Re: [abc] Jastadd compile bugs

Neil Ongkingco wrote:

> Attached are some jastadd bugs found when compiling using Jigsaw.
> Perhaps they will help in deciding whether we should switch. All
> examples compile fine in both javac and abc. I'm still just halfway
> through jastadd's error dump, so I might add some more bugs later today.
>
Some remarks on the test cases below.

>class CharByteTest {
> public static void main(String args[]) {
> byte a = 100;
> System.out.println("Hello world");
> //Jastadd does not properly recognize the cast, and insists
> //that the type of the expression is still a char
> a = (byte) (a - 'a');
> }
>}
>
>
The comment about the problem in JastAdd is misleading. It's not the
cast that confuses it; rather, it tries to add the byte a to the char
'a' without performing binary numeric promotion
(http://java.sun.com/docs/books/jls/second_edition/html/conversions.doc.html#170983),
i.e. without converting both to int first. Thus it ends up with
mismatched types on the '-' operation, not on the assignment back to a.
Casting the first operand to char or the second to byte ensures
compilability with jastadd. It is interesting that if one changes the
expression to "a = (char) a - 'a';", then jastadd complains that it
can't assign int to byte, i.e. it does do some promotion of its
arguments, but seemingly too late.

>import java.io.*;
>
>class FinallyTest {
> //Jastadd gives a false error report if more than one exception
> //is specifified in the signature. Removing InterruptedException
> //allows Jastadd to compile properly
> public static void a() throws IOException, InterruptedException {
> throw new IOException("FinallyTest.a");
> }
>
> public static void main(String args[]) throws Exception{
> try {
> a();
> } finally {
> try {
> a();
> } catch (IOException e) {}
> }
> }
>}
>
>
This is indeed rather strange -- the error JastAdd reports is that the
body of the catch block is unreachable, and indeed this is reported
whenever a() is declared to throw more than one exception, whether or
not it is declared to throw IOException. No idea about this one.

>class ClassTest {
> static {
> //Jastadd doesn't lookup names properly in static blocks
> Class c = ClassTest.class;
> }
> public static void main(String args[]) {
> }
>}
>
>
Again, the comment about the problem is misleading. The problem here is
in JastAdd-generated code. The exact error message is "*** Semantic
Error: variable class$ClassTest is used in static { Class c =
class$ClassTest != null ? class$ClassTest : class$ClassTest =
class$("ClassTest"); } before it is declared". Interestingly if one
drops the static block, this works correctly -- perhaps the (generated)
class$ClassTest variable isn't static itself?

>class ClassTest2 {
> //Jastadd also has lookup problems for static variables
> static Class cn = ClassTest2.class;
> public static void main(String args[]) {
> }
>}
>
>
This is identical to above (in fact, though I may be wrong, I think the
bottom form is syntactic sugar for the top). It certainly produces the
same error message.

The problem me and Julian have been talking about is demonstrated by
this simple test case:

public class Test {
    public static void main(String[] args) {
        StringBuffer b = new StringBuffer();
        b.append('-');
    }
}

When compiled with a Java 1.4 JVM, this works fine. Under a Java5 JVM,
JastAdd can't compile it -- it complains about an uncaught IOException
on the call to append(). The reason is that the Java 5 API introduces an
Appendable interface with a method 'Appendable append(char c) throws
IOException', and StringBuffer implements Appendable. The method lookup
ends up returning the method from the interface rather than the
StringBuffer class, and so exception behaviour is considered incorrect.
This is actually quite nasty, because a lot of String manipulations get
compiled down to StringBuffer actions, so this prevents the compilation
of many many programs in a Java5 environment.

I'll summarise any other problems in this thread; we should send it to
Torbjorn for comments/fixes when we've finalised it.

- P
Received on Thu Jan 12 14:24:35 2006