Re: can't build for z8k-coff

[Jeff Johnston <jjohnstn at redhat dot com>]

Christian Groessler wrote:
> > If you didn't specify an option like -i on the build, then check which make
> > you are using and if there are any updates available.  You can set up a small
> > test case to reproduce the problem and if still present you should open a bug.
>
> I'm using gmake 3.81. I didn't specify -i.
> According to the man page of gmake return code 1 should only happen
> if I specify -q, which I didn't. I will check with gmake 3.80 to see
> if it happens there too.

Christian, I should clarify my earlier statements regarding error code. 
 I was referring to the typical percolation I expect to see on a newlib 
compile error.  I have an example below for mn10300-elf whereby I simply 
added a non-existent type to the file:

/home/jjohnstn/sourceware/src/newlib/libc/argz/argz_insert.c: In 
function ‘argz_insert’:
/home/jjohnstn/sourceware/src/newlib/libc/argz/argz_insert.c:25: error: 
‘blahblah’ undeclared (first use in this function)
/home/jjohnstn/sourceware/src/newlib/libc/argz/argz_insert.c:25: error: 
(Each undeclared identifier is reported only once
/home/jjohnstn/sourceware/src/newlib/libc/argz/argz_insert.c:25: error: 
for each function it appears in.)
/home/jjohnstn/sourceware/src/newlib/libc/argz/argz_insert.c:25: error: 
expected ‘;’ before ‘k’
make[7]: *** [lib_a-argz_insert.o] Error 1
make[7]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/am33/newlib/libc/argz'
make[6]: *** [all-recursive] Error 1
make[6]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/am33/newlib/libc'
make[5]: *** [all-recursive] Error 1
make[5]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/am33/newlib'
make[4]: *** [all] Error 2
make[4]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/am33/newlib'
make[3]: *** [multi-do] Error 1
make[3]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/newlib'
make[2]: *** [all-multi] Error 2
make[2]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/newlib'
make[1]: *** [all] Error 2
make[1]: Leaving directory 
`/notnfs/jjohnstn/sourceware/mn10300/mn10300-elf/newlib'
make: *** [all-target-newlib] Error 2

The error 1 code for lib_a-argz_insert.o from gcc is expected.  It 
should have percolated up to [all] with Error 2 unless you specified -i 
or -k.

> > It is likely that I gummed this up when I brought the code in originally and
> > modified it to work for newlib.  Obviously, this new code has never been
> > compiled on the z8k.  I've made up a proposed patch to fix this. Try it out
> > and let me know.
>
> It works. But in order for the compile to succeed, I had to do another
> change (appended).
> It is also a variable declaration in the middle of a function.
> I think there is no parameter to tell current gcc to report an error for
> such things, even with -std=c89 :-(   This is very annoying...

Thanks, will check in the patch.

-- Jeff J.

> regards,
> chris
>
> Index: ldtoa.c
> ===================================================================
> RCS file: /cvs/src/src/newlib/libc/stdlib/ldtoa.c,v
> retrieving revision 1.11
> diff -u -r1.11 ldtoa.c
> --- ldtoa.c     16 May 2007 19:59:40 -0000      1.11
> +++ ldtoa.c     6 Jun 2007 18:15:18 -0000
> @@ -2868,9 +2868,6 @@
>  LDPARMS rnd;
>  LDPARMS *ldp = &rnd;
>  
> -rnd.rlast = -1;
> -rnd.rndprc = NBITS;
> -
>  union uconv du;
>  du.d = *d;
>  #if LDBL_MANT_DIG == 24
> @@ -2883,6 +2880,9 @@
>  e113toe( &du.pe, e, ldp );
>  #endif
>  
> +rnd.rlast = -1;
> +rnd.rndprc = NBITS;
> +
>  if( (e[NE-1] & 0x7fff) == 0x7fff )
>         {
>  #ifdef NANS