Re: [patch, libfortran] [patch, fortran] Fix PR 79956, part two, attempt 2

[Markus Trippelsdorf <markus at trippelsdorf dot de>]

On 2017.03.13 at 23:24 +0100, Thomas Koenig wrote:
> Hello world,
> 
> Following Richard's suggestion, I have implemented the GFC_ASSERT macro
> and used it to (hopefully) silence the ominous warning and allow
> further optimization.

Unfortunately the single GFC_ASSERT isn't enough:

libgfortran/intrinsics/reshape_generic.c: In function ‘reshape_internal’:
libgfortran/intrinsics/reshape_generic.c:257:21: warning: ‘sstride[0]’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   sstride0 = sstride[0] * size;
              ~~~~~~~^~~
The following patch is needed to silence them all.

diff --git a/libgfortran/intrinsics/reshape_generic.c b/libgfortran/intrinsics/reshape_generic.c
index 43a822f87ae..d7c9e74853a 100644
--- a/libgfortran/intrinsics/reshape_generic.c
+++ b/libgfortran/intrinsics/reshape_generic.c
@@ -66,6 +66,7 @@ reshape_internal (parray *ret, parray *source, shape_type *shape,
   index_type shape_data[GFC_MAX_DIMENSIONS];
 
   rdim = GFC_DESCRIPTOR_EXTENT(shape,0);
+  GFC_ASSERT(rdim>0);
   if (rdim != GFC_DESCRIPTOR_RANK(ret))
     runtime_error("rank of return array incorrect in RESHAPE intrinsic");
 
@@ -158,6 +159,10 @@ reshape_internal (parray *ret, parray *source, shape_type *shape,
 
       source_extent = 1;
       sdim = GFC_DESCRIPTOR_RANK (source);
+      /* sdim is always > 0; this lets the compiler optimize more and
+         avoids a warning.  */
+      GFC_ASSERT(sdim>0);
+
       for (n = 0; n < sdim; n++)
 	{
 	  index_type se;
@@ -219,6 +224,7 @@ reshape_internal (parray *ret, parray *source, shape_type *shape,
     }
 
   sdim = GFC_DESCRIPTOR_RANK (source);
+  GFC_ASSERT(sdim>0);
   ssize = 1;
   sempty = 0;
   for (n = 0; n < sdim; n++)

-- 
Markus