Re: [PATCH] change the condition preventing symbol replacement in snapshot_symbol()

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Jul 27, 2009 at 9:46 AM, Nick Clifton<nickc@redhat.com> wrote:
> Hi Jan,
>
>> The question is whether an equate counts as a local symbol.
>
>> I'd really hope that Alan and or Nick could jump in here clarifying
>> expected
>> behavior.
>
> I didn't reply because I do not know the answer.
>
> The documentation already states that equating one symbol with another is
> the same as using the .set directive. ?The .set directive says that the new
> symbol inherits the value and type of the original symbol. ?Thus one would

Yes, "the value and type". It doesn't include other properties like binding and
visibility.

> expect that in:
>
> ?.global foo
> ?bar = foo
>
> That bar behaves like a global symbol, not a local one.
>
> We can change the documentation of course, and as H.J. has already pointed
> out, if foo is defined then the current behaviour is to treat bar as a local
> symbol.
>
> To me though this seems wrong. ?As a programmer I would expect that equated
> symbols are the same. ?Ie that they can be treated as aliases of each other.
> ?I do not see the benefit of changing the nature of equated symbol.

The .set directive creates an alias if the source is undefined. Otherwise,
many softwares which depend on the current behavior will be broken.

> I extended H.J.'s demonstration slightly:
>
> ?bar = foo
> ?baz = globule
>
> ? ? ? ?.globl foo
> ? ? ? ?.globl globule
> ?foo:
> ? ? ? ?jmp bar
> ? ? ? ?jmp foo
> ? ? ? ?jmp baz
> ? ? ? ?jmp globule
>
> Which when assembled and dumped with objdump -dr gives:
>
> 00000000 <foo>:
> ? 0: ? eb fe ? ? ? ? ? ? ? ? ? jmp ? ?0 <foo>
> ? 2: ? e9 fc ff ff ff ? ? ? ? ?jmp ? ?3 <foo+0x3>
> ? ? ? ? ? ? ? ? ? ? ? ?3: R_386_PC32 ? foo
> ? 7: ? e9 fc ff ff ff ? ? ? ? ?jmp ? ?8 <foo+0x8>
> ? ? ? ? ? ? ? ? ? ? ? ?8: R_386_PC32 ? globule
> ? c: ? e9 fc ff ff ff ? ? ? ? ?jmp ? ?d <foo+0xd>
> ? ? ? ? ? ? ? ? ? ? ? ?d: R_386_PC32 ? globule
>
> The first jmp just seems plain wrong to me. ?I would expect it to have a
> relocation against the foo symbol just like all of the others.

That is a bug. I will look into it.

> So, in my opinion: "no, equating a symbol does not make it local".
>

-- 
H.J.