Re: [PATCH 1/5] emit-rtl.c: more typesafety

[David Malcolm <dmalcolm at redhat dot com>]

On Tue, 2018-06-12 at 15:23 -0600, Jeff Law wrote:
> On 06/12/2018 03:16 PM, David Malcolm wrote:
> > On Tue, 2018-06-12 at 14:50 -0600, Jeff Law wrote:
> > 
> > Ultimately these boil down to:
> > 
> > gcc_checking_assert (INSN_P (rt)
> >                      || NOTE_P (rt)
> >                      || JUMP_TABLE_DATA_P (rt)
> >                      || BARRIER_P (rt)
> >                      || LABEL_P (rt));
> > 
> > (with an "if (rt)" for the safe_a_a variant), and hence they're
> > runtime
> > checks (and only enabled when #if CHECKING_P).
> 
> Ack.
> 
> > 
> > > Obviously your testing ought to be sufficient for the
> > > former.  Mine will help with the latter since it'll build arc,
> > > bfin
> > > and
> > > c6x elf targets through newlib.  For sh it'll build a mixture of
> > > -elf
> > > things through newlib and it'll bootstrap sh4.
> > 
> > Is this testing something that you've already triggered, or does it
> > happen upon commit to trunk?
> 
> THe jobs poll the various relevant git servers (gcc, binutils,
> newlib,
> glibc, linux kernel) once per day .  When a change is detected in any
> of
> the repos a full build is started.
> 
> 
> > 
> > (The sh changes are the only aspect of this kit that I'm keen for
> > review/testing of, as it needed extra casts; I *think* everywhere I
> > added them are places where rtx are already being unconditionally
> > treated as instruction nodes, but your eyes/testing would be
> > helpful).
> 
> They looked reasonable to me.  Thankfully sh is one that we can
> actually
> bootstrap GCC within a chroot using qemu, so it'll get fairly
> extensive
> testing.   It takes ~8hrs, but that's not too bad :-)

Thanks.  For reference, I've committed the combination of the patches
to trunk as r261547.

Dave

(I also smoketested the build of arc, since I noticed some changes
there since my earlier testing).