Re: [PATCH] add support for strnlen (PR 81384)

[Martin Sebor <msebor at gmail dot com>]

On 06/19/2018 02:10 PM, Martin Sebor wrote:
> On 06/19/2018 01:33 PM, David Edelsohn wrote:
> > Martin,
> >
> > Does attr-nonstring-3.c assume a 64 bit environment?
>
> I don't think so.  The error below suggests a problem during
> the expansion of the strncmp built-in.  The patch didn't
> change that.  I see GCC 8 ICE on the test as well so it must
> be a latent bug that the test somehow manages to tickle.  Let
> me try to reduce it to a small test case and open a bug.

I opened 86222 and CC'd you on it.  The problem can be triggered
by calling strncpy() with a constant bound between PTRDIFF_MAX
+ 1 and PTRDIFF_MAX + 32.

Martin

> Martin
>
> > I'm seeing new errors on the rs6000 port
> >
> > FAIL: gcc.dg/attr-nonstring-3.c (internal compiler error)
> > /nasfarm/edelsohn/src/src/gcc/testsuite/gcc.dg/attr-nonstring-3.c:73:1:
> > error: could not split insn
> > (insn 3244 3245 3246 (set (reg:SI 5 5)
> >         (const_int -2147483680 [0xffffffff7fffffe0]))
> > "/nasfarm/edelsohn/src/src/gcc/testsuite/gcc.dg/attr-nonstring-3.c":72
> > 446 {*movsi_internal1}
> >      (nil))
> > during RTL pass: final
> > /nasfarm/edelsohn/src/src/gcc/testsuite/gcc.dg/attr-nonstring-3.c:73:1:
> > internal compiler error: in final_scan_insn_1, at final.c:3140
> > ranges offset out of range
> >
> > Something is assuming that a 64 bit value can fit in a 32 bit register.
> >
> > Thanks, David