Re: proposed new module intprops-test

[Paul Eggert]

> With just the first 2 out of 3 commits (partial revert and work around IRIX 
> 6.5
> cc bug), IRIX 6.5 cc would still have complained ("The indicated constant 
> value
> is not known.") about a combination of signed and unsigned integers, this time
> with '|', not '+' ... and HP-UX cc would still complain

Thanks for checking that.  These complaints are still worrisome,
because INT_ADD_OVERFLOW etc. can't be used as integer constant
expressions on those platforms.  But the diagnostics you're sending
suggest that a new intprops.h implementation, which I've just pushed,
might fix things for those two platforms.  Could you please give the
latest gnulib a try, with the following quick change?

--- a/tests/test-intprops.c
+++ b/tests/test-intprops.c
@@ -31,7 +31,7 @@
    These tests should be checkable via 'verify' rather than 'ASSERT', but
    using 'verify' would run into a bug with HP-UX 11.23 cc; see
    <http://lists.gnu.org/archive/html/bug-gnulib/2011-05/msg00401.html>.  */
-#if __GNUC__ || __SUNPRO_C
+#if __GNUC__ || __SUNPRO_C || 1
 # define VERIFY(x) do { verify (x); } while (0)
 #else
 # define VERIFY(x) ASSERT (x)



Or, if the above doesn't work, could you also try it after replacing
lib/intprop.h's two instances of

  (b) - (b) + (a)

with

  (b) - (b) + (a) - (a)

or with

  (b) - (b) + ((a) - (a))

?  Thanks again.


Here's the intprops.h patch I just pushed:

intprops: remove assumption about A|B representation
* lib/intprops.h (_GL_BINARY_OP_OVERFLOW): Do not assume that A|B
is a valid integer if both A and B are.  Although this is true for
all known practical hosts, the C standard doesn't guarantee it,
and the code need not assume it.  Also, this change may work around
HP-UX 11.23 and IRIX 6.5 cc bugs reported by Bruno Haible in
<http://lists.gnu.org/archive/html/bug-gnulib/2011-05/msg00426.html>.
diff --git a/lib/intprops.h b/lib/intprops.h
index bd9f040..293204a 100644
--- a/lib/intprops.h
+++ b/lib/intprops.h
@@ -311,13 +311,10 @@
 /* Return 1 if the expression A <op> B would overflow,
    where OP_RESULT_OVERFLOW (A, B, MIN, MAX) does the actual test,
    assuming MIN and MAX are the minimum and maximum for the result type.
-
-   This macro assumes that A | B is a valid integer if both A and B are,
-   which is true of all known practical hosts.  If this is a problem
-   for you, please let us know how to fix it for your host.  */
+   Arguments should be free of side effects.  */
 #define _GL_BINARY_OP_OVERFLOW(a, b, op_result_overflow)        \
   op_result_overflow (a, b,                                     \
-                      _GL_INT_MINIMUM ((a) | (b)),              \
-                      _GL_INT_MAXIMUM ((a) | (b)))
+                      _GL_INT_MINIMUM ((b) - (b) + (a)),        \
+                      _GL_INT_MAXIMUM ((b) - (b) + (a)))

 #endif /* _GL_INTPROPS_H */