Re: saving AC development strength

[Bruce Korb]

Paul Eggert wrote:
> OK, but that hand-holding teaches people to replace every `[' and `]'
> with a quadrigraph.  So:
> 
> * They learn the wrong quoting rules.

Not wrong, pug-ugly.

> * They say "It's sooo ugly; you can't be serious!".

It is the consequence of choosing a quoting phrase that is
indistinguishable from square brackets used in user code.
All solutions are pug-ugly.  The only approach is to choose
one that does not have harmful and obtuse consequences.
Blowing up with an orthoginal error message because someone
supplied an unmatched square bracket is, to me, completely
unacceptable.  Better the uglitude.  Anyway, it is tooo ugly,
but better than a failure.

> In this particular case it may be better to teach them to fish
> than to give them a McDonald's Filet-O-Fish Sandwich.

Well, that it should be so hard to drop a fish pellet into
boiling fat, slap it in bread and push it through the window
bespeaks of some severe design issues.  It would be far better
(in my opinion) to change the quoting phrases to something
that is unlikely to be seen as a normal token sequence.  e.g., >>>
and <<< -- but anything will do.  I have seen many references
to "no, no, no" with no clear explanation as to "why not?"

Tom wrote:
> it's worse than that: the prevailing philosophy of autoconf 2.5x
> is that user-defined macros are neither desirable nor supported.

It would seem so, sometimes.  ;-)

Akim wrote:
> No, I'm saying that with a small scale project, problems are less
> likely to appear than with a full scale Autoconf.  In addition, the
> figures are not faithful: 2.5x is *much* more factored than 2.13.
> Some day I should do statistics on the average nested levels.

I do not think "average nested levels" applies to my application.
Basicly, the conftest generated macros would be basically prohibited
from directly referencing autoconf macros.

> Also, 2.12 and 2.13 [were] incredibly badly quoted.

I am certain you fixed that problem.  It would still be enlightening
to know what the issues are with wrapping autoconf-macro-free text
inside changequote-s.  I am convinced it is the wrong answer to
expect conftest users to always and only enter balanced square bracket
text.  The failure mode is too awful and all the other alternatives
are ugly.  Ugly all ways around.

-- 

Bruce Korb <first initial + last name at gnu dot org>
AG URL: http://autogen.sourceforge.net