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

[Ralf Corsepius]

On Wed, 2004-01-07 at 22:51, Harlan Stenn wrote:

> > A configure script that has to check for 125 brand names, only for Linux,
> > is not only unmaintainable, it also limits the freedom to fork a new
> > distribution.
> 
> So for this reason people who write scripts (autoconf or otherwise) who can
> successfully use config.guess to decide what the CPU, Vendor, and OS are
> (which is what the original specs declared) are now in the position to only
> have the OS info for non-gnu-linux boxes, because on gnu-linux boxes
> everybody who wants to make OS-level decisions has to code their own tests
> to figure out the OS name.
IMO, you are forgetting some essential points:
 
* config.guess is a last resort means to help  escaping from "helpless
situations"

* Using config.guess'ed values means table lookup based configurations.
This approach more than once has proven to be unmaintainable.

* autoconf's working principle is feature checking.

* autoconf is not in a position to dictate the tuples - Distributors
dictate them.

* There is no strong relationship between "OS" and features. Any system
admin (Esp. under Linux) has the freedom to change a particular
installation at his/her free will. I.e. any guess on a system's feature
based on a config.guess'ed "OS" value is flaky and basically
meaningless.

> And yes, I am aware that distributions are not OSes, but hey, some of them
> are getting close.
> 
> It is a cost shift that is really stupid and wasteful of a lot of resources.
I think, your are confused.

> In my opinion.
Sorry for having to say this, but IMO, configure scripts relying on
config.guess'ed values are "badly designed and fundamentally flawed".

BTW: If you really think, you want/need different target tuples, nothing
prevents you from implementing your own extended/modified config.guess. 

Ralf