[Boost-users] [mpl][fusion] Issues in the mpl/fusion interface

[Gabriel Redner]

Hi all,

I am writing some code which uses MPL and fusion together, and in
general I am happy with their interoperability. However, I have
stumbled on a few areas in which functionality I expected was missing,
or where expressions I expected to work did not. All of these issues
have workarounds, but I'd appreciate some input on whether the problem
is with the libraries, or with my expectations.

1. mpl::pair and fusion::pair

Fusion sequences are conforming MPL sequences, and MPL sequences can
be transmuted into fusion sequences. However, the same is not true of
their pair classes. I feel that fusion::pair ought to conform to some
"MPL pair" concept such that:

- I can access its types by the same 'first' and 'second' members as
in mpl::pair
- mpl::first and mpl::second metafunctions should work against it

For instance, I'd like to be able to extract a sequence of key types from a map:

// OK!
typedef mpl::map<mpl::pair<int, char>, mpl::pair<bool, short> > m1;
typedef mpl::transform<
m1,
mpl::first<mpl::placeholders::_1>,
mpl::back_inserter<mpl::vector<> > >::type keys1;

// Not OK!
typedef fusion::map<fusion::pair<int, char>, fusion::pair<bool, short> > m2;
typedef mpl::transform<
m2,
mpl::first<mpl::placeholders::_1>,
mpl::back_inserter<mpl::vector<> > >::type keys2;

// OK, but requires different code for mpl and fusion maps
typedef mpl::transform<
m2,
fusion::result_of::first<mpl::placeholders::_1>,
mpl::back_inserter<mpl::vector<> > >::type keys3;

Also, I feel there should be a canned way to translate mpl::pair into
fusion::pair. fusion::result_of::as_pair?

2. mpl::map and fusion::map

This code works nicely:

typedef mpl::vector<int, char, bool> v1;
typedef fusion::result_of::as_vector<v1>::type v2;
v2 v(1, 'a', false);

However, this does not:

typedef mpl::map<mpl::pair<int, char>, mpl::pair<bool, short> > m1;
typedef fusion::result_of::as_map<m1>::type m2;
m2 m('a', 3);

The type 'm2' ends up as fusion::map<mpl::pair<int, char>,
mpl::pair<bool, short> >, which clearly is no good. I feel that
as_map should automatically translate mpl::pair to fusion::pair so
that the result is a usable container.

Please let me know if I am barking up the wrong tree, or whether these
seem like useful features to anyone else. If there is consensus that
they are useful, I would be interested in developing patches for them.

Thanks,
-Gabe
_______________________________________________
Boost-users mailing list
Boost...@lists.boost.org
http://lists.boost.org/mailman/listinfo.cgi/boost-users

[Joel de Guzman]

On 8/9/2011 6:12 AM, Gabriel Redner wrote:
> Hi all,
>
> I am writing some code which uses MPL and fusion together, and in
> general I am happy with their interoperability. However, I have
> stumbled on a few areas in which functionality I expected was missing,
> or where expressions I expected to work did not. All of these issues
> have workarounds, but I'd appreciate some input on whether the problem
> is with the libraries, or with my expectations.
>
> 1. mpl::pair and fusion::pair
>
> Fusion sequences are conforming MPL sequences, and MPL sequences can
> be transmuted into fusion sequences. However, the same is not true of
> their pair classes. I feel that fusion::pair ought to conform to some
> "MPL pair" concept such that:
>
> - I can access its types by the same 'first' and 'second' members as
> in mpl::pair
> - mpl::first and mpl::second metafunctions should work against it

That sounds reasonable.

> Also, I feel there should be a canned way to translate mpl::pair into
> fusion::pair. fusion::result_of::as_pair?

Ok. Sounds good.

> 2. mpl::map and fusion::map
>
> This code works nicely:
>
> typedef mpl::vector<int, char, bool> v1;
> typedef fusion::result_of::as_vector<v1>::type v2;
> v2 v(1, 'a', false);
>
> However, this does not:
>
> typedef mpl::map<mpl::pair<int, char>, mpl::pair<bool, short> > m1;
> typedef fusion::result_of::as_map<m1>::type m2;
> m2 m('a', 3);
>
> The type 'm2' ends up as fusion::map<mpl::pair<int, char>,
> mpl::pair<bool, short> >, which clearly is no good. I feel that
> as_map should automatically translate mpl::pair to fusion::pair so
> that the result is a usable container.
>
> Please let me know if I am barking up the wrong tree, or whether these
> seem like useful features to anyone else. If there is consensus that
> they are useful, I would be interested in developing patches for them.

All these sound reasonable. I'd definitely welcome a patch,
but I have to insist on docs too.

Regards,
--
Joel de Guzman
http://www.boostpro.com
http://boost-spirit.com

[Gabriel Redner]

Hi Joel,

OK, thanks for the responses. When I have some free time I'll look
into implementing this, and I'm happy to write docs.

-Gabe