Restore gnu.gcj.runtime.NameFinder.remove_unknown

Restore gnu.gcj.runtime.NameFinder.remove_unknown

[Andrew Haley]

This fixes a regression in stack traces generation whereby the system
property gnu.gcj.runtime.NameFinder.remove_unknown no longer works.
(Until it was removed, this system property allowed a user to print
all stack frames rather than just a subset that wouoldn't worry a Java
programmer.)

Why this is useful: I was debugging a very weird problem in gij when I
realized that the stack trace was misleading because so much
information had been removed.

Restoring this functionality turns this

Exception in thread "main" java.lang.NoClassDefFoundError: Foo
   at java.lang.Class.initializeClass (natClass.cc:690)
   at java.lang.Class.isAssignableFrom (Class.h:599)
   at Test.main (Test.java:10)
Caused by: java.lang.ClassNotFoundException: javax.management.ObjectName not found in gnu.gcj.runtime.SystemClassLoader{urls=[file:./], parent=gnu.gcj.runtime.ExtensionClassLoader{urls=[], parent=null}}
   at java.net.URLClassLoader.findClass (URLClassLoader.java:1080)
   at java.lang.ClassLoader.loadClass (ClassLoader.java:317)
   at java.lang.ClassLoader.loadClass (ClassLoader.java:260)
   at java.lang.Class.forName (natClass.cc:88)
   at java.lang.Class.initializeClass (natClass.cc:684)
   ...2 more

into 

Exception in thread "main" java.lang.NoClassDefFoundError: Foo
   at java.lang.Class.initializeClass (natClass.cc:690)
   at java.lang.Class.isAssignableFrom (Class.h:599)
   at ffi_call_unix64 (unix64.S:73)
   at ffi_call (ffi64.c:428)
   at ffi_java_raw_call (java_raw_api.c:295)
   at Test.main (Test.java:10)
   at  (java_raw_api.c:309)
   at ffi_closure_unix64_inner (ffi64.c:563)
   at ffi_closure_unix64 (unix64.S:228)
Caused by: java.lang.ClassNotFoundException: javax.management.ObjectName not found in gnu.gcj.runtime.SystemClassLoader{urls=[file:./], parent=gnu.gcj.runtime.ExtensionClassLoader{urls=[], parent=null}}
   at java.net.URLClassLoader.findClass (URLClassLoader.java:1080)
   at java.lang.ClassLoader.loadClass (ClassLoader.java:317)
   at java.lang.ClassLoader.loadClass (ClassLoader.java:260)
   at _Jv_FindClass(_Jv_Utf8Const*, java::lang::ClassLoader*) (natClassLoader.cc:291)
   at java.lang.Class.forName (natClass.cc:88)
   at _Jv_BytecodeVerifier::pop_init_ref(_Jv_BytecodeVerifier::type) (verify.cc:379)
   at _Jv_BytecodeVerifier::verify_instructions_0() (verify.cc:2895)
   at _Jv_VerifyMethod(_Jv_InterpMethod*) (verify.cc:3090)
   at _Jv_InterpreterEngine::do_verify(java::lang::Class*) (interpret.cc:3757)
   at _Jv_Linker::wait_for_state(java::lang::Class*, int) (link.cc:1770)
   at java.lang.Class.initializeClass (natClass.cc:684)
   ...8 more

and you can see at once the ClassNotFoundException in triggered by the verifier.

Andrew.


2006-02-16  Andrew Haley  <aph@redhat.com>

	* stacktrace.cc (GetStackTraceElements): Call
	gnu::gcj::runtime::NameFinder::removeUnknown() to determine if
	non-Java frames should be removed from a printed stack trace.
	Pass methodName to getLineNumberForFrame().
	(getLineNumberForFrame): Set method_name from info.dli_sname.
	* gnu/gcj/runtime/NameFinder.java (removeUnknown): New method.
	(remove_unknown): New variable.
	* include/java-stack.h (_Jv_StackTrace::getLineNumberForFrame):
	Add methodName arg.

Index: gnu/gcj/runtime/NameFinder.java
===================================================================
--- gnu/gcj/runtime/NameFinder.java	(revision 110794)
+++ gnu/gcj/runtime/NameFinder.java	(working copy)
@@ -34,6 +34,9 @@
  *     source file and line number info. Throwable.printStackTrace() will
  *     be faster if this property is set to 'false'.
  * </ul>
+ * <ul><code>gnu.gcj.runtime.NameFinder.remove_unknown</code>
+ *     Whether calls to unknown functions (class and method names are unknown)
+ *     should be removed from the stack trace. </ul>
  * </li>
  *
  * <code>close()</code> should be called to get rid of all resources.
@@ -57,6 +60,18 @@
                 ("gnu.gcj.runtime.NameFinder.use_addr2line", "true")
             ).booleanValue();
 
+  private static final boolean remove_unknown
+	  = Boolean.valueOf(System.getProperty
+		("gnu.gcj.runtime.NameFinder.remove_unknown", "true")
+	    ).booleanValue();
+
+  // Return true if non-Java frames should be removed from stack
+  // traces.
+  static final boolean removeUnknown()
+  {
+    return remove_unknown;
+  }
+
   class Addr2Line
   {
     Process proc;
Index: stacktrace.cc
===================================================================
--- stacktrace.cc	(revision 110794)
+++ stacktrace.cc	(working copy)
@@ -171,7 +171,8 @@
 
 void
 _Jv_StackTrace::getLineNumberForFrame(_Jv_StackFrame *frame, NameFinder *finder, 
-		 jstring *sourceFileName, jint *lineNum)
+				      jstring *sourceFileName, jint *lineNum,
+				      jstring *methodName)
 {
 #ifdef INTERPRETER
   if (frame->type == frame_interpreter)
@@ -200,6 +201,9 @@
       else
         return;
 
+      if (*methodName == NULL && info.dli_sname)
+	*methodName = JvNewStringUTF (info.dli_sname);
+
       // addr2line expects relative addresses for shared libraries.
       if (strcmp (info.dli_fname, argv0) == 0)
         offset = (_Unwind_Ptr) ip;
@@ -323,16 +327,22 @@
 	end_idx = i - 1;
     }
   
+  const jboolean remove_unknown 
+    = gnu::gcj::runtime::NameFinder::removeUnknown();
+
   // Second pass: Look up line-number info for remaining frames.
   for (int i = start_idx; i <= end_idx; i++)
     {
       _Jv_StackFrame *frame = &trace->frames[i];
       
-      if (frame->klass == NULL)
-        // Not a Java frame.
+      if (frame->klass == NULL && remove_unknown)
+	// Not a Java frame.
 	continue;
-      
-      jstring className = frame->klass->getName ();
+
+      jstring className = NULL;
+      if (frame->klass != NULL)
+	className = frame->klass->getName ();
+
       jstring methodName = NULL;
       if (frame->meth)
         methodName = JvNewStringUTF (frame->meth->name->chars());
@@ -340,7 +350,8 @@
       jstring sourceFileName = NULL;
       jint lineNum = -1;
       
-      getLineNumberForFrame(frame, finder, &sourceFileName, &lineNum);
+      getLineNumberForFrame(frame, finder, &sourceFileName, &lineNum, 
+			    &methodName);
       
       StackTraceElement *element = new StackTraceElement (sourceFileName, lineNum,
         className, methodName, 0);
Index: include/java-stack.h
===================================================================
--- include/java-stack.h	(revision 110794)
+++ include/java-stack.h	(working copy)
@@ -105,7 +105,8 @@
   static jclass ClassForFrame (_Jv_StackFrame *frame);
   static void FillInFrameInfo (_Jv_StackFrame *frame);
   static void getLineNumberForFrame(_Jv_StackFrame *frame, NameFinder *finder, 
-			     jstring *sourceFileName, jint *lineNum);
+				    jstring *sourceFileName, jint *lineNum,
+				    jstring *methodName);
   
   static _Unwind_Reason_Code UnwindTraceFn (struct _Unwind_Context *context, 
     void *state_ptr);

Re: Restore gnu.gcj.runtime.NameFinder.remove_unknown

[Michael Koch]

On Thu, Feb 16, 2006 at 12:24:21PM +0000, Andrew Haley wrote:
> +  private static final boolean remove_unknown
> +	  = Boolean.valueOf(System.getProperty
> +		("gnu.gcj.runtime.NameFinder.remove_unknown", "true")
> +	    ).booleanValue();

Shouldnt this use SystemProperties.getProperty() to circumvent the
security check? Just an idea...


Cheers,
Michael
-- 
Escape the Java Trap with GNU Classpath!
http://www.gnu.org/philosophy/java-trap.html

oin the community at http://planet.classpath.org/

Re: Restore gnu.gcj.runtime.NameFinder.remove_unknown

[Andrew Haley]

Michael Koch writes:
 > On Thu, Feb 16, 2006 at 12:24:21PM +0000, Andrew Haley wrote:
 > > +  private static final boolean remove_unknown
 > > +	  = Boolean.valueOf(System.getProperty
 > > +		("gnu.gcj.runtime.NameFinder.remove_unknown", "true")
 > > +	    ).booleanValue();
 > 
 > Shouldnt this use SystemProperties.getProperty() to circumvent the
 > security check? Just an idea...

I guess so.  There's also use_addr2line in the same class.  

Andrew.