• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*!
2 @file
3 Forward declares `boost::hana::Sequence`.
4 
5 @copyright Louis Dionne 2013-2017
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_SEQUENCE_HPP
11 #define BOOST_HANA_FWD_CONCEPT_SEQUENCE_HPP
12 
13 #include <boost/hana/config.hpp>
14 #include <boost/hana/core/when.hpp>
15 
16 
17 BOOST_HANA_NAMESPACE_BEGIN
18     //! @ingroup group-concepts
19     //! @defgroup group-Sequence Sequence
20     //! The `Sequence` concept represents generic index-based sequences.
21     //!
22     //! Compared to other abstract concepts, the Sequence concept is very
23     //! specific. It represents generic index-based sequences. The reason
24     //! why such a specific concept is provided is because there are a lot
25     //! of models that behave exactly the same while being implemented in
26     //! wildly different ways. It is useful to regroup all those data types
27     //! under the same umbrella for the purpose of generic programming.
28     //!
29     //! In fact, models of this concept are not only _similar_. They are
30     //! actually _isomorphic_, in a sense that we define below, which is
31     //! a fancy way of rigorously saying that they behave exactly the same
32     //! to an external observer.
33     //!
34     //!
35     //! Minimal complete definition
36     //! ---------------------------
37     //! `Iterable`, `Foldable`, and `make`
38     //!
39     //! The `Sequence` concept does not provide basic methods that could be
40     //! used as a minimal complete definition; instead, it borrows methods
41     //! from other concepts and add laws to them. For this reason, it is
42     //! necessary to specialize the `Sequence` metafunction in Hana's
43     //! namespace to tell Hana that a type is indeed a `Sequence`. Explicitly
44     //! specializing the `Sequence` metafunction can be seen like a seal
45     //! saying "this data type satisfies the additional laws of a `Sequence`",
46     //! since those can't be checked by Hana automatically.
47     //!
48     //!
49     //! Laws
50     //! ----
51     //! The laws for being a `Sequence` are simple, and their goal is to
52     //! restrict the semantics that can be associated to the functions
53     //! provided by other concepts. First, a `Sequence` must be a finite
54     //! `Iterable` (thus a `Foldable` too). Secondly, for a `Sequence` tag
55     //! `S`, `make<S>(x1, ..., xn)` must be an object of tag `S` and whose
56     //! linearization is `[x1, ..., xn]`. This basically ensures that objects
57     //! of tag `S` are equivalent to their linearization, and that they can
58     //! be created from such a linearization (with `make`).
59     //!
60     //! While it would be possible in theory to handle infinite sequences,
61     //! doing so complicates the implementation of many algorithms. For
62     //! simplicity, the current version of the library only handles finite
63     //! sequences. However, note that this does not affect in any way the
64     //! potential for having infinite `Searchable`s and `Iterable`s.
65     //!
66     //!
67     //! Refined concepts
68     //! ----------------
69     //! 1. `Comparable` (definition provided automatically)\n
70     //! Two `Sequence`s are equal if and only if they contain the same number
71     //! of elements and their elements at any given index are equal.
72     //! @include example/sequence/comparable.cpp
73     //!
74     //! 2. `Orderable` (definition provided automatically)\n
75     //! `Sequence`s are ordered using the traditional lexicographical ordering.
76     //! @include example/sequence/orderable.cpp
77     //!
78     //! 3. `Functor` (definition provided automatically)\n
79     //! `Sequence`s implement `transform` as the mapping of a function over
80     //! each element of the sequence. This is somewhat equivalent to what
81     //! `std::transform` does to ranges of iterators. Also note that mapping
82     //! a function over an empty sequence returns an empty sequence and never
83     //! applies the function, as would be expected.
84     //! @include example/sequence/functor.cpp
85     //!
86     //! 4. `Applicative` (definition provided automatically)\n
87     //! First, `lift`ing a value into a `Sequence` is the same as creating a
88     //! singleton sequence containing that value. Second, applying a sequence
89     //! of functions to a sequence of values will apply each function to
90     //! all the values in the sequence, and then return a list of all the
91     //! results. In other words,
92     //! @code
93     //!     ap([f1, ..., fN], [x1, ..., xM]) == [
94     //!         f1(x1), ..., f1(xM),
95     //!         ...
96     //!         fN(x1), ..., fN(xM)
97     //!     ]
98     //! @endcode
99     //! Example:
100     //! @include example/sequence/applicative.cpp
101     //!
102     //! 5. `Monad` (definition provided automatically)\n
103     //! First, `flaten`ning a `Sequence` takes a sequence of sequences and
104     //! concatenates them to get a larger sequence. In other words,
105     //! @code
106     //!     flatten([[a1, ..., aN], ..., [z1, ..., zM]]) == [
107     //!         a1, ..., aN, ..., z1, ..., zM
108     //!     ]
109     //! @endcode
110     //! This acts like a `std::tuple_cat` function, except it receives a
111     //! sequence of sequences instead of a variadic pack of sequences to
112     //! flatten.\n
113     //! __Example__:
114     //! @include example/sequence/monad.ints.cpp
115     //! Also note that the model of `Monad` for `Sequence`s can be seen as
116     //! modeling nondeterminism. A nondeterministic computation can be
117     //! modeled as a function which returns a sequence of possible results.
118     //! In this line of thought, `chain`ing a sequence of values into such
119     //! a function will return a sequence of all the possible output values,
120     //! i.e. a sequence of all the values applied to all the functions in
121     //! the sequences.\n
122     //! __Example__:
123     //! @include example/sequence/monad.types.cpp
124     //!
125     //! 6. `MonadPlus` (definition provided automatically)\n
126     //! `Sequence`s are models of the `MonadPlus` concept by considering the
127     //! empty sequence as the unit of `concat`, and sequence concatenation
128     //! as `concat`.
129     //! @include example/sequence/monad_plus.cpp
130     //!
131     //! 7. `Foldable`\n
132     //! The model of `Foldable` for `Sequence`s is uniquely determined by the
133     //! model of `Iterable`.
134     //! @include example/sequence/foldable.cpp
135     //!
136     //! 8. `Iterable`\n
137     //! The model of `Iterable` for `Sequence`s corresponds to iteration over
138     //! each element of the sequence, in order. This model is not provided
139     //! automatically, and it is in fact part of the minimal complete
140     //! definition for the `Sequence` concept.
141     //! @include example/sequence/iterable.cpp
142     //!
143     //! 9. `Searchable` (definition provided automatically)\n
144     //! Searching through a `Sequence` is equivalent to just searching through
145     //! a list of the values it contains. The keys and the values on which
146     //! the search is performed are both the elements of the sequence.
147     //! @include example/sequence/searchable.cpp
148     //!
149     //!
150     //! Concrete models
151     //! ---------------
152     //! `hana::tuple`
153     //!
154     //!
155     //! [1]: http://en.wikipedia.org/wiki/Isomorphism#Isomorphism_vs._bijective_morphism
156 #ifdef BOOST_HANA_DOXYGEN_INVOKED
157     template <typename S>
158     struct Sequence;
159 #else
160     template <typename S, typename = void>
161     struct Sequence : Sequence<S, when<true>> { };
162 #endif
163 BOOST_HANA_NAMESPACE_END
164 
165 #endif // !BOOST_HANA_FWD_CONCEPT_SEQUENCE_HPP
166