work around compiler bug on CX/UX (perl5.004_01)

This patch provides a work-around for a compiler bug on CX/UX systems (which
shows up as a failure in the 'w' format of pack). The
CXUX_BROKEN_CONSTANT_CONVERT ifdef flag is added to the hints/cxux.sh
compiler and pp.c is modified to avoid a compile time constant conversion
which fails based on that ifdef.

While I was in the hints file, I also added the magical
-Qtarget=M88110compat compiler option which makes it build code that will
run on both 88110 and 88100 CX/UX machines interchangably.

This patch was generated from a brand new copy of perl5.004_01, so I'm
confident there are no extraneous changes that slipped in. I even built
and tested and it passed all tests.

(I decided to go with option #3 in my previous mail about how to do the patch).

If its too late for 5.004_02, I wouldn't worry - it isn't very critical.


p5p-msgid: 9707301934.AA18594@amber.ssd.hcsc.com

1 files changed, 15 insertions, 0 deletions

diff --git a/pp.c b/pp.c
index d791c4b456..a0b337a6e8 100644
--- a/pp.c
+++ b/pp.c
@@ -16,6 +16,17 @@
 #include "perl.h"
 
 /*
+ * The compiler on Concurrent CX/UX systems has a subtle bug which only
+ * seems to show up when compiling pp.c - it generates the wrong double
+ * precision constant value for (double)UV_MAX when used inline in the body
+ * of the code below, so this makes a static variable up front (which the
+ * compiler seems to get correct) and uses it in place of UV_MAX below.
+ */
+#ifdef CXUX_BROKEN_CONSTANT_CONVERT
+static double UV_MAX_cxux = ((double)UV_MAX);
+#endif
+
+/*
  * Types used in bitwise operations.
  *
  * Normally we'd just use IV and UV.  However, some hardware and
@@ -3714,8 +3725,12 @@ PP(pp_pack)
 #ifdef BW_BITS
 		    adouble <= BW_MASK
 #else
+#ifdef CXUX_BROKEN_CONSTANT_CONVERT
+		    adouble <= UV_MAX_cxux
+#else
 		    adouble <= UV_MAX
 #endif
+#endif
 		    )
 		{
 		    char   buf[1 + sizeof(UV)];