// Copyright 2012 Google Inc. All Rights Reserved.
//
// This file declares INTERNAL parts of the Split API that are inline/templated
// or otherwise need to be available at compile time. The main two abstractions
// defined in here are
//
//   - SplitIterator<>
//   - Splitter<>
//
// Everything else is plumbing for those two.
//
// DO NOT INCLUDE THIS FILE DIRECTLY. Use this file by including
// strings/split.h.
//
// IWYU pragma: private, include "strings/split.h"

#ifndef STRINGS_SPLIT_INTERNAL_H_
#define STRINGS_SPLIT_INTERNAL_H_

#include <iterator>
#include <map>
#include <vector>

#include "kudu/gutil/port.h"  // for LANG_CXX11
#include "kudu/gutil/strings/stringpiece.h"

#ifdef LANG_CXX11
// This must be included after "kudu/gutil/port.h", which defines LANG_CXX11.
#include <array>
#include <initializer_list>
#endif  // LANG_CXX11

namespace strings {

namespace internal {

// The default Predicate object, which doesn't filter out anything.
struct NoFilter {
  bool operator()(StringPiece /* ignored */) {
    return true;
  }
};

// This class splits a string using the given delimiter, returning the split
// substrings via an iterator interface. An optional Predicate functor may be
// supplied, which will be used to filter the split strings: strings for which
// the predicate returns false will be skipped. A Predicate object is any
// functor that takes a StringPiece and returns bool. By default, the NoFilter
// Predicate is used, which does not filter out anything.
//
// This class is NOT part of the public splitting API.
//
// Usage:
//
//   using strings::delimiter::Literal;
//   Literal d(",");
//   for (SplitIterator<Literal> it("a,b,c", d), end(d); it != end; ++it) {
//     StringPiece substring = *it;
//     DoWork(substring);
//   }
//
// The explicit single-argument constructor is used to create an "end" iterator.
// The two-argument constructor is used to split the given text using the given
// delimiter.
template <typename Delimiter, typename Predicate = NoFilter>
class SplitIterator
    : public std::iterator<std::input_iterator_tag, StringPiece> {
 public:
  // Two constructors for "end" iterators.
  explicit SplitIterator(Delimiter d)
      : delimiter_(std::move(d)), predicate_(), is_end_(true) {}
  SplitIterator(Delimiter d, Predicate p)
      : delimiter_(std::move(d)), predicate_(std::move(p)), is_end_(true) {}
  // Two constructors taking the text to iterator.
  SplitIterator(StringPiece text, Delimiter d)
      : text_(std::move(text)),
        delimiter_(std::move(d)),
        predicate_(),
        is_end_(false) {
    ++(*this);
  }
  SplitIterator(StringPiece text, Delimiter d, Predicate p)
      : text_(std::move(text)),
        delimiter_(std::move(d)),
        predicate_(std::move(p)),
        is_end_(false) {
    ++(*this);
  }

  StringPiece operator*() { return curr_piece_; }
  StringPiece* operator->() { return &curr_piece_; }

  SplitIterator& operator++() {
    do {
      if (text_.end() == curr_piece_.end()) {
        // Already consumed all of text_, so we're done.
        is_end_ = true;
        return *this;
      }
      StringPiece found_delimiter = delimiter_.Find(text_);
      assert(found_delimiter.data() != NULL);
      assert(text_.begin() <= found_delimiter.begin());
      assert(found_delimiter.end() <= text_.end());
      // found_delimiter is allowed to be empty.
      // Sets curr_piece_ to all text up to but excluding the delimiter itself.
      // Sets text_ to remaining data after the delimiter.
      curr_piece_.set(text_.begin(), found_delimiter.begin() - text_.begin());
      text_.remove_prefix(found_delimiter.end() - text_.begin());
    } while (!predicate_(curr_piece_));
    return *this;
  }

  SplitIterator operator++(int /* postincrement */) {
    SplitIterator old(*this);
    ++(*this);
    return old;
  }

  bool operator==(const SplitIterator& other) const {
    // Two "end" iterators are always equal. If the two iterators being compared
    // aren't both end iterators, then we fallback to comparing their fields.
    // Importantly, the text being split must be equal and the current piece
    // within the text being split must also be equal. The delimiter_ and
    // predicate_ fields need not be checked here because they're template
    // parameters that are already part of the SplitIterator's type.
    return (is_end_ && other.is_end_) ||
           (is_end_ == other.is_end_ &&
            text_ == other.text_ &&
            text_.data() == other.text_.data() &&
            curr_piece_ == other.curr_piece_ &&
            curr_piece_.data() == other.curr_piece_.data());
  }

  bool operator!=(const SplitIterator& other) const {
    return !(*this == other);
  }

 private:
  // The text being split. Modified as delimited pieces are consumed.
  StringPiece text_;
  Delimiter delimiter_;
  Predicate predicate_;
  bool is_end_;
  // Holds the currently split piece of text. Will always refer to string data
  // within text_. This value is returned when the iterator is dereferenced.
  StringPiece curr_piece_;
};

// Declares a functor that can convert a StringPiece to another type. This works
// for any type that has a constructor (explicit or not) taking a single
// StringPiece argument. A specialization exists for converting to string
// because the underlying data needs to be copied. In theory, these
// specializations could be extended to work with other types (e.g., int32), but
// then a solution for error reporting would need to be devised.
template <typename To>
struct StringPieceTo {
  To operator()(StringPiece from) const {
    return To(from);
  }
};

// Specialization for converting to string.
template <>
struct StringPieceTo<std::string> {
  std::string operator()(StringPiece from) const {
    return from.ToString();
  }
};

// Specialization for converting to *const* string.
template <>
struct StringPieceTo<const std::string> {
  std::string operator()(StringPiece from) const {
    return from.ToString();
  }
};

#ifdef LANG_CXX11
// IsNotInitializerList<T>::type exists iff T is not an initializer_list. More
// details below in Splitter<> where this is used.
template <typename T>
struct IsNotInitializerList {
  typedef void type;
};
template <typename T>
struct IsNotInitializerList<std::initializer_list<T> > {};
#endif  // LANG_CXX11

// This class implements the behavior of the split API by giving callers access
// to the underlying split substrings in various convenient ways, such as
// through iterators or implicit conversion functions. Do not construct this
// class directly, rather use the Split() function instead.
//
// Output containers can be collections of either StringPiece or string objects.
// StringPiece is more efficient because the underlying data will not need to be
// copied; the returned StringPieces will all refer to the data within the
// original input string. If a collection of string objects is used, then each
// substring will be copied.
//
// An optional Predicate functor may be supplied. This predicate will be used to
// filter the split strings: only strings for which the predicate returns true
// will be kept. A Predicate object is any unary functor that takes a
// StringPiece and returns bool. By default, the NoFilter predicate is used,
// which does not filter out anything.
template <typename Delimiter, typename Predicate = NoFilter>
class Splitter {
 public:
  typedef internal::SplitIterator<Delimiter, Predicate> Iterator;

  Splitter(StringPiece text, Delimiter d)
      : begin_(text, d), end_(d) {}

  Splitter(StringPiece text, Delimiter d, Predicate p)
      : begin_(text, d, p), end_(d, p) {}

  // Range functions that iterate the split substrings as StringPiece objects.
  // These methods enable a Splitter to be used in a range-based for loop in
  // C++11, for example:
  //
  //   for (StringPiece sp : my_splitter) {
  //     DoWork(sp);
  //   }
  const Iterator& begin() const { return begin_; }
  const Iterator& end() const { return end_; }

#ifdef LANG_CXX11
// Support for default template arguments for function templates was added in
// C++11, but it is not allowed if compiled in C++98 compatibility mode. Since
// this code is under a LANG_CXX11 guard, we can safely ignore the
// -Wc++98-compat flag and use default template arguments on the implicit
// conversion operator below.
//
// This use of default template arguments on a function template was approved
// by tgs and sanjay on behalf of the c-style-arbiters in email thread
//
// All compiler flags are first saved with a diagnostic push and restored with a
// diagnostic pop below.
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wpragmas"
#pragma GCC diagnostic ignored "-Wc++98-compat"

  // Uses SFINAE to restrict conversion to container-like types (by testing for
  // the presence of a const_iterator member type) and also to disable
  // conversion to an initializer_list (which also has a const_iterator).
  // Otherwise, code compiled in C++11 will get an error due to ambiguous
  // conversion paths (in C++11 vector<T>::operator= is overloaded to take
  // either a vector<T> or an initializer_list<T>).
  //
  // This trick was taken from util/gtl/container_literal.h
  template <typename Container,
            typename IsNotInitializerListChecker =
                typename IsNotInitializerList<Container>::type,
            typename ContainerChecker =
                typename Container::const_iterator>
  operator Container() {
    return SelectContainer<Container, is_map<Container>::value>()(this);
  }

  template <typename ArrayType, size_t ArraySize>
  operator std::array<ArrayType, ArraySize>() {
    return ToArray<ArrayType, ArraySize>();
  }

// Restores diagnostic settings, i.e., removes the "ignore" on -Wpragmas and
// -Wc++98-compat.
#pragma GCC diagnostic pop

#else
  // Not under LANG_CXX11
  template <typename Container>
  operator Container() {
    return SelectContainer<Container, is_map<Container>::value>()(this);
  }
#endif  // LANG_CXX11

  template <typename First, typename Second>
  operator std::pair<First, Second>() {
    return ToPair<First, Second>();
  }

 private:
  // is_map<T>::value is true iff there exists a type T::mapped_type. This is
  // used to dispatch to one of the SelectContainer<> functors (below) from the
  // implicit conversion operator (above).
  template <typename T>
  struct is_map {
    template <typename U> static base::big_ test(typename U::mapped_type*);
    template <typename> static base::small_ test(...);
    static const bool value = (sizeof(test<T>(0)) == sizeof(base::big_));
  };

  // Base template handles splitting to non-map containers
  template <typename Container, bool>
  struct SelectContainer {
    Container operator()(Splitter* splitter) const {
      return splitter->template ToContainer<Container>();
    }
  };

  // Partial template specialization for splitting to map-like containers.
  template <typename Container>
  struct SelectContainer<Container, true> {
    Container operator()(Splitter* splitter) const {
      return splitter->template ToMap<Container>();
    }
  };

  // Inserts split results into the container. To do this the results are first
  // stored in a vector<StringPiece>. This is where the input text is actually
  // "parsed". This vector is then used to possibly reserve space in the output
  // container, and the StringPieces in "v" are converted as necessary to the
  // output container's value type.
  //
  // The reason to use an intermediate vector of StringPiece is so we can learn
  // the needed capacity of the output container. This is needed when the output
  // container is a vector<string> in which case resizes can be expensive due to
  // copying of the ::string objects.
  //
  // At some point in the future we might add a C++11 move constructor to
  // ::string, in which case the vector resizes are much less expensive and the
  // use of this intermediate vector "v" can be removed.
  template <typename Container>
  Container ToContainer() {
    std::vector<StringPiece> v;
    for (Iterator it = begin(); it != end_; ++it) {
      v.push_back(*it);
    }
    typedef typename Container::value_type ToType;
    internal::StringPieceTo<ToType> converter;
    Container c;
    ReserveCapacity(&c, v.size());
    std::insert_iterator<Container> inserter(c, c.begin());
    for (const auto& sp : v) {
      *inserter++ = converter(sp);
    }
    return c;
  }

  // The algorithm is to insert a new pair into the map for each even-numbered
  // item, with the even-numbered item as the key with a default-constructed
  // value. Each odd-numbered item will then be assigned to the last pair's
  // value.
  template <typename Map>
  Map ToMap() {
    typedef typename Map::key_type Key;
    typedef typename Map::mapped_type Data;
    Map m;
    StringPieceTo<Key> key_converter;
    StringPieceTo<Data> val_converter;
    typename Map::iterator curr_pair;
    bool is_even = true;
    for (Iterator it = begin(); it != end_; ++it) {
      if (is_even) {
        curr_pair = InsertInMap(std::make_pair(key_converter(*it), Data()), &m);
      } else {
        curr_pair->second = val_converter(*it);
      }
      is_even = !is_even;
    }
    return m;
  }

  // Returns a pair with its .first and .second members set to the first two
  // strings returned by the begin() iterator. Either/both of .first and .second
  // will be empty strings if the iterator doesn't have a corresponding value.
  template <typename First, typename Second>
  std::pair<First, Second> ToPair() {
    StringPieceTo<First> first_converter;
    StringPieceTo<Second> second_converter;
    StringPiece first, second;
    Iterator it = begin();
    if (it != end()) {
      first = *it;
      if (++it != end()) {
        second = *it;
      }
    }
    return std::make_pair(first_converter(first), second_converter(second));
  }

#ifdef LANG_CXX11
  // Return an array from the split results.
  template <typename ArrayType, size_t ArraySize>
  std::array<ArrayType, ArraySize> ToArray() {
    StringPieceTo<ArrayType> converter;
    std::array<ArrayType, ArraySize> ret;

    typename std::array<ArrayType, ArraySize>::iterator out_it = ret.begin();
    Iterator it = begin();
    while (it != end() && out_it != ret.end()) {
      *out_it++ = converter(*it++);
    }
    return ret;
  }
#endif

  // Overloaded InsertInMap() function. The first overload is the commonly used
  // one for most map-like objects. The second overload is a special case for
  // multimap, because multimap's insert() member function directly returns an
  // iterator, rather than a pair<iterator, bool> like map's.
  template <typename Map>
  typename Map::iterator InsertInMap(
      const typename Map::value_type& value, Map* map) {
    return map->insert(value).first;
  }

  // InsertInMap overload for multimap.
  template <typename K, typename T, typename C, typename A>
  typename std::multimap<K, T, C, A>::iterator InsertInMap(
      const typename std::multimap<K, T, C, A>::value_type& value,
      typename std::multimap<K, T, C, A>* map) {
    return map->insert(value);
  }

  // Reserves the given amount of capacity in a vector<string>
  template <typename A>
  void ReserveCapacity(std::vector<std::string, A>* v, size_t size) {
    v->reserve(size);
  }
  void ReserveCapacity(...) {}

  const Iterator begin_;
  const Iterator end_;
};

}  // namespace internal

}  // namespace strings

#endif  // STRINGS_SPLIT_INTERNAL_H_