fix some 32/64-bit compiler warnings

Some bits of code don't do well on a 32-bit system with 64-bit ints
(-Duse64bitint)

In particular:

_MEM_WRAP_NEEDS_RUNTIME_CHECK:
    if sizeof(MEM_SIZE) > sizeof(n), then the shift count could be
        negative

S_regmatch:
    ln and n were two different sizes and signesses, so comparing them
    warned. Since they were being mis-used as two convenient temporary
    booleans anyway, just use temporary booleans instead.

Perl_sv_vcatpvfn_flags:
    the test/assertion (IV)elen < 0 was (I think) being used to test for
    signed/unsigned conversion wrap-around. elen is of type STRLEN which
    is a pointer-based type, so can be 32-bit while IV is 64-bit. Instead
    compare it to half the maximum value of a STRLEN var to see if it may
    have wrapped.

1 files changed, 4 insertions, 1 deletions

diff --git a/handy.h b/handy.h
index 031850492c..ffb8e2fc3d 100644
--- a/handy.h
+++ b/handy.h
@@ -1917,10 +1917,13 @@ PoisonWith(0xEF) for catching access to freed memory.
  * As well as avoiding the need for a run-time check in some cases, it's
  * designed to avoid compiler warnings like:
  *     comparison is always false due to limited range of data type
+ * It's mathematically equivalent to
+ *    max(n) * sizeof(t) > MEM_SIZE_MAX
  */
 
 #  define _MEM_WRAP_NEEDS_RUNTIME_CHECK(n,t) \
-    (sizeof(t) > ((MEM_SIZE)1 << 8*(sizeof(MEM_SIZE) - sizeof(n))))
+    (  sizeof(MEM_SIZE) < sizeof(n) \
+    || sizeof(t) > ((MEM_SIZE)1 << 8*(sizeof(MEM_SIZE) - sizeof(n))))
 
 /* This is written in a slightly odd way to avoid various spurious
  * compiler warnings. We *want* to write the expression as