Re: New members in struct format_ops in gas/obj.h used with multiformattargets

[Alan Modra <alan at linuxcare dot com dot au>]

On Sun, 21 May 2000, Hans-Peter Nilsson wrote:

> "env GNUTARGET=a.out-i386 AS_EMULATION=i386aout make check".
> Slightly better with these patches: the ld test "check sections" now
> assemble and run correctly (test 1 and 2).  ELF-specific tests break with
> and without the patches of course, but then they are expected to.  Perhaps
> the testsuite should be corrected to look at the current emulation, not
> the configured target for which tests to run.  There are two testfailures
> (with and without the patches) that look real: "i386 reloc" and "i386
> white".  They complain about "cannot represent relocation type
> BFD_RELOC_8", which seems unhealthy.

These failures are just like the ELF ones in that the testsuite is fooled
into thinking that the target format is ELF, when the emulation code has
changed the output to something else.  See gas/testsuite/gas/i386/i386.exp

> mipsel-linux-gnuecoff:  (Is that really a valid target?  There's
> alpha*-*-linuxecoff* in gas/configure.in, but I dreamt this one up.)
> 
> "make check": No differences.  Note that there are 9 unexpected failures
> with the gas tests with and without these patches.  To the untrained eye
> they look like real problems.

The mips testsuite needs some tidying.  I think most of the failures are
due to wrong padding, different nops, and other fairly simple problems.