RISCV changes broke 32-bit --enable-targets=all

[Nick Alcock]

On 31 Jul 2020, Maciej W. Rozycki via Binutils uttered the following:

> On Sat, 11 Jul 2020, Palmer Dabbelt wrote:
>
>> > By the look of it, git commit 39ff0b812324 broke 32-bit host
>> > --enable-targets=all binutils builds.
>> >
>> > /usr/local/bin/ld: ../opcodes/.libs/libopcodes.a(riscv-dis.o): in function `parse_riscv_dis_option':
>> > /home/alan/src/binutils-gdb/opcodes/riscv-dis.c:102: undefined reference to `riscv_get_priv_spec_class'
>> > collect2: error: ld returned 1 exit status
>> > Makefile:925: recipe for target 'objdump' failed
>> 
>> Sorry about that.  I know I haven't been checking --enable-targets=all builds,
>> and I'm not sure if anyone else has either.
>
>  FWIW I routinely have (also with GDB), though not with a 32-bit host due 
> to their mostly diminished role as processing work force.  In reality I 

It's also part of my standard test matrix before pushing CTF stuff (both
on 32- and 64-bit x86 hosts because I broke non-ELF in the past and I
don't want to do it again, and becaue the CTF deduplicator contains some
32-bit-specific code...)

(I also test that 'make -k check-binutils check-gas check-ld' on a long
list of cross targets (which is meant to be more or less every target we
implement) doesn't regress across the range of each series I work on. It
takes about an hour even parallelized, but the number of times it's told
me that I regressed 'make check' or simply broke compilation on some
obscure target I'd never otherwise have thought of testing makes it time
well spent. This is not my idea: hat tip to Jose Marchesi for both the
idea and 90% of the code...)