[rfc] [0/8] Get rid of current_gdbarch

[rfc] [0/8] Get rid of current_gdbarch

[Markus Deuling]

Hi,

this is the next series of 8 mostly mechnical patches to get rid of of some trivial to replace occurences of current_gdbarch. 

Tested by running testuite no x86 and gdb_mbuild. Is this ok?


-- 
Markus Deuling
GNU Toolchain for Linux on Cell BE
deuling@de.ibm.com

Re: [rfc] [0/8] Get rid of current_gdbarch

[Ulrich Weigand]

Markus Deuling wrote:

> this is the next series of 8 mostly mechnical patches to get rid of of some trivial to replace occurences of current_gdbarch. 

I've checked in all of those except:

- the xtensa-tdep.c patch (to give Maxim a chance to comment), and
- the stack.c patch, because of this:

@@ -1854,10 +1856,11 @@ If you continue, the return value that y
   if (return_value != NULL)
     {
       struct type *return_type = value_type (return_value);
-      gdb_assert (gdbarch_return_value (current_gdbarch, return_type,
+      gdb_assert (gdbarch_return_value (get_frame_arch (get_current_frame ()),
+					return_type,
 					NULL, NULL, NULL)
 		  == RETURN_VALUE_REGISTER_CONVENTION);
-      gdbarch_return_value (current_gdbarch, return_type,
+      gdbarch_return_value (get_frame_arch (get_current_frame ()), return_type,
 			    get_current_regcache (), NULL /*read*/,
 			    value_contents (return_value) /*write*/);
     }

At this point, there's no need to involve get_current_frame.  The
routine operates on the current regcache, and should simply consult
its architecture.

Thanks,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [rfc] [0/8] Get rid of current_gdbarch

[Daniel Jacobowitz]

On Tue, Oct 09, 2007 at 08:22:45PM +0200, Ulrich Weigand wrote:
> At this point, there's no need to involve get_current_frame.  The
> routine operates on the current regcache, and should simply consult
> its architecture.

As a more general note, replacing one global state with another is of
limited value.  It's not completely useless, but let's avoid it when
we can :-)

-- 
Daniel Jacobowitz
CodeSourcery

Re: [rfc] [0/8] Get rid of current_gdbarch

[Ulrich Weigand]

Daniel Jacobowitz wrote:
> On Tue, Oct 09, 2007 at 08:22:45PM +0200, Ulrich Weigand wrote:
> > At this point, there's no need to involve get_current_frame.  The
> > routine operates on the current regcache, and should simply consult
> > its architecture.
> 
> As a more general note, replacing one global state with another is of
> limited value.  It's not completely useless, but let's avoid it when
> we can :-)

Well, in some cases you have to have some global state ...  At this
point, which is the implementation of the "return" command, you have
no other context -- the command by design operates on the "current"
register set ...  (That still doesn't mean it should refer to
current_gdbarch at this point, that makes a lot more assumptions.)

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [rfc] [0/8] Get rid of current_gdbarch

[Maxim Grigoriev]

>> I've checked in all of those except:
>>
>> - the xtensa-tdep.c patch (to give Maxim a chance to comment), and

I've tested xtensa-tdep.* patch. It works fine.

Getting rid of defines looks somewhat less elegant. But, I think it's 
been done for a reason.

Please, go ahead and check it in.

Thanks much for doing this.

-- Maxim


Ulrich Weigand wrote:
> Markus Deuling wrote:
>
>   
>> this is the next series of 8 mostly mechnical patches to get rid of of some trivial to replace occurences of current_gdbarch. 
>>     
>
> I've checked in all of those except:
>
> - the xtensa-tdep.c patch (to give Maxim a chance to comment), and
> - the stack.c patch, because of this:
>
> @@ -1854,10 +1856,11 @@ If you continue, the return value that y
>    if (return_value != NULL)
>      {
>        struct type *return_type = value_type (return_value);
> -      gdb_assert (gdbarch_return_value (current_gdbarch, return_type,
> +      gdb_assert (gdbarch_return_value (get_frame_arch (get_current_frame ()),
> +					return_type,
>  					NULL, NULL, NULL)
>  		  == RETURN_VALUE_REGISTER_CONVENTION);
> -      gdbarch_return_value (current_gdbarch, return_type,
> +      gdbarch_return_value (get_frame_arch (get_current_frame ()), return_type,
>  			    get_current_regcache (), NULL /*read*/,
>  			    value_contents (return_value) /*write*/);
>      }
>
> At this point, there's no need to involve get_current_frame.  The
> routine operates on the current regcache, and should simply consult
> its architecture.
>
> Thanks,
> Ulrich
>
>   

Re: [rfc] [0/8] Get rid of current_gdbarch

[Ulrich Weigand]

Maxim Grigoriev wrote:

> >> I've checked in all of those except:
> >>
> >> - the xtensa-tdep.c patch (to give Maxim a chance to comment), and
> 
> I've tested xtensa-tdep.* patch. It works fine.
> 
> Getting rid of defines looks somewhat less elegant. But, I think it's 
> been done for a reason.
> 
> Please, go ahead and check it in.
Done.

> Thanks much for doing this.

Thanks for testing!

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [rfc] [0/8] Get rid of current_gdbarch

[Markus Deuling]

Ulrich Weigand schrieb:
> Maxim Grigoriev wrote:
> 
>>>> I've checked in all of those except:
>>>>
>>>> - the xtensa-tdep.c patch (to give Maxim a chance to comment), and
>> I've tested xtensa-tdep.* patch. It works fine.
>>
>> Getting rid of defines looks somewhat less elegant. But, I think it's 
>> been done for a reason.
>>
>> Please, go ahead and check it in.
> Done.
> 
>> Thanks much for doing this.
> 
> Thanks for testing!
> 

Thanks Maxim for testing and Uli for checking in !



-- 
Markus Deuling
GNU Toolchain for Linux on Cell BE
deuling@de.ibm.com

Re: [rfc] [0/8] Get rid of current_gdbarch

[Markus Deuling]

Ulrich Weigand schrieb:
> 
> @@ -1854,10 +1856,11 @@ If you continue, the return value that y
>    if (return_value != NULL)
>      {
>        struct type *return_type = value_type (return_value);
> -      gdb_assert (gdbarch_return_value (current_gdbarch, return_type,
> +      gdb_assert (gdbarch_return_value (get_frame_arch (get_current_frame ()),
> +					return_type,
>  					NULL, NULL, NULL)
>  		  == RETURN_VALUE_REGISTER_CONVENTION);
> -      gdbarch_return_value (current_gdbarch, return_type,
> +      gdbarch_return_value (get_frame_arch (get_current_frame ()), return_type,
>  			    get_current_regcache (), NULL /*read*/,
>  			    value_contents (return_value) /*write*/);
>      }
> 
> At this point, there's no need to involve get_current_frame.  The
> routine operates on the current regcache, and should simply consult
> its architecture.

Hm, ok, thank you very much. I'll come up with another patch.


-- 
Markus Deuling
GNU Toolchain for Linux on Cell BE
deuling@de.ibm.com