egcs 1.0.3a build report

[Martin Andrews <andrewm at eeg dot ccf dot org>]

I have succesfully buil egcs-1.0.3a on the following platforms:

*** hppa1.1-hp-hpux10.20 --with-gnu-as

output from "gcc -v":

Reading specs from
/net/appl/egcs-1.0/lib/gcc-lib/hppa1.1-hp-hpux10.20/egcs-2.90.29/specs
gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)

output from "uname -a":

HP-UX dev2 B.10.20 A 9000/778 2003168385 two-user license

*** hppa1.1-hp-hpux9.05 --with-gnu-as

output from "gcc -v":

Reading specs from
/net/appl/egcs-1.0/lib/gcc-lib/hppa1.1-hp-hpux9.05/egcs-2.90.29/specs
gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)

output from "uname -a":

HP-UX imadev A.09.05 A 9000/715 2009123119 two-user license

*** m68k-hp-hpux9.03 --with-gnu-as --with-gnu-ld

output from "gcc -v":

Reading specs from
/net/appl/egcs-1.0/lib/gcc-lib/m68k-hp-hpux9.03/egcs-2.90.29/specs
gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)

output from "uname -a":

HP-UX vue9 B.09.03 A 9000/42T 080009323173 two-user license

--------------------

I have run into two minor problems:

1) There is a bug in varargs handling for hppa - you cannot loop through
the arguments multiple times (this problem  was in gcc 2.7 too).  The
following program will core dump when compiled by egcs for HP PA-RISC:

#include <stdio.h>
#include <stdarg.h>

void
stub(int num, ...)
{
    va_list ap;
    char *end;
    int i;

    for (i = 0; i < 100; i++) {
        va_start(ap, num);
        while ( 1 ) {
            end = va_arg(ap, char *);
            if (!end) break;
            printf("end is %s\n", end);
        }
        va_end(ap);
    }
}

int
main()
{
    stub(1, "ab", "bc", "cx", NULL);
    return 0;
}

2) I built binutils-2.8.1 in the same directory as egcs in a single
"make bootstrap" - I ran into a problem with the HP machines in the
second stage where the gnu assembler was not being found (the HP as was
being run) - I fixed this problem with the following hack:

if [ ! -h gas/as-new -a ! -f gas/as-new ]
then
    ln -s as.new gas/as-new
fi



-- 
Martin Andrews                      andrewm@eeg.ccf.org
Section of Neurological Computing, Cleveland Clinic Foundation