Re: Autoconf, DOS and NT (was Ebcdic rule)

[Paul Eggert]

> From: Mike Castle <address@hidden>
> Date: Thu, 4 Oct 2001 11:28:55 -0700
> 
> On Wed, Oct 03, 2001 at 11:52:11PM -0700, Paul Eggert wrote:
> > Autoconf is merely a tool for configuring source code for UNIX-like
> > systems.  That problem is hard enough.
> 
> I didn't realize that DOS and NT were considered UNIX-like systems.

Actually, I was paraphrasing the first sentence of the Autoconf manual,
which is:

  Autoconf is a tool for producing shell scripts that automatically
  configure software source code packages to adapt to many kinds of
  UNIX-like systems.

My understanding is that bare DOS is not UNIX-like, but it gets
reasonably UNIX-like if you add enough 3rd-party software.  Similarly
for NT.

Autoconf has some support for Microsoft's proprietary operating
systems.  Maintaining this support is not a primary goal of the GNU
project, but if it's easy to separate the Microsoft-related code and
documentation so that it doesn't get in the way of the principal goals
of Autoconf, it's OK to leave it in.

However, what's _not_ OK is when the Microsoft support breaks things
on UNIX-like systems.  Unfortunately the current Autoconf snapshot has
a fair number of these problems, mostly having to do with placing
arbitrary restrictions on file names.  Autoconf gets away with this
only because Autoconf historically has not allowed arbitrary UNIX
pathnames in filenames, and the Microsoft naming restrictions tend to
squeeze into the cracks.  However, the "right" thing to do would be to
remove all these restrictions.

Also, it's definitely not OK for the Autoconf code to get littered
with special cases for Microsoft.  As much as possible, the
Microsoft-related stuff should be segregated out, and the mainline
Autoconf code should use a plain POSIX-subset style.
For more on this subject (in the C context, but the same idea applies
to the shell), please see Spencer and Collyer's classic paper "#ifdef
Considered Harmful, or Portability Experience with C News" (Summer
1992 USENIX, pages 185-198)
<http://www.cs.umd.edu/users/cml/cstyle/ifdefs.pdf>.
(Maybe the Autoconf manual should refer to this paper?)