Re: [patch, rfc] Semantics of attribute (aligned), misscompilation of crtstuff

[Mark Mitchell <mark at codesourcery dot com>]

Daniel Jacobowitz wrote:
> On Tue, Oct 23, 2007 at 03:50:15PM +0200, Zdenek Dvorak wrote:
>>> How about __attribute__((aligned(sizeof(func_ptr)),packed)) instead of
>>> the runtime test?
>> do we support attribute packed on variables? I get
>>
>> ../../gcc/gcc/crtstuff.c:196: warning: ?packed? attribute ignored
>> ../../gcc/gcc/crtstuff.c:209: warning: ?packed? attribute ignored
>>
>> when I try to use it.
> 
> :-(
> 
> The documentation isn't clear... it reads like that should work
> (and be right).

I think this is a documentation bug.  In particular, the manual does not
list "packed" under "Variable Attributes" -- but it does say, in the
description of the "aligned" attribute that "packed" can decrease
alignment.  Is that what suggested to you that the combination was
valid?  I think what's happened is that the "aligned" documentation is
shared between "Type Attributes" and "Variable Attributes"; we probably
need to change the "Variable Attributes" copy not to mention "packed"
since it doesn't make sense in this context.

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713