[RFA] Java DWARF fixes (was Re: [RFA] dwarf2_physname FINAL)

[Keith Seitz <keiths@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1259 bytes --]

On 03/25/2010 03:20 AM, Ulrich Weigand wrote:
>> I should think so... :-) If maintainers wish it so, it will be done.
>
> I'd appreciate that, thanks!

I've appended this to the attached patch.

>> No, it's not. I have a patch for this, but I've noticed some other
>> little java regressions (which are not tested by the test suite), so I
>> am going to delay submitting my patch until I can get the problems all
>> worked out.

My problems turned out to stem from another GCC debuginfo problem, 
described in gcc/43521 (out-of-line debuginfo for class methods does not 
mark "this" ptr with DW_AT_artificial).

I've worked around the issue in the attached patch, which should fix all 
the problems with Java that the dwarf2_physname patch introduced (or at 
least all the problems that anyone has yet noticed :-).

Ok?
Keith

ChangeLog
2010-03-25  Keith Seitz  <keiths@redhat.com>

	* dwarf2read.c (read_subroutine_type): If the compilation unit
	language is Java, mark any formal parameter named "this" as
	artificial (GCC/43521).
	(dwarf2_name): Add special handling for Java constructors.

testsuite/ChangeLog
2010-03-25  Keith Seitz  <keiths@redhat.com>

	* gdb.java/jprint.exp: XFAIL printing of static class members
	because of GCC debuginfo problem.


[-- Attachment #2: java-dwarf.patch --]
[-- Type: text/plain, Size: 2913 bytes --]

Index: dwarf2read.c
===================================================================
RCS file: /cvs/src/src/gdb/dwarf2read.c,v
retrieving revision 1.372
diff -u -p -r1.372 dwarf2read.c
--- dwarf2read.c	22 Mar 2010 13:21:39 -0000	1.372
+++ dwarf2read.c	25 Mar 2010 18:25:53 -0000
@@ -5932,7 +5932,18 @@ read_subroutine_type (struct die_info *d
 	      if (attr)
 		TYPE_FIELD_ARTIFICIAL (ftype, iparams) = DW_UNSND (attr);
 	      else
-		TYPE_FIELD_ARTIFICIAL (ftype, iparams) = 0;
+		{
+		  TYPE_FIELD_ARTIFICIAL (ftype, iparams) = 0;
+
+		  /* GCC/43521: In java, the formal parameter
+		     "this" is sometimes not marked with DW_AT_artificial.  */
+		  if (cu->language == language_java)
+		    {
+		      const char *name = dwarf2_name (child_die, cu);
+		      if (name && !strcmp (name, "this"))
+			TYPE_FIELD_ARTIFICIAL (ftype, iparams) = 1;
+		    }
+		}
 	      TYPE_FIELD_TYPE (ftype, iparams) = die_type (child_die, cu);
 	      iparams++;
 	    }
@@ -9159,6 +9170,34 @@ dwarf2_name (struct die_info *die, struc
       /* These tags always have simple identifiers already; no need
 	 to canonicalize them.  */
       return DW_STRING (attr);
+    case DW_TAG_subprogram:
+      /* Java constructors will all be named "<init>", so return
+	 the class name when we see this special case.  */
+      if (cu->language == language_java
+	  && DW_STRING (attr) != NULL
+	  && strcmp (DW_STRING (attr), "<init>") == 0)
+	{
+	  struct dwarf2_cu *spec_cu = cu;
+	  struct die_info *spec_die;
+
+	  /* GCJ will output '<init>' for Java constructor names.
+	     For this special case, return the name of the parent class.  */
+
+	  /* GCJ may output suprogram DIEs with AT_specification set.
+	     If so, use the name of the specified DIE.  */
+	  spec_die = die_specification (die, &spec_cu);
+	  if (spec_die != NULL)
+	    return dwarf2_name (spec_die, spec_cu);
+
+	  do
+	    {
+	      die = die->parent;
+	      if (die->tag == DW_TAG_class_type)
+		return dwarf2_name (die, cu);
+	    }
+	  while (die->tag != DW_TAG_compile_unit);
+	}
+      /* fall through */
     default:
       if (!DW_STRING_IS_CANONICAL (attr))
 	{
Index: testsuite/gdb.java/jprint.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.java/jprint.exp,v
retrieving revision 1.12
diff -u -p -r1.12 jprint.exp
--- testsuite/gdb.java/jprint.exp	9 Mar 2010 18:08:05 -0000	1.12
+++ testsuite/gdb.java/jprint.exp	25 Mar 2010 18:25:53 -0000
@@ -86,5 +86,8 @@ if ![set_lang_java] then {
     gdb_test "call x.addk(44)" "adding k gives 121\r\n.*= 121.*" "inherited virtual fn call"
 
     # Regression test for a crasher.
+    # GCC does not output location information for static class members,
+    # so GDB will report these as "optimized out". See gcc/43260.
+    setup_xfail *-*-* gcc/43260
     gdb_test "print *jprint.props" " = .*" "print a java.util.Properties"
 }