Re: [RFC} Supporting aligned commons in PE.

[Dave Korn <dave dot korn dot cygwin at googlemail dot com>]

Danny Smith wrote:
> Dave Korn wrote
> 
> 
>> 2)  Add a new type of statement to the .drectve section, with a syntax such as
>> "-aligncomm:<symbolname>,<alignment>".  This might cause noisy warnings from
>> the vendors tools, but I think it's a cleaner interface than inventing our own
>> non-approved aux symbol format, so up til now it's my preferred option.
> 
> 
> This is my preferred option too, since the vendors own spec
> documentation says that the .drectve section is meant to handle
> compiler specific features.

  It does come with one slight disadvantage built in; putting what is
effectively metadata about symbols into a section, rather than into the symbol
table itself, where it belongs, might increase the chance it'll somehow come
adrift from the symbols to which it refers, which just couldn't happen if we
invented a new aux symbol entry type.  But I guess that's a theoretical risk
to be contrasted with the fact that the technique works well in practice.

> I think that other features (eg, visibility attributes to hide symbols
> from --export-all) could also be incorporated into .drectve.

  Yes, definitely.  I guess I should add dynamic-relinking info there too in
my grand-plan-of-things.

> I actually starting working up a patch along these lines but real-
> (aka making-a-living-) life got in the way.

  I am currently fortunate enough to be blessed with a period of
unemployment(*); leave it to me!

    cheers,
      DaveK
-- 
(*) - aka "freedom and opportunity" :)