Re: PATCH to fix bogus warning with -Wstringop-truncation -g (PR tree-optimization/84228)

[Martin Sebor <msebor at gmail dot com>]

> > > Ok for trunk, though generally looking at just next stmt is very
> > > fragile, might be
> > > better to look at the strncpy's vuse immediate uses if they are
> > > within the
> > > same basic block and either don't alias with it, or are the store it is
> > > looking for, or something similar.
> >
> > I guess some
> > FOR_EACH_IMM_USE_FAST (...)
> >   {
> >     if (is_gimple_debug (USE_STMT (use_p)))
> >       continue;
> > ...
> > would be better.
>
> Jeff and I had a fairly extensive discussion about how best to
> handle debug statements.  I had prototyped his suggestion along
> these lines but ran into false positives and other difficulties.
> The details are here:
>
>   https://gcc.gnu.org/ml/gcc-patches/2017-11/msg00805.html
>
> I thought I ended up just using gsi_next_nondebug() and added
> a test for it but I guess I must be misremembering.  The patch
> review spanned a couple of months so it's even possible I forgot
> to make the change.

I looked though my hard drive for the test I remember writing
for this.  I couldn't find it but from memory here's roughly
what it did:

  char d[3];

  void f (const char *s)
  {
    int i;
    __builtin_strncpy (d, s, sizeof d);   // false positive with -g
    i = 0;
    d[sizeof d - 1] = 0;
  }

Martin