[RFA] gcj PR debug/40990

[Dodji Seketeli]

Le 10/08/2009 19:10, Dave Korn a écrit :
> Dodji Seketeli wrote:
>> Le 10/08/2009 18:03, Andrew Haley a écrit :
>>> Andrew Haley wrote:
>>>> Dodji Seketeli wrote:
>>>>> I noticed that gcj was emitting some "unnamed" methods for the class
>>>>> java.lang.String. I am not sure what those methods are. Maybe the
>>>>> contructors ? I am talking about the methods represented the DIE 0x88, for
>>>>> instance. In any case, before this patch, gcj was emitting the string "()"
>>>>> as method name, in the .debug_pubnames section. Now with the patch, it just
>>>>> emits an empty string. This is weird, but I think the patch doesn't make
>>>>> the situation worse.
>>>> I don't know.  I'll have a look.
>>> It's quite explicit:
>>>
>>> 	  if (! DECL_CONSTRUCTOR_P (node))
>>> 	    put_decl_node (DECL_NAME (node));
>>>
>>> So, constructors don't have names.  What does C++ do for constructors?
>> In c++, for a class named "s", the name of the constructor in the dwarf
>> output would be the string "s".
> 
>   Are you sure?  It looks to me like it uses the fully-qualified name "s::s"
> 
> $ cat s.cxx
> 
> #include <stdlib.h>
> #include <string.h>
> 
> class s
> {
>   int foo;
>   char *bar;
>   s(const char *);
> };
> 
> s::s(const char *x)
> {
>   foo = strlen (x);
>   bar = (char *) malloc (foo);
>   strcpy (bar, x);
> }
> 
> $ g++-4 -S s.cxx -g -O0  -o s.s
> 
> $ grep -A20 debug_pubname s.s
>         .section        .debug_pubnames,"dr"
>         .long   0x20
>         .word   0x2
>         .secrel32       Ldebug_info0
>         .long   0x1c1
>         .long   0x178
>         .ascii "s::s\0"
>         .long   0x19e
>         .ascii "s::s\0"
>         .long   0x0
>         .section        .debug_pubtypes,"dr"
> 

Ah, for the pubname section, you are correct yes. I was talking about the
DW_AT_name attribute of the constructor DIE in the .debug_info section. Not
about constructor name in the .debug_pubnames section.

-- 
Dodji Seketeli
Red Hat