This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA]: Java Inferior Call Take 3


Jeff Johnston <jjohnstn@redhat.com> writes:
> Here is the revised patch.
> 
> 2004-08-24  Jeff Johnston  <jjohnstn@redhat.com>
> 
>      * dwarf2read.c (typename_concat): Change prototype to accept dwarf2_cu
>      struct pointer as argument.  Change function to use new argument to
>      determine language.  If language is Java, use "." as separator,
>      otherwise, use "::".
>      (partial_die_parent_scope): Change comment to include java.  Check
>      language to determine if Java "." or C++ "::" separator should be used.
>      (add_partial_symbol): Enhance tests for C++ to also test for Java.
>      (guess_structure_name): Ditto.
>      (read_subroutine_type): Ditto.
>      (new_symbol): Ditto.
>      (read_structure_type): Add Java vtable support.
>      (read_namespace): Add Java support.
>      * jv-exp.y (FuncStart): New pattern.
>      (MethodInvocation): Add support for simple function calls.  Change
>      warning message for other forms of inferior call currently not
>      supported.
>      * valarith.c (value_subscript): Treat an array with upper-bound
>      of -1 as unknown size.

For the dwarf2read.c part of this patch:

There are a lot of places where we're selecting the name component
separator based on the language; I'd like compound name construction
abstracted out into its own function.  Two possible approaches:

- typename_concat could take, in addition to the cu, a pointer to an
  obstack.  When the obstack pointer is null, typename_concat would
  use xmalloc as it does now; otherwise it'd use obconcat.

- If you don't like overloading the behavior of typename_concat that
  way, you could define a new function altogether that takes a prefix,
  a name, a cu, and an obstack, and returns the name with the prefix
  properly attached, allocated in the obstack.

But once there's a function that does this, I think all the 'if java
then "." else "::"' can be neatened up quite a bit.


In this change:

-	    /* The semantics of C++ state that "struct foo { ... }" also
+	    /* The semantics of C++ and Java state that "struct foo { ... }" also

'struct foo { ... }' isn't valid Java; go ahead and say what you mean:

    /* The semantics of C++ state that "struct foo { ... }" also
       defines a typedef for "foo".  A Java class declaration also
       defines a typedef for the class.  Synthesize a typedef symbol
       so that "ptype foo" works as expected.  */


The new comment for typename_concat should explain what its 'cu'
argument is used for.

The vtable pointer recognition code is kind of weird.  The use of
'strlen (vptr_name) - 1' looks like a bug: don't we want to include
that last character in the comparison?  I've committed the patch
below; could you adapt your patch to apply on top of that?

2004-08-31  Jim Blandy  <jimb@redhat.com>

	* dwarf2read.c (is_vtable_name): New function, based on logic from
	read_structure_type, but passing the correct length to strncmp,
	and using 'sizeof' instead of 'strlen'.
	(read_structure_type): Call it.

Index: gdb/dwarf2read.c
===================================================================
RCS file: /cvs/src/src/gdb/dwarf2read.c,v
retrieving revision 1.159
diff -c -p -r1.159 dwarf2read.c
*** gdb/dwarf2read.c	29 Aug 2004 10:12:14 -0000	1.159
--- gdb/dwarf2read.c	31 Aug 2004 22:25:05 -0000
*************** dwarf2_attach_fn_fields_to_type (struct 
*** 3265,3270 ****
--- 3265,3287 ----
    TYPE_NFN_FIELDS_TOTAL (type) = total_length;
  }
  
+ 
+ /* Returns non-zero if NAME is the name of a vtable member in CU's
+    language, zero otherwise.  */
+ static int
+ is_vtable_name (const char *name, struct dwarf2_cu *cu)
+ {
+   static const char vptr[] = "_vptr";
+ 
+   /* C++ and some implementations of Java use this name.  */
+   if (strncmp (name, vptr, sizeof (vptr) - 1) == 0
+       && is_cplus_marker (name[sizeof (vptr) - 1]))
+     return 1;
+ 
+   return 0;
+ }
+ 
+ 
  /* Called when we find the DIE that starts a structure or union scope
     (definition) to process all dies that define the members of the
     structure or union.
*************** read_structure_type (struct die_info *di
*** 3403,3410 ****
  	      TYPE_VPTR_BASETYPE (type) = t;
  	      if (type == t)
  		{
- 		  static const char vptr_name[] =
- 		  {'_', 'v', 'p', 't', 'r', '\0'};
  		  int i;
  
  		  /* Our own class provides vtbl ptr.  */
--- 3420,3425 ----
*************** read_structure_type (struct die_info *di
*** 3414,3423 ****
  		    {
  		      char *fieldname = TYPE_FIELD_NAME (t, i);
  
! 		      if ((strncmp (fieldname, vptr_name,
!                                     strlen (vptr_name) - 1)
!                            == 0)
! 			  && is_cplus_marker (fieldname[strlen (vptr_name)]))
  			{
  			  TYPE_VPTR_FIELDNO (type) = i;
  			  break;
--- 3429,3435 ----
  		    {
  		      char *fieldname = TYPE_FIELD_NAME (t, i);
  
!                       if (is_vtable_name (fieldname, cu))
  			{
  			  TYPE_VPTR_FIELDNO (type) = i;
  			  break;


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]