question about method naming

question about method naming

[Tom Tromey]

I compiled the appended code with `gcj -g'.  I'm using the trunk cvs
gcj.  Then I ran gdb on the program.

If I type:

    b x.main

gdb tells me that there is no such method.

I tracked the problem down to linespec.c:find_methods().

Here we end up in the loop, and method_name for the 0th method is:

    (gdb) p method_name
    $27 = 0x83e9a52 "x.main(java.lang.String[])"

So of course we fail to match this against "main".

Is this a gdb bug or a gcj bug?  Or just a disagreement?

I'm guessing that method_name should not be fully qualified.
Is that correct?

Tom

public class x
{
  public static void main (String[] args)
  {
    char c = args[0].charAt(0);
    System.out.println(c);
  }
}

Re: question about method naming

[Daniel Jacobowitz]

On Mon, May 06, 2002 at 11:53:52AM -0600, Tom Tromey wrote:
> I compiled the appended code with `gcj -g'.  I'm using the trunk cvs
> gcj.  Then I ran gdb on the program.
> 
> If I type:
> 
>     b x.main
> 
> gdb tells me that there is no such method.

Are you working on a very current gcj?  I.E. one that defaults to
DWARF-2 output?  I observe the same thing in 3.0.4 with -gdwarf-2, but
not with -gstabs+.

In this version, the stabs debug info claims that 'class x' has a
member function 'main'.  In DWARF-2 it claims that 'class x' has a
member function 'x.main(java.lang.String[])'.  If this is still the
case in your compiler (check readelf -w output, search for a
DW_TAG_subprogram whose name contains main) then I believe it is
a gcj bug.

 <2><16d>: Abbrev Number: 4 (DW_TAG_subprogram)
     DW_AT_sibling     : <1c4>  
     DW_AT_external    : 1      
     DW_AT_name        : x.main(java.lang.String[])     
     DW_AT_decl_file   : 1      
     DW_AT_decl_line   : 458756 
     DW_AT_MIPS_linkage_name: _ZN1x4mainEP6JArrayIPN4java4lang6StringEE 
     DW_AT_declaration : 1      


Also amusing:

 <2><1c4>: Abbrev Number: 6 (DW_TAG_subprogram)
     DW_AT_external    : 1
     DW_AT_name        : ()
     DW_AT_MIPS_linkage_name: _ZN1xC1Ev
     DW_AT_artificial  : 1
     DW_AT_declaration : 1

The constructor's name is obviously wrong...

> 
> I tracked the problem down to linespec.c:find_methods().
> 
> Here we end up in the loop, and method_name for the 0th method is:
> 
>     (gdb) p method_name
>     $27 = 0x83e9a52 "x.main(java.lang.String[])"
> 
> So of course we fail to match this against "main".
> 
> Is this a gdb bug or a gcj bug?  Or just a disagreement?
> 
> I'm guessing that method_name should not be fully qualified.
> Is that correct?

I believe so.

> 
> Tom
> 
> public class x
> {
>   public static void main (String[] args)
>   {
>     char c = args[0].charAt(0);
>     System.out.println(c);
>   }
> }
> 

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: question about method naming

[Tom Tromey]

>>>>> "Daniel" == Daniel Jacobowitz <drow@mvista.com> writes:

Tom> If I type:
Tom> b x.main
Tom> gdb tells me that there is no such method.

Daniel> Are you working on a very current gcj?  I.E. one that defaults
Daniel> to DWARF-2 output?  I observe the same thing in 3.0.4 with
Daniel> -gdwarf-2, but not with -gstabs+.

Yes.  I'm using the trunk cvs gcj.

Daniel> In this version, the stabs debug info claims that 'class x'
Daniel> has a member function 'main'.  In DWARF-2 it claims that
Daniel> 'class x' has a member function 'x.main(java.lang.String[])'.
Daniel> If this is still the case in your compiler (check readelf -w
Daniel> output, search for a DW_TAG_subprogram whose name contains
Daniel> main) then I believe it is a gcj bug.

Ok.  What is the correct output?  `main(java.lang.String[])'?

I wonder if this was done because of the lack of namespace support in
gdb?  In this case that would be strange (`x' is a class, not a
namespace).  This is something I'll note in the bug report.

Daniel> Also amusing:
Daniel>  <2><1c4>: Abbrev Number: 6 (DW_TAG_subprogram)
Daniel>      DW_AT_external    : 1
Daniel>      DW_AT_name        : ()
Daniel>      DW_AT_MIPS_linkage_name: _ZN1xC1Ev
Daniel>      DW_AT_artificial  : 1
Daniel>      DW_AT_declaration : 1
Daniel> The constructor's name is obviously wrong...

I saw this too.

Thanks for your insight,
Tom

Re: question about method naming

[Daniel Jacobowitz]

On Mon, May 06, 2002 at 05:03:19PM -0600, Tom Tromey wrote:
> >>>>> "Daniel" == Daniel Jacobowitz <drow@mvista.com> writes:
> 
> Tom> If I type:
> Tom> b x.main
> Tom> gdb tells me that there is no such method.
> 
> Daniel> Are you working on a very current gcj?  I.E. one that defaults
> Daniel> to DWARF-2 output?  I observe the same thing in 3.0.4 with
> Daniel> -gdwarf-2, but not with -gstabs+.
> 
> Yes.  I'm using the trunk cvs gcj.
> 
> Daniel> In this version, the stabs debug info claims that 'class x'
> Daniel> has a member function 'main'.  In DWARF-2 it claims that
> Daniel> 'class x' has a member function 'x.main(java.lang.String[])'.
> Daniel> If this is still the case in your compiler (check readelf -w
> Daniel> output, search for a DW_TAG_subprogram whose name contains
> Daniel> main) then I believe it is a gcj bug.
> 
> Ok.  What is the correct output?  `main(java.lang.String[])'?

No, `main'.  The type should not be there at all.  Breaking on
`x.main(java.lang.String[])' or something similar should still work,
because the mangled name is in scope...

> I wonder if this was done because of the lack of namespace support in
> gdb?  In this case that would be strange (`x' is a class, not a
> namespace).  This is something I'll note in the bug report.

That may be why.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer