1 files changed, 38 insertions, 0 deletions
diff --git a/gcc/testsuite/g++.old-deja/g++.bugs/900227_01.C b/gcc/testsuite/g++.old-deja/g++.bugs/900227_01.C
new file mode 100644
index 00000000000..77ac1002893
--- /dev/null
+++ b/gcc/testsuite/g++.old-deja/g++.bugs/900227_01.C
@@ -0,0 +1,38 @@
+// g++ 1.37.1 bug 900227_01
+
+// g++ allows pointer type values to be converted to integral types which are
+// not actually large enough to hold the converted values.
+
+// Section 3.3.4 of the ANSI C standard says:
+
+//	A pointer may be converted to an integral type.  The size of the
+//	integer required and the results are implementation defined.  If
+//	the space provided is not long enough, the behavior is undefined.
+
+// I believe that the only proper thing to do in such cases is to generate
+// errors.  After all, if the converted value gets truncated, it is not
+// likely to be useful after that.
+
+// Furthermore, as the following example demonstrates, allowing pointers
+// to be converted to integral types which are not of sufficient size to
+// completely hold the converted values may cause additional troubles.
+
+// I tried the following code on 5 different machines and it failed on
+// all five (unless I also use the GNU assembler and the GNU linker).  Three
+// of the five (Sun3, Sun4, and Symmetry) got link-time errors about byte
+// offset overflows.  The other two (368/SystemV and AViiON) got assembly
+// time errors about relocatable names used in "constant" expressions.
+
+// keywords: casts, pointer types, integral types
+
+// Update 2/10/95: The compiler will now compute these expressions at
+// runtime.  I think this is in the spirit of the GNU compilers (jason).
+
+// Special g++ Options:
+
+int main ();
+
+short s = (short) &main;
+char c = (char) &main;
+
+int main () { return 0; }