Re: Four problems with Scheme 48 1.8

Michael Sperber <[email protected]> writes:

> Martin Ward <[email protected]> writes:
>
>> (1) Compiling with gcc version 4.4.0 gives an error on "make test":
>>
>> FAILURES:
>> Test case *3 [misc-big-tests] FAILED:
>>>From expression (* 47123 46039) EXPECTED value 2169495797 of 2169495797
>> INSTEAD got -2125471499
>>
>> '(#{Check-failure #{Test-case *3} (actual (* 47123 46039) -2125471499) 
>> (expected 2169495797 2169495797) #{Procedure 983 (equal? in 
>> scheme-level-1)}})
>>
>> Compiling with gcc 4.0.1 seems to work OK.
>
> This appears to be a bug in gcc 4.4, I'm afraid.  

Scratch that and substitute "interesting interpretation of the ANSI C
standard."  The attached patch (which is in the development code) works
around the problem.

-- 
Cheers =8-} Mike
Friede, Völkerverständigung und überhaupt blabla

changeset:   1289:ab92a31cec79
user:        Mike Sperber <[email protected]>
date:        Thu Jul 30 10:40:00 2009 +0200
summary:     Work around overflow problem in SMALL_MULTIPLY.

diff --git a/c/scheme48vm-prelude.h b/c/scheme48vm-prelude.h
--- a/c/scheme48vm-prelude.h
+++ b/c/scheme48vm-prelude.h
@@ -5,7 +5,15 @@
 #include "c-mods.h"
 #include "scheme48write-barrier.h"
 
-#define SMALL_MULTIPLY(x,y) ((x) * (y))
+/* 
+ * This is to tell the C compiler that the result may be negative even
+ * though the operands are unsigned.  It is still dubious: A cast from
+ * an unsigned to an int that involves an overflow pretty much has
+ * undefined result - the C standard does not give us two's complement
+ * arithmetic.
+ */
+
+#define SMALL_MULTIPLY(x,y) ((long)((unsigned long)(x) * (unsigned long)(y)))
 
 #define	NO_ERRORS	0		/* extension to errno.h */