Re: [boost] nested BOOST_FOREACH and -Wshadow

Subject: Re: [boost] nested BOOST_FOREACH and -Wshadow
From: Eric Niebler (eric_at_[hidden])
Date: 2009-02-09 19:55:59

Dustin Spicuzza wrote:
> David Abrahams wrote:
>> on Sat Feb 07 2009, Sebastian Redl <sebastian.redl-AT-getdesigned.at> wrote:
>>> John Bytheway wrote:
>>>> I'd suggest a simpler (from the user perspective) solution would be to
>>>> have BOOST_FOREACH paste __LINE__ into its variable names.  Then the
>>>> problem would only occur if nested BOOST_FOREACHs were used on the same
>>>> line.  Asking users not to do that seems a lot more reasonable than
>>>> asking them not to use -Wshadow or not to use nested BOOST_FOREACHs at all.
>>>>   
>>> I'm pretty sure you'd need a special variant for MS compilers, since
>>> their __LINE__ expansion is not cleanly pasteable. (You can use their
>>> counting macro instead.)
>> 
>> Doesn't using BOOST_PP_CAT work?
> 
> Ok, so attached is a simple patch to BOOST_FOREACH (v1.37) using the
> __LINE__ mechanism suggested. It works for gcc 4.3, but I don't have
> access to other compilers at the moment -- and even if I did, I'm not
> quite sure of the best way to mix the __LINE__ and __COUNTER__
> implementations correctly.

Thanks for the patch. I've made my own minor modifications and committed 
it to trunk. Since I also recall there being problems on msvc with 
__LINE__ and msvc's edit-and-continue feature, I'm simply not mangling 
names for msvc. That's ok because msvc doesn't emit shadow warnings, 
AFAIK. (Using __COUNTER__ on msvc would actually be wrong because it 
would become impossible to refer back to an identifier created previously.)

-- 
Eric Niebler
BoostPro Computing
http://www.boostpro.com