[Bug libgcj/17021] libgcj verifier resolves classes too eagerly

["thebohemian at gmx dot net" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #11 from thebohemian at gmx dot net  2005-10-19 16:14 -------
In the JavaUtils class of axis 1.2 RC2 contains the following instruction:

639: invokevirtual <Method javax.activation.DataHandler.getContent
()java.lang.Object>

With a missing DataHandler class file this JavaUtils class can be verified
successfully on sun' s jvm but fails at this instruction on gij.

I traced the problem back to a call to in 

_Jv_BytecodeVerifier::ref_intersection::compatible(_Jv_BytecodeVerifier::ref_intersection*,
_Jv_BytecodeVerifier*)

In that method the names of the two classes (which are not resolved) are the
following:

(gdb) putf self->data.name
$152 = 0x121054 "javax.activation.DataHandler"
(gdb) putf other_iter->data.name
$153 = 0x80736ac "Ljavax.activation.DataHandler;"

Obviously this denotes the same classnames but for Jv_equalUtf8Consts which is
called in _Jv_BytecodeVerifier::ref_intersection::compatible() the strings are
different.

I propose to fix the problem in the following way: Instead of
if (! self->is_resolved
  && ! other_iter->is_resolved
  && _Jv_equalUtf8Consts (self->data.name,
  other_iter->data.name))

do

if (! self->is_resolved
  && ! other_iter->is_resolved
  && ( _Jv_equalUtf8Consts (self->data.name, other_iter->data.name) ||
match_classname(self->data.name, other_iter->data.name ) )

match_classname is then implemented in a way that it would accept all these
names as equal:
java/lang/Class
Ljava/lang/Class;
java.lang.Class
Ljava.lang.Class;

However there might be another approach to the problem: One could make sure
that class names in _Jv_UTF8_Const objects are always in the
Lp1/p2/p3/classname; form.

Please comment.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17021