test suite which makes testing harder than for Autoconf.

Paolo