Re: converting netkit to autoconf?

[Bruce Korb]

Akim Demaille wrote:

> >> M4's rules are sometimes painful, but the shell's are often worse.
> 
> Bruce> No way.  What is crucial in quoting rules is that they be
> Bruce> consistent.  M4's are not.  Second in importance is that they
> Bruce> be few in number.  M4, si, shell, no!
> 
> Your sentence makes no sense to me.  M4 quotes are the most consistent
> one can design.

The rules are consistent.  The quoting is not.
The author of a macro must know what the current quoting sequence
is and the invoker of that macro must ensure that that is, in fact,
the current sequence when the macro is invoked.  If handling the
quoting is so simple, why do you suppose so many people find it so
hard?  Why is it that the history of autoconf is littered with
people bashing their brains out over broken quoting?  I submit that
consistent quoting rules does not necessarily make for consistent
quoting.

> [[...]]  But
> once that problem solved with quadrigraph, you end up with the most
> simple system you could imagine.

"quadrigraph" and "simple" in the same sentence?  8-O
I'm pretty sure we have different ideas about "simple".
It is hard enough to read and understand macros without
the additional obfuscation of quadrigraphs.

> My own grief wrt M4 is that programming properly in M4 is programming
> in a lazy language.  But (i) the language itself does not promote Lazy
> Functional Programming enough, and (ii) (this is bound to (i)) the
> performances are very bad since it does not take advantage of all that
> can be done when compiling such languages.

Performance issues are vastly overwhelmed by programming difficulty.

> I expect a lot from shell functions.  Including saving
> space, and improving performances  (with good shells).

With configure scripts buring half a meg or more of space,
I'm thinking space might be the greater concern. ...