Use of pointer casts in the interpreter

Sun Aug 8 23:14:00 GMT 1999

I've gotten to the point that Kawa now crashes in the Interpreter,
which led me to look at interpret.cc. There are a bunch of
pointer casts in there. Such casts are unsafe, as the gcc
optimizer makes aliasing assumptions based on the types of
pointers. In general, I think it is cleaner to use a union
anyway. I suggest:
typedef union {
 jint i;
 jfloat f;
 jint ia[1]; // Half of _Jv_word2.
 void* p;
 _Jv_Field *fld;
 ... 
} _Jv_word;
typedef union {
 jint ia[2];
 jlong l;
 jdouble d;
} _Jv_word2;
class _Jv_InterpMethodInvocation {
 _Jv_InterpMethod *running;
 _Jv_word *sp;
 ...
};
#define PUSHA(V) ((sp++)->p = (V))
#define PUSHI(V) ((sp++)->i = (V))
#define PUSHF(V) ((sp++)->f = (V))
#define PUSHL(V) ({ _Jv_word2 word2_tmp; \
 word2_tmp.l = (V); (sp++)->ia[0] = word2_tmp.ia[0]; (sp++)->ia[0] = word2_tmp.ia[1]; })
I also noticed a comment in _Jv_InterpMethod::continue1 about huge stack frames. That
suggests avoiding local variables in macros. Instead make word2_tmp a local
variable in _Jv_InterpMethod::continue1. (Admittedly, this is less clean.)
-- 
	--Per Bothner
bothner@pacbell.net per@bothner.com http://home.pacbell.net/bothner/