]> git.proxmox.com Git - ceph.git/blob - ceph/src/boost/libs/hana/include/boost/hana/fwd/concept/struct.hpp
add subtree-ish sources for 12.0.3
[ceph.git] / ceph / src / boost / libs / hana / include / boost / hana / fwd / concept / struct.hpp
1 /*!
2 @file
3 Forward declares `boost::hana::Struct`.
4
5 @copyright Louis Dionne 2013-2016
6 Distributed under the Boost Software License, Version 1.0.
7 (See accompanying file LICENSE.md or copy at http://boost.org/LICENSE_1_0.txt)
8 */
9
10 #ifndef BOOST_HANA_FWD_CONCEPT_STRUCT_HPP
11 #define BOOST_HANA_FWD_CONCEPT_STRUCT_HPP
12
13 #include <boost/hana/config.hpp>
14
15
16 BOOST_HANA_NAMESPACE_BEGIN
17 //! @ingroup group-concepts
18 //! @defgroup group-Struct Struct
19 //! The `Struct` concept represents `struct`-like user-defined types.
20 //!
21 //! The `Struct` concept allows restricted compile-time reflection over
22 //! user-defined types. In particular, it allows accessing the names of
23 //! the members of a user-defined type, and also the value of those
24 //! members. `Struct`s can also be folded, searched and converted to
25 //! some types of containers, where more advanced transformations can
26 //! be performed.
27 //!
28 //! While all types can _in theory_ be made `Struct`s, only a subset of
29 //! them are actually interesting to see as such. More precisely, it is
30 //! only interesting to make a type a `Struct` when it is conceptually
31 //! a C++ `struct`, i.e. a mostly dumb aggregate of named data. The way
32 //! this data is accessed is mostly unimportant to the `Struct` concept;
33 //! it could be through getters and setters, through public members,
34 //! through non-member functions or it could even be generated on-the-fly.
35 //! The important part, which is made precise below, is that those accessor
36 //! methods should be move-independent.
37 //!
38 //! Another way to see a `Struct` is as a map where the keys are the names
39 //! of the members and the values are the values of those members. However,
40 //! there are subtle differences like the fact that one can't add a member
41 //! to a `Struct`, and also that the order of the members inside a `Struct`
42 //! plays a role in determining the equality of `Struct`s, which is not
43 //! the case for maps.
44 //!
45 //!
46 //! Minimal complete definition
47 //! ---------------------------
48 //! `accessors`
49 //!
50 //! A model of `Struct` is created by specifying a sequence of key/value
51 //! pairs with the `accessors` function. The first element of a pair in
52 //! this sequence represents the "name" of a member of the `Struct`, while
53 //! the second element is a function which retrieves this member from an
54 //! object. The "names" do not have to be in any special form; they just
55 //! have to be compile-time `Comparable`. For example, it is common to
56 //! provide "names" that are `hana::string`s representing the actual names
57 //! of the members, but one could provide `hana::integral_constant`s just
58 //! as well. The values must be functions which, when given an object,
59 //! retrieve the appropriate member from it.
60 //!
61 //! There are several ways of providing the `accessors` method, some of
62 //! which are more flexible and others which are more convenient. First,
63 //! one can define it through tag-dispatching, as usual.
64 //! @snippet example/struct.mcd.tag_dispatching.cpp main
65 //!
66 //! Secondly, it is possible to provide a nested `hana_accessors_impl`
67 //! type, which should be equivalent to a specialization of
68 //! `accessors_impl` for tag-dispatching. However, for a type `S`, this
69 //! technique only works when the data type of `S` is `S` itself, which
70 //! is the case unless you explicitly asked for something else.
71 //! @snippet example/struct.mcd.nested.cpp main
72 //!
73 //! Finally, the most convenient (but least flexible) option is to use
74 //! the `BOOST_HANA_DEFINE_STRUCT`, the `BOOST_HANA_ADAPT_STRUCT` or the
75 //! `BOOST_HANA_ADAPT_ADT` macro, which provide a minimal syntactic
76 //! overhead. See the documentation of these macros for details on how
77 //! to use them.
78 //!
79 //! Also note that it is not important that the accessor functions retrieve
80 //! an actual member of the struct (e.g. `x.member`). Indeed, an accessor
81 //! function could call a custom getter or even compute the value of the
82 //! member on the fly:
83 //! @snippet example/struct.custom_accessor.cpp main
84 //!
85 //! The only important thing is that the accessor functions are
86 //! move-independent, a notion which is defined below.
87 //!
88 //!
89 //! @anchor move-independence
90 //! Move-independence
91 //! -----------------
92 //! The notion of move-independence presented here defines rigorously
93 //! when it is legitimate to "double-move" from an object.
94 //!
95 //! A collection of functions `f1, ..., fn` sharing the same domain is
96 //! said to be _move-independent_ if for every fresh (not moved-from)
97 //! object `x` in the domain, any permutation of the following statements
98 //! is valid and leaves the `zk` objects in a fresh (not moved-from) state:
99 //! @code
100 //! auto z1 = f1(std::move(x));
101 //! ...
102 //! auto zn = fn(std::move(x));
103 //! @endcode
104 //!
105 //! @note
106 //! In the special case where some functions return objects that can't be
107 //! bound to with `auto zk =` (like `void` or a non-movable, non-copyable
108 //! type), just pretend the return value is ignored.
109 //!
110 //! Intuitively, this ensures that we can treat `f1, ..., fn` as
111 //! "accessors" that decompose `x` into independent subobjects, and
112 //! that do so without moving from `x` more than that subobject. This
113 //! is important because it allows us to optimally decompose `Struct`s
114 //! into their subparts inside the library.
115 //!
116 //!
117 //! Laws
118 //! ----
119 //! For any `Struct` `S`, the accessors in the `accessors<S>()` sequence
120 //! must be move-independent, as defined above.
121 //!
122 //!
123 //! Refined concepts
124 //! ----------------
125 //! 1. `Comparable` (free model)\n
126 //! `Struct`s are required to be `Comparable`. Specifically, two `Struct`s
127 //! of the same data type `S` must be equal if and only if all of their
128 //! members are equal. By default, a model of `Comparable` doing just that
129 //! is provided for models of `Struct`. In particular, note that the
130 //! comparison of the members is made in the same order as they appear in
131 //! the `hana::members` sequence.
132 //! @include example/struct/comparable.cpp
133 //!
134 //! 2. `Foldable` (free model)\n
135 //! A `Struct` can be folded by considering it as a list of pairs each
136 //! containing the name of a member and the value associated to that
137 //! member, in the same order as they appear in the `hana::members`
138 //! sequence. By default, a model of `Foldable` doing just that is
139 //! provided for models of the `Struct` concept.
140 //! @include example/struct/foldable.cpp
141 //! Being a model of `Foldable` makes it possible to turn a `Struct`
142 //! into basically any `Sequence`, but also into a `hana::map` by simply
143 //! using the `to<...>` function!
144 //! @include example/struct/to.cpp
145 //!
146 //! 3. `Searchable` (free model)\n
147 //! A `Struct` can be searched by considering it as a map where the keys
148 //! are the names of the members of the `Struct`, and the values are the
149 //! members associated to those names. By default, a model of `Searchable`
150 //! is provided for any model of the `Struct` concept.
151 //! @include example/struct/searchable.cpp
152 template <typename S>
153 struct Struct;
154 BOOST_HANA_NAMESPACE_END
155
156 #endif // !BOOST_HANA_FWD_CONCEPT_STRUCT_HPP