on Fri Mar 23 2007, Scott Meyers wrote:

David Abrahams wrote:

...

Nothing; this is clearly a bug in MPL

As a workaround, I thought I'd write an erase_first function as shown below, but this seems to have the same problem. Perhaps this is because I'm using erase in my implementation, and erase_key may, I suppose, be implemented on top of erase.

It is *extremely* likely that erase_key and erase produce the same result.

More likely, I'm doing something wrong.

Not all that likely.

More likely still, the way I'm doing it is gross.

Well yeah. The un-gross way is to wait for Aleksey to fix it. He's unfortunately in a work crunch at the moment, but I'd guess he could address it within a couple of days.

I'd appreciate feedback on the implementation of my erase_first function, where I've put comments showing the procedural equivalent of what I'm trying to do in each statement.

Also, does anybody have a workaround for the bug in erase_key?

It's highly unlikely that any opportunity for such a workaround exists in the set code.

Thanks,

Scott

#include #include #include #include #include #include

namespace mpl = boost::mpl; using mpl::_1; using mpl::_2;

struct A {}; struct B {};

typedef mpl::set set1; typedef mpl::set set2;

// return Seq with the first occurrance of T removed template struct erase_first { // i = Seq.find(T) typedef typename mpl::find::type i;

// found = (i == Seq.end()) ? false : true typedef typename mpl::if_< typename boost::is_same::type, mpl::false_, mpl::true_ >::type found;

mpl::if_< pred, mpl::false_, mpl::true_>::type is usually better written mpl::not_<pred>

// type = found ? Seq : Seq.erase(i) typedef typename mpl::if_::type >::type type; };

From scratch (untested):

template struct erase_first { typedef typename mpl::find::type i; typedef typename eval_if< boost::is_same , mpl::identity<Seq> , mpl::erase >::type type; }; HTH, -- Dave Abrahams Boost Consulting www.boost-consulting.com

On 3/23/07, Scott Meyers wrote:

// return Seq with the first occurrance of T removed template struct erase_first { // i = Seq.find(T) typedef typename mpl::find::type i;

// found = (i == Seq.end()) ? false : true typedef typename mpl::if_< typename boost::is_same::type, mpl::false_, mpl::true_ >::type found;

// type = found ? Seq : Seq.erase(i)

// type = found ? Seq.erase(i) : Seq

typedef typename mpl::if_::type >::type type;

typedef typename if_::type, Seq >::type type;

};

BOOST_MPL_ASSERT(( mpl::equal< erase_first::type, erase_first::type

...

));

Or did I misunderstand? Tony

On 3/23/07, Scott Meyers wrote:

Any other suggestions?

Thanks,

Scott

Do you know what the bug in mpl is? we have, essentially: : eval_if< has_key_implaux::set_tag::apply , eval_if< is_same< T,typename Set::item_type_ > , base<Set> , identity< s_mask > > , identity<Set> > Can we assume kas_key_impl works? I would hope so, else mpl::set is a bit useless. The if (is_same) then base<Set> is pretty simple, so I suspect it is the s_mask that is causing problems. And, unfortunately, although staring for the usual requisite hour, s_mask, although small (or because it is small), makes no sense to me. In particular, I thought it would have a 'type' typedef'd somewhere in it, but I don't see it. Tony