Re: MANUFACTURER in canonicalization triple

2000-04-28 Thread Peter Eisentraut

On Thu, 27 Apr 2000, Ralf Corsepius wrote:

 What is the MANUFACTURER aka VENDOR field in a canonicalization triple
 meant to contain?

AFAIC, it has always been pretty useless, especially since the
config.guess maintainers don't seem to be very sure about it themselves
and change it back and forth. For example, the "vendor" of a i386-*-qnx*
machine has changed from "pc" to "qnx" back to "pc". Also there seem to be
*-unknown-linux* and *-pc-linux* machines out there.

I wonder what it really might be useful for. If it's the processor vendor,
the encode that in the cpu name. (Why would you care?) If it's the os
vendor, then encode that in the os name. (What OS has two different
vendors?) That leaves packagers, but if you depend on that information
without testing for features, you're really pushing your luck.

IMO, config.{guess|sub} are sometimes a painful hogwash of overly specific
and overly general, so using the output for anything but naming your
cross-toolchain isn't going to work very well in general.


-- 
Peter Eisentraut  Sernanders väg 10:115
[EMAIL PROTECTED]   75262 Uppsala
http://yi.org/peter-e/Sweden




MANUFACTURER in canonicalization triple

2000-04-27 Thread Ralf Corsepius

Hi,

What is the MANUFACTURER aka VENDOR field in a canonicalization triple
meant to contain?

My understanding is that it shall describe the vendor/manufacturer of a
board or better the board family, being used to guess/preset settings
for a specific setups, primarily in near-hardware packages (e.g.
binutils).
I know the VENDOR field is rarely used at all, therefore the value won't
matter in most cases, but as we're  at clarifying canonicalization, this
could become an issue.

I am asking, because i386-pc-linux distributors have started to set this
field to their name for the gnu-toolchains distributed with their
distributions (IMHO, an uncredible decision ;).

However, a PC still remains a PC and is not a SuSE, RH, Debian or
whatsoever machine. 

Imagine the consequences of:
./configure --build=i386-suse-linux-gnu --host=i386-redhat-linux-gnu

Ralf.

-- 
Ralf Corsepius 
Forschungsinstitut fuer Anwendungsorientierte Wissensverarbeitung (FAW)
Helmholtzstr. 16, 89081 Ulm, Germany Tel: +49/731/501-8690
mailto:[EMAIL PROTECTED]   FAX: +49/731/501-999  
http://www.faw.uni-ulm.de




Re: MANUFACTURER in canonicalization triple

2000-04-27 Thread Ian Lance Taylor

   Date: Thu, 27 Apr 2000 13:52:53 +0200
   From: Ralf Corsepius [EMAIL PROTECTED]

   What is the MANUFACTURER aka VENDOR field in a canonicalization triple
   meant to contain?

It normally contains the manufacturer or vendor of the hardware.

   I am asking, because i386-pc-linux distributors have started to set this
   field to their name for the gnu-toolchains distributed with their
   distributions (IMHO, an uncredible decision ;).

   However, a PC still remains a PC and is not a SuSE, RH, Debian or
   whatsoever machine. 

   Imagine the consequences of:
   ./configure --build=i386-suse-linux-gnu --host=i386-redhat-linux-gnu

I think those configuration triplets are wrong.  I think they should
be something like `i386-pc-linuxsuse7.0-gnu' or
`i386-pc-linux-gnususe7.0'.

Ian




Re: MANUFACTURER in canonicalization triple

2000-04-27 Thread Pavel Roskin

Hello!

Imagine the consequences of:
./configure --build=i386-suse-linux-gnu --host=i386-redhat-linux-gnu

It doesn't scare me :-)
You get what you are asking for.

 I think those configuration triplets are wrong.  I think they should
 be something like `i386-pc-linuxsuse7.0-gnu' or
 `i386-pc-linux-gnususe7.0'.

It it not better. "linux" should have the kernel version at the end, as
this is much more relevant. Maybe "gnu" should be GLibc version:

i386-pc-linux2.3.99_pre5-gnu2.1.3_pre3

We need to choose whether VENDOR is software vendor or hardware vendor. In
the first case "redhat" and "suse" are absolutely legal. Otherwise it
should be "intel" or "amd". Note that in either case I don't expect from 
config.guess to distinguish between vendors, at least on Linux.

I vote for "hardware vendors". The argument for that is that we should
favor individual developers and users, not companies. SuSE and RedHat have
package managers in their hands, so the can easily deal with the absence
of the "software vendor" field.

Software developers, on another hand, should never target RedHat or
SuSE explicitly, i.e. they should not treat $host_vendor as "software
vendor"

I don't think that RedHat will be happy if e.g. Lynx developers will
decide that lynx.cfg goes to /etc/ just because gcc is configured by
RedHat. It should be RedHat that decides how to make software fit they
distribution guidelines.

$host_vendor could be of some help for people who cross-compile for
embedded boards. In fact, I would rather have "board name" in that place
as a nice complement to "CPU name", but it's a different story.

Regards,
Pavel Roskin




Re: MANUFACTURER in canonicalization triple

2000-04-27 Thread Bob Friesenhahn

On Thu, 27 Apr 2000, Pavel Roskin wrote:
 
  I think those configuration triplets are wrong.  I think they should
  be something like `i386-pc-linuxsuse7.0-gnu' or
  `i386-pc-linux-gnususe7.0'.
 
 It it not better. "linux" should have the kernel version at the end, as
 this is much more relevant. Maybe "gnu" should be GLibc version:
 
 i386-pc-linux2.3.99_pre5-gnu2.1.3_pre3

You guys are plumb crazy.  Maybe we don't really need Autoconf anymore
since we can just encode everything we need in the output of
config.guess.  Everyone is moving to Linux, after all, so there is
really no need to know anything more than the kernel version, the name
of the distribution, the processor family, and the C library version.  
Once we eliminate all those annoying feature tests, then 'configure'
will run a whole lot faster.

Bob
==
Bob Friesenhahn
[EMAIL PROTECTED]
http://www.simplesystems.org/users/bfriesen