Re: config.guess and freedom (was: 1.8 and mkdir_p)

[Harlan Stenn]

> Harlan Stenn wrote (meaning "Linux distribution" when he writes "OS"):
> > help tool maintainers make choices
> > about how things that are hard to find out otherwise (like OS-based
> > choices).
> > ...
> > everybody who wants to make OS-level decisions has to code their own tests
> > to figure out the OS name.
> 
> The point is: Don't make choices at all based on the distribution's name!
> Not only it limits the freedom to create new distributions, it's also more
> costly to maintain a list of features/commands/install-locations/etc.
> depending on the distribution, than to write an autoconf test.

The good news and bad news is that your position is a POLICY decision.

I am talking about a MECHANISM tool.

It is unreasonable to expect tool/package maintainers to do this.

It is unreasonable to expect that tool/package maintainers are the only ones
who need to do this.

config.guess is also useful (on non-"gnu-linux" systems) for specifying what
OS partition to mount for automounters, just to name one example.

Different OS versions often have different libc versions (for example), so
while it may be easy to run a SunOS5 binary on later versions of the OS, it
may not be possible to run a SunOS9 binary on earlier versions of the OS.

> autoconf was developed 10 years ago when there were dozens of different
> Unix variants, and it helped a lot. Now we have dozens of Linux
> distributions, and autoconf helps again. Why would you need to check for
> SuSE Linux in order to decide whether to install a startup script in
> /etc/init.d or /etc/rc.d? Better test whether each of these directory
> exists. Then you won't have to change your test the day SuSE Linux is
> renamed into Novell Linux (just a hypothetical example).

Well, in the old days we used sysIII and sysV in various incarnations to
avoid coding a different value for every reseller of those OSes.  Simliarly
for the different BSD releases.

If the releases are all that similar, why not use:

 i686-GnuLinux-*

as your test, and provide the "popular" distributions in the 3rd field?

The "magic" command has a large database of selections on it; using this
sort of mechanism should greatly ease the burder on the config.*
maintainers.

config.guess started out life as a useful infrastructure tool, and the
arrival of the 4-part format and the generic "gnu-linux" tag renders it
quite useless for anything other than configure.

A *ton* of people have to work around the current behavior, and there isn't
a really useful and easy place to share knowledge about how to determine and
test for additional knowledge.  Mailing lists just don't do it.  Subject
threads are problematic, and one often has to plow thru tons of cruft to get
to useful nuggets that are often strewn across many messages.

Can somebody put up a twiki for auto* macros and addons?

H