binds_local question

binds_local question

[Mike Stump]

In varasm.c we have:

   /* A variable is local if the user explicitly tells us so.  */
   else if (DECL_VISIBILITY_SPECIFIED (exp) && DECL_VISIBILITY (exp) != 
VISIBILITY_DEFAULT)
     local_p = true;
   /* Otherwise, variables defined outside this object may not be local. 
  */
   else if (DECL_EXTERNAL (exp))
     local_p = false;

I was wondering if a variable that was marked with a visibility for 
documentation purposes in a header file that was external, should be 
considered local or not.  I'd expect that it would not be local, but 
for that to be true, the two lines above would have to be swapped.

?  Any downside to swapping them?

Construct found in some darwin header files.

Re: binds_local question

[Daniel Jacobowitz]

On Mon, Aug 30, 2004 at 05:12:56PM -0700, Mike Stump wrote:
> In varasm.c we have:
> 
>   /* A variable is local if the user explicitly tells us so.  */
>   else if (DECL_VISIBILITY_SPECIFIED (exp) && DECL_VISIBILITY (exp) != 
> VISIBILITY_DEFAULT)
>     local_p = true;
>   /* Otherwise, variables defined outside this object may not be local. 
>  */
>   else if (DECL_EXTERNAL (exp))
>     local_p = false;
> 
> I was wondering if a variable that was marked with a visibility for 
> documentation purposes in a header file that was external, should be 
> considered local or not.  I'd expect that it would not be local, but 
> for that to be true, the two lines above would have to be swapped.
> 
> ?  Any downside to swapping them?
> 
> Construct found in some darwin header files.

A variable with specified visibility should be "local" whether or not
it's defined in this translation unit - that's most of the point.  We
know it will bind within this dynamic object.

-- 
Daniel Jacobowitz

Re: binds_local question

[Mike Stump]

On Aug 30, 2004, at 5:33 PM, Daniel Jacobowitz wrote:
> A variable with specified visibility should be "local" whether or not
> it's defined in this translation unit - that's most of the point.

Ok, then Mark got this slightly wrong?!

*** ./config/darwin.c.~1~       Sat Aug 21 23:36:18 2004
--- ./config/darwin.c   Mon Aug 30 18:29:43 2004
*************** name_needs_quotes (const char *name)
*** 141,149 ****
   static int
   machopic_symbol_defined_p (rtx sym_ref)
   {
!   return ((SYMBOL_REF_FLAGS (sym_ref) & MACHO_SYMBOL_FLAG_DEFINED)
!         /* Local symbols must always be defined.  */
!         || SYMBOL_REF_LOCAL_P (sym_ref));
   }

   /* This module assumes that (const (symbol_ref "foo")) is a legal pic
--- 141,148 ----
   static int
   machopic_symbol_defined_p (rtx sym_ref)
   {
!   /* APPLE LOCAL machopic_symbol_defined_p bugfix */
!   return (SYMBOL_REF_FLAGS (sym_ref) & MACHO_SYMBOL_FLAG_DEFINED);
   }

   /* This module assumes that (const (symbol_ref "foo")) is a legal pic


:-(  Mark, was there some other spelling you liked/wanted/needed?

Re: binds_local question

[Richard Henderson]

On Mon, Aug 30, 2004 at 06:57:02PM -0700, Mike Stump wrote:
>   machopic_symbol_defined_p (rtx sym_ref)
>   {
> -   return ((SYMBOL_REF_FLAGS (sym_ref) & MACHO_SYMBOL_FLAG_DEFINED)
> -         /* Local symbols must always be defined.  */
> -         || SYMBOL_REF_LOCAL_P (sym_ref));
> +   /* APPLE LOCAL machopic_symbol_defined_p bugfix */
> +   return (SYMBOL_REF_FLAGS (sym_ref) & MACHO_SYMBOL_FLAG_DEFINED);
>   }

From the names of the symbols involved, this looks correct.


r~

Re: binds_local question

[Mark Mitchell]

Mike Stump wrote:

> On Aug 30, 2004, at 5:33 PM, Daniel Jacobowitz wrote:
>
>> A variable with specified visibility should be "local" whether or not
>> it's defined in this translation unit - that's most of the point.
>
>
> Ok, then Mark got this slightly wrong?!

Mike, you sound so amazed.  I've gotten things wrong before. :-)

> :-(  Mark, was there some other spelling you liked/wanted/needed?

I think I thought that "local" meant "within this translation unit".

-- 
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com