Re: 1215 build successes

Re: 1215 build successes

[Mike Stump]

> To: egcs@cygnus.com
> From: jhpb@sarto.gaithersburg.md.us (Joseph H. Buehler)
> Date: 01 Jan 1998 02:51:21 -0500

> I finally came up with a (non-trivial!) shell script that builds the
> following properly:

>     HOST hppa1.0-hp-hpux10.10       TARGET rs6000-ibm-aix3.2.5
[ ... ]

I would like to see the intelligence embodied in your script built
into gcc.  I think it would be good, as then more people would do up
up cross compilation systems for systems they don't have, which gains
them the ability to produce binaries for machines they don't have
(useful).  Also, bootstrapping then becomes easier on machines without
compilers, as all you need is one machine with any C compiler to
bootstrap up to all machines will all (GNU) compilers.

The good news is that quite a lot of the requisite intelligence is
already in gcc, it would be good at get this last bit in... so that
the next person to do it, has to just supply some information, and
type make (opps, I mean configure; make).

If you have the time and desire, would love to see the work folded in.

Re: 1215 build successes

[Joseph H. Buehler]

>>>>> "MS" == Mike Stump <mrs@wrs.com> writes:

    >> I finally came up with a (non-trivial!) shell script that builds the
    >> following properly:

    >> HOST hppa1.0-hp-hpux10.10       TARGET rs6000-ibm-aix3.2.5
    MS> [ ... ]

    MS> I would like to see the intelligence embodied in your script built
    MS> into gcc.  I think it would be good, as then more people would do up
    MS> up cross compilation systems for systems they don't have, which gains
    MS> them the ability to produce binaries for machines they don't have
    MS> (useful).  Also, bootstrapping then becomes easier on machines without
    MS> compilers, as all you need is one machine with any C compiler to
    MS> bootstrap up to all machines will all (GNU) compilers.

    MS> The good news is that quite a lot of the requisite intelligence is
    MS> already in gcc, it would be good at get this last bit in... so that
    MS> the next person to do it, has to just supply some information, and
    MS> type make (opps, I mean configure; make).

    MS> If you have the time and desire, would love to see the work folded in.

The C library is currently a problem.  The current gcc/egcs requires a
C library (and associated includes) that matches the target platform.
I presume that the GNU C library will eventually be ported to more
platforms, but it only builds for a few platforms at the moment.  So
the compilers I build use NFS-mounted libraries and include files from
target machines that *do* have native compilers.  (And I haven't even
checked with the manufacturers yet to see if this arrangement violates
any licenses.)

The amount of work involved in getting my egcs cross-compilers built
got me thinking about all this.  The industry is wasting *incredible*
amounts of time.  How many new instruction sets do we really need?  Do
we really need new OS releases that introduce yet more header and
library incompatibilities?  Do we really need 37 different kinds of
UNIX using 37 different compilers?

Seems to me there must be a real market for cross-compilation
environments.  It is very attractive to be able to build for target
machines, using whatever you have handy.  In my case, we have an
RS6000 that is bogged down, and an SMP HPUX box that screams, so a
cross-compiler is an obvious solution.

It is also nice to be able to use just one compiler on all platforms.
We build on HPUX 9, HPUX 10, AIX 3 and AIX 4, and the compilers are
all different, and this causes major pain, we constantly run into
something that builds on one machine but not on another, because of
the way the compiler does template instantiation, how signal handlers
are defined, or whatever.

Joe Buehler

1215 build successes

[Joseph H. Buehler]

After working on various versions of egcs since early November or so,
I finally came up with a (non-trivial!) shell script that builds the
following properly:

    HOST hppa1.0-hp-hpux10.10       TARGET hppa1.0-hp-hpux10.10
    HOST hppa1.0-hp-hpux10.10       TARGET rs6000-ibm-aix3.2.5

    HOST hppa1.1-hp-hpux9.04        TARGET hppa1.1-hp-hpux9.04
    HOST hppa1.1-hp-hpux9.04        TARGET rs6000-ibm-aix3.2.5

    HOST powerpc-ibm-aix4.2.0.0     TARGET powerpc-ibm-aix4.2.0.0
    HOST powerpc-ibm-aix4.2.0.0     TARGET rs6000-ibm-aix3.2.5

    HOST rs6000-ibm-aix3.2.5        TARGET rs6000-ibm-aix3.2.5

- I did not build either fortran or objective-c.

- The native compilers are configured like the cross-compilers, taking
  the includes and libraries from under /usr/local, so I can run them
  on machines that have no other installed compiler.

Joe Buehler