[parisc-linux] Architecture string change

[parisc-linux] Architecture string change

[David Huggins-Daines]

Hi,

In keeping with the way things are done on other Linux platforms, I
would like us to start using the following architecture strings for
configuring our compilers:

32-bit userland and kernel, PA1.1: hppa-linux
64-bit kernel, PA2.0:              hppa64-linux

I guess this means that I volunteer to update the recipes on the
website :-)

config.guess will return a more detailed string, such as
'hppa2.0-unknown-linux-gnu', but these should be the canonical,
lowest-common-denominator strings.  In particular, we should revise
the recipes and such to use them and configure dpkg-architecture to
set DEB_GNU_HOST_ARCH to 'hppa-linux'.

I have not been able to get a solid answer out of anyone regarding
whether it is possible or desireable to use 'parisc-' instead of
'hppa-', so I will leave it as 'hppa' for the time being.  It is easy
enough to make 'parisc' an alias for 'hppa' if we want to.

Some pros and cons of using 'parisc' vs. 'hppa':

 * Pro: hppa*-linux targets produce very different 32-bit object code,
   as well as different 64-bit assembly code, so it may be beneficial
   to distinguish them from other PA-RISC targets.

 * Pro: I'm told that people at HP want to use 'parisc'.  Could those
   people please stand up :-)

 * Con: config.guess [1] will presumably return hppa*-*-linux-gnu.

 * Con: Every other operating system on PA-RISC uses hppa*-*-*.

(I know there are some more but I can't think of them right now)

Please remember that, now that we are entering userspace, any
decisions of this sort that we make are going to stick with us
forever. [2]

[1] Which I can't actually run since our page fault handler is broken:

> /bin/sh config.guess

do_page_fault() pid=8 command='sh'
                                  
Bad Address (null pointer deref?): Code=15 regs=c7e5c248 (Addr=00000001)
                                                                        
PSW  : 0004f80b  GR 1 : 00090350  GR 2 : 00007f4f  GR 3 : 00092e28  
GR 4 : 00000000  GR 5 : 00092b50  GR 6 : 00000019  GR 7 : 0000000e  
GR 8 : 00000001  GR 9 : 00092e28  GR10 : ffffffff  GR11 : 200206e2  
GR12 : 0000000a  GR13 : 00094b50  GR14 : 0007f621  GR15 : 00000000  
GR16 : 00094f0c  GR17 : 00000000  GR18 : 0000000e  GR19 : 00000001  
GR20 : 40100868  GR21 : 00000001  GR22 : 00000000  GR23 : 40100030  
GR24 : 40100848  GR25 : 00000020  GR26 : 40100010  GR27 : 0008f350  
GR28 : 40100850  GR29 : 00000001  GR30 : 20020880  GR31 : 000171d7  
SR0  : 00002001  SR1  : 00002001  SR2  : 00000100  SR3  : 00002001  
SR4  : 00002001  SR5  : 00002001  SR6  : 00002001  SR7  : 00002001  
                                                                    
IASQ : 00002001 00002001 IAOQ : 00007f53 00007f57
 IIR : 0e7c1280 ISR : 00002001 IOR : 00000001    

[2] A good example of this is how the soname for glibc is
'libc.so.6.1' on Linux/alpha instead of 'libc.so.6' due to a screwup
in an early Red Hat distribution...

-- 
dhd@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Re: [parisc-linux] Architecture string change

[Grant Grundler]

David Huggins-Daines wrote:
>  * Pro: I'm told that people at HP want to use 'parisc'.  Could those
>    people please stand up :-)

I'm one of them. But I'm not religous about it.

Just seems goofy to have "hp" in the name.
IMHO, It'd be comparable to using sunsparc or decalpha.

grant

Grant Grundler
Unix Development Lab
+1.408.447.7253

Re: [parisc-linux] Architecture string change

[John David Anglin]

> 
> David Huggins-Daines wrote:
> >  * Pro: I'm told that people at HP want to use 'parisc'.  Could those
> >    people please stand up :-)
> 
> I'm one of them. But I'm not religous about it.
> 
> Just seems goofy to have "hp" in the name.
> IMHO, It'd be comparable to using sunsparc or decalpha.

If it is going to change, why not just 'pa', for precision architecture.
Everbody knows it is risc.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: [parisc-linux] Architecture string change

[Bdale Garbee]

In article <87og45bqxh.fsf@linuxcare.com> you wrote:

> Some pros and cons of using 'parisc' vs. 'hppa':

I noted the last time I looked that all of HP's web-based documentation on 
the architecture seems to use 'parisc' pretty religiously.  

When I asked about the apparent confusion a while back, I got the very
strong indication that the parisc-linux community wanted to use 'parisc' for
the arhitecture string in Debian space... so that's the direction I've been
heading (mailing list, web pages, etc).  If we want to change, I can get it
done (no messages have passed the debian-parisc list yet despite over 50
subscribers, for example), but the sooner the better.

I have no emotional loading on this one, I just want to know the answer.  :-)

We *must* decide this once and for all *very* soon.  Pretty please.

Bdale

Re: [parisc-linux] Architecture string change

[Corne Beerse]

David Huggins-Daines wrote:
> 
> Hi,
> 
> In keeping with the way things are done on other Linux platforms, I
> would like us to start using the following architecture strings for
> configuring our compilers:
> 
> 32-bit userland and kernel, PA1.1: hppa-linux
> 64-bit kernel, PA2.0:              hppa64-linux
> 
> I guess this means that I volunteer to update the recipes on the
> website :-)
> 
> config.guess will return a more detailed string, such as
> 'hppa2.0-unknown-linux-gnu', but these should be the canonical,
> lowest-common-denominator strings.  In particular, we should revise
> the recipes and such to use them and configure dpkg-architecture to
> set DEB_GNU_HOST_ARCH to 'hppa-linux'.
> 
> I have not been able to get a solid answer out of anyone regarding
> whether it is possible or desireable to use 'parisc-' instead of
> 'hppa-', so I will leave it as 'hppa' for the time being.  It is easy
> enough to make 'parisc' an alias for 'hppa' if we want to.
> 
> Some pros and cons of using 'parisc' vs. 'hppa':
> 
>  * Pro: hppa*-linux targets produce very different 32-bit object code,
>    as well as different 64-bit assembly code, so it may be beneficial
>    to distinguish them from other PA-RISC targets.
> 
>  * Pro: I'm told that people at HP want to use 'parisc'.  Could those
>    people please stand up :-)
> 
>  * Con: config.guess [1] will presumably return hppa*-*-linux-gnu.
> 
>  * Con: Every other operating system on PA-RISC uses hppa*-*-*.
> 
> (I know there are some more but I can't think of them right now)
> 
> Please remember that, now that we are entering userspace, any
> decisions of this sort that we make are going to stick with us
> forever. [2]
> 
>>>>>> details on [1] and [2] removed <<<<<<<<<<<<<

My 2 cents on this architecture string:
I see the string contains 4 fields. For other linux implementations, what do
these fields mean? From what I see of it, is it:
<processorfamily>-<supplier>-<operatingsystem>-<???> ??

For the processorfamily, I would say 'parisk', without the HP in the name.
That's for the next field: it is an HP processor.

Compare it with the use on other platforms: For the base PC, it would be
i386-intel-linux-gnu or i386-amd-linux-gnu if there is difference between
the 2 i386 processors between intel and linux which makes a difference in
the configuration. If there is no difference between intel and amd at this
level, leave the supplier as unknown. For the 80386 type processors, there
is no difference to linux, hence: i386-unknown-linux-gnu. For the i686 type
processors it would be i686-intel-linux-gnu for Intels pentiumII and
i686-amd-linux-gnu for the amds K6.

How is it done on other platforms as the sparc, alpha and i386?



CB



-- 
RAM = Rarely Adequate Memory
Corne' Beerse					| Alcatel Telecom Nederland
mailto:beerse@ats.nld.alcatel.nl		| Postbus 3292
talkto:+31(70)3079108 faxto:+31(70)3079191	| NL-2280 GG  Rijswijk

Re: [parisc-linux] Architecture string change

[David Huggins-Daines]

Bdale Garbee <bdale@gag.com> writes:

> When I asked about the apparent confusion a while back, I got the very
> strong indication that the parisc-linux community wanted to use 'parisc' for
> the arhitecture string in Debian space... so that's the direction I've been

Well, there's no reason why DEB_HOST_GNU_CPU has to be the same as
DEB_HOST_ARCH.  Although I think I've seen a few debian/rules files
(including one of my own at one point) that do things like this:

build:
        ./configure --host=$(DEB_HOST_ARCH)-linux

Just to make it clear, this discussion is about DEB_HOST_GNU_CPU,
i.e. the CPU part of the string returned by config.guess and accepted
by configure, autoconf, and automake.

> I have no emotional loading on this one, I just want to know the answer.  :-)

I'm a bit biased towards 'hppa' just because it is the currently
accepted convention with all the GNU software out there.  Also it will
be a bit of a hack to config.guess to return parisc*, because we'll
end up only doing it on Linux.  Also, if we go with 'parisc', then:

(a) We need to send patches for config.guess and config.sub to the
    automake and autoconf people NOW.
(b) We must be prepared to manually update config.guess and config.sub
    in basically every piece of software we build. [1]

> We *must* decide this once and for all *very* soon.  Pretty please.

Yup :-)

[1] On Alpha, there continue to be annoying problems when building GNU
software because config.guess (which is part of autoconf) knew about
the PCA56 and EV6 subarchitectures long before config.sub (which is
part of automake) did, and thus configuring would always fail unless
--host=alpha-linux (or some other recognizable architecture string)
was specified.  Because the maintainer has to manually update
config.guess and config.sub, lots of packages still have old versions
that break...

-- 
dhd@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Re: [parisc-linux] Architecture string change

[Andrew Shugg]

One more vote for 'parisc'.  Is anyone counting?

I haven't seen good, clear reasoning yet as to why 'parisc' does not
adequately fit the requirements and 'hppa' does.

Bdale said:
> ... no messages have passed the debian-parisc list yet despite over 50
> subscribers ...

Surely you exaggerate, I see _ten_ messages in my debian-parisc box, not
including the subscription confirmation emails!

But seriously, I would guess that most of the subscribers are userland
hackers, not kernel people ... or, like me, aren't hackers at all,
they're just sad lonely people who own an HP box and like being on
mailing lists.  =)

When the announcement is made "okay, userland works and we're not going
to change anything any more, at least not without big huge announcements"
and people can start working with gusto on building packages, I'm sure
the list traffic will go up considerably.

We're just waiting for you to show us the way, in other words.  =)

Andrew.

--
Andrew Shugg <andrew@neep.com.au>                http://www.neep.com.au/

  "Just remember Basil, there's always someone worse off than yourself."
  "Oh, really?  I'd like to meet him ... I could do with a good laugh."
                 [ Sybil and Basil Fawlty, "Fawlty Towers" ]

Re: [parisc-linux] Architecture string change

[John David Anglin]

> 
> One more vote for 'parisc'.  Is anyone counting?
> 
> I haven't seen good, clear reasoning yet as to why 'parisc' does not
> adequately fit the requirements and 'hppa' does.