Re: [open-cobol-list] Re: possibly bug: Compare: numeric (USAGE COMP ...) with alphanumeric

[Keisuke Nishida]

At Mon, 17 May 2004 16:33:42 +0200,
Thomas Biehler wrote:
> 
> But i believe that allowing the "critical compare":
>   "NON - USAGE DISPLAY" numeric  and  alphanumeric
> is NOT ANSI 1985 or ISO 2002  conform! 

Oh, you are right.  We need another config option here..

> So far i unterstand the compiler-source, the error reporting here is not 
> trivial to implement.

Actually this is trivial.  Try this patch:

--- typeck.c.~1.84.~    2004-05-15 19:11:02.000000000 +0900
+++ typeck.c    2004-05-18 22:28:08.000000000 +0900
@@ -1235,6 +1235,18 @@
              }
            else
              {
+               if ((CB_TREE_CLASS (p->x) == CB_CLASS_NUMERIC
+                    && CB_REFERENCE_P (p->x)
+                    && CB_FIELD (cb_ref (p->x))->usage != CB_USAGE_DISPLAY
+                    && (CB_TREE_CLASS (p->y) == CB_CLASS_ALPHABETIC
+                        || CB_TREE_CLASS (p->y) == CB_CLASS_ALPHANUMERIC))
+                   || (CB_TREE_CLASS (p->y) == CB_CLASS_NUMERIC
+                       && CB_REFERENCE_P (p->y)
+                       && CB_FIELD (cb_ref (p->y))->usage != CB_USAGE_DISPLAY
+                       && (CB_TREE_CLASS (p->x) == CB_CLASS_ALPHABETIC
+                           || CB_TREE_CLASS (p->x) == CB_CLASS_ALPHANUMERIC)))
+                 cb_error_x (p->x, "nonnumeric item cannot be compared with "
+                             "numeric item of usage non-DISPLAY");
                if (cb_flag_runtime_inlining)
                  {
                    if (CB_LITERAL_P (p->y))

Currently there is no error check for comparison.
I'll implement better checks and test suite later.

Keisuke