quasardb 3.13.6.post1__cp310-cp310-win_amd64.whl → 3.13.7__cp310-cp310-win_amd64.whl

quasardb/convert/unicode.hpp

@@ -1,25 +1,29 @@
 #pragma once
 
+#include "../concepts.hpp"
 #include <range/v3/range_fwd.hpp>
 #include <range/v3/view/adaptor.hpp>
 #include <range/v3/view/cache1.hpp>
+#include <range/v3/view/transform.hpp>
+#include <iostream>
+#include <optional>
 
 namespace qdb::convert::unicode
 {
 
-typedef std::uint32_t cp_type;
-typedef std::uint32_t u32_type; // UTF32 == CodePoint
+typedef std::uint32_t u32_type;
 typedef std::uint8_t u8_type;
 
-namespace encode
+namespace detail
 {
 
-template <typename CharT>
-struct fn_;
-
-template <typename CharT, std::size_t Width = (sizeof(cp_type) / sizeof(CharT))>
-    requires(1 <= Width && Width <= 4)
-struct next_chars
+/**
+ * General purpose buffer which holds the "next" characters. This is useful for
+ * buffering purposes, e.g. when encoding codepoints to UTF-8, where advancing the
+ * iterator forward by 1 can actually emit 4 characters.
+ */
+template <typename CharT, std::size_t Width = (sizeof(u32_type) / sizeof(CharT))>
+requires(1 <= Width && Width <= 4) struct next_chars
 {
 public:
     constexpr next_chars() = default;
@@ -29,45 +33,91 @@ public:
         , n_{1}
     {}
 
-    constexpr next_chars(CharT _1, CharT _2)
-        requires(Width >= 2)
-    : xs_{{_1, _2}}
-    , n_{2}
+    constexpr next_chars(CharT _1, CharT _2) requires(Width >= 2)
+        : xs_{{_1, _2}}
+        , n_{2}
     {}
 
-    constexpr next_chars(CharT _1, CharT _2, CharT _3)
-        requires(Width >= 3)
-    : xs_{{_1, _2, _3}}
-    , n_{3}
+    constexpr next_chars(CharT _1, CharT _2, CharT _3) requires(Width >= 3)
+        : xs_{{_1, _2, _3}}
+        , n_{3}
     {}
 
-    constexpr next_chars(CharT _1, CharT _2, CharT _3, CharT _4)
-        requires(Width == 4)
-    : xs_{{_1, _2, _3, _4}}
-    , n_{4}
+    constexpr next_chars(CharT _1, CharT _2, CharT _3, CharT _4) requires(Width == 4)
+        : xs_{{_1, _2, _3, _4}}
+        , n_{4}
     {}
+
     constexpr inline CharT pop() noexcept
     {
         assert(p_ < n_);
         return xs_[p_++];
     };
 
+    constexpr inline CharT top() const noexcept
+    {
+        assert(p_ < n_);
+        return xs_[p_];
+    };
+
     constexpr inline bool empty() const noexcept
     {
         return p_ == n_;
     };
 
+    constexpr inline std::size_t size() const noexcept
+    {
+        return n_ - p_;
+    }
+
 private:
     std::array<CharT, Width> xs_{{}};
     std::size_t n_ = 0;
     std::size_t p_ = 0;
 };
 
-// codepoint -> utf8
+class code_point
+{
+private:
+    std::uint32_t value_;
+
+public:
+    constexpr explicit code_point(std::uint32_t value)
+        : value_{value} {};
+
+    constexpr explicit code_point(std::uint32_t && value)
+        : value_{std::move(value)} {};
+
+    constexpr inline std::uint32_t get() const noexcept
+    {
+        return value_;
+    }
+
+    friend constexpr bool operator==(code_point const & lhs, code_point const & rhs)
+    {
+        return lhs.value_ == rhs.value_;
+    }
+    friend constexpr bool operator!=(code_point const & lhs, code_point const & rhs)
+    {
+        return !(lhs == rhs);
+    }
+};
+
+static_assert(sizeof(code_point) == sizeof(std::uint32_t));
+
+/**
+ * Functor that accepts codepoints and emits encoded UTF characters.
+ */
+template <typename CharT>
+struct encode_fn_;
+
+/**
+ * Functor for encoding codepoints to UTF-8.
+ */
 template <>
-struct fn_<u8_type>
+struct encode_fn_<u8_type>
 {
-    inline next_chars<u8_type> operator()(cp_type cp) const noexcept
+    inline next_chars<u8_type> operator()(std::uint32_t cp) const noexcept
     {
         if (cp >= (1L << 16)) [[unlikely]]
         {
@@ -92,199 +142,457 @@ struct fn_<u8_type>
             return {static_cast<u8_type>(cp)};
         }
     }
+
+    inline next_chars<u8_type> operator()(detail::code_point cp) const noexcept
+    {
+        return operator()(cp.get());
+    }
 };
 
 /**
- * UTF32->UTF8 encoding view
- *
- * Reads codepoints (i.e. UTF32/UCS4) and emits UTF8. Effectively used for
- * conversion from numpy to qdb.
+ * Functor that accepts UTF encoded characters and yields codepoints.
  */
-template <typename OutCharT, typename Rng>
-    requires(ranges::sized_range<Rng>)
-class view_ : public ranges::view_facade<view_<OutCharT, Rng>, ranges::finite>
-{
-    friend ranges::range_access;
-    using iterator_t = ranges::iterator_t<Rng>;
-    using value_type = OutCharT;
-
-public:
-    view_() = default;
-    explicit view_(Rng && rng) noexcept
-        : rng_{rng}
-        , iter_{ranges::begin(rng)}
-    {}
-
-    constexpr inline bool equal(ranges::default_sentinel_t) const noexcept
-    {
-        return iter_ == ranges::end(rng_) && next_.empty();
-    }
+template <typename CharT>
+struct decode_fn_;
 
-public:
-    // Actual encoding logic
-    inline value_type read() const noexcept
+/**
+ * UTF-8 -> CodePoint functor.
+ */
+template <>
+struct decode_fn_<u8_type>
+{
+    template <ranges::input_iterator I>
+    inline detail::code_point operator()(I it) const noexcept
     {
-        // Because our iterator may emit 1..4 positions for every read(), we cannot
-        // easily separate read() and next() without taking a performance hit.
-        //
-        // As such, we'll const_cast ourselves to make our internal 'next' chars
-        // writable.
-        view_ * this_ = const_cast<view_ *>(this);
-        return this_->read();
-    };
+        std::uint32_t cp;
 
-    inline value_type read() noexcept
-    {
-        if (next_.empty() == true) [[likely]]
+        if (*it < 0x80) [[likely]]
         {
-            assert(iter_ != ranges::end(rng_));
-            next_ = encode_(*(iter_++));
-        };
-
-        assert(next_.empty() == false);
-        return next_.pop();
-    };
-
-    constexpr inline void next() noexcept {};
+            cp = *it++;
+        }
+        else if ((*it & 0xe0) == 0xc0)
+        {
+            cp = ((long)(*it++ & 0x1f) << 6) | ((long)(*it++ & 0x3f) << 0);
+        }
+        else if ((*it & 0xf0) == 0xe0)
+        {
+            cp = ((long)(*it++ & 0x0f) << 12) | ((long)(*it++ & 0x3f) << 6)
+                 | ((long)(*it++ & 0x3f) << 0);
+        }
+        else if ((*it & 0xf8) == 0xf0 && (*it <= 0xf4))
+        {
+            cp = ((long)(*it++ & 0x07) << 18) | ((long)(*it++ & 0x3f) << 12)
+                 | ((long)(*it++ & 0x3f) << 6) | ((long)(*it++ & 0x3f) << 0);
+        }
+        else [[unlikely]]
+        {
+            cp = -1;
+            ++it;
+        }
 
-private:
-    Rng rng_;
-    iterator_t iter_;
-    next_chars<OutCharT> next_;
-    fn_<OutCharT> encode_;
+        return detail::code_point{cp};
+    }
 };
 
-template <typename OutCharT, class Rng>
-view_<OutCharT, Rng> view(Rng && rng)
-{
-    return view_<OutCharT, Rng>{std::forward<Rng>(rng)};
-}
-
-template <class Rng>
-view_<u8_type, Rng> utf8_view(Rng && rng)
-{
-    return view<u8_type, Rng>(std::forward<Rng>(rng));
-}
-
-}; // namespace encode
+/**
+ * Counts how many UTF characters are represented by a code point.
+ */
+template <typename CharT>
+struct count_fn_;
 
-namespace decode
+/**
+ * Counts how many UTF-8 characters are represented by a code point.
+ */
+template <>
+struct count_fn_<u8_type>
 {
+    inline std::size_t operator()(detail::code_point cp) const noexcept
+    {
+        if (cp.get() >= (1L << 16)) [[unlikely]]
+        {
+            return 4;
+        }
+        else if (cp.get() >= (1L << 11))
+        {
+            return 3;
+        }
+        else if (cp.get() >= (1L << 7))
+        {
+            return 2;
+        }
+        else [[likely]]
+        {
+            return 1;
+        }
+    }
+};
 
+/**
+ * Skips as many UTF characters as necessary to read a single code point.
+ */
 template <typename CharT>
-struct fn_;
+struct skip_fn_;
 
+/**
+ * Skips as many UTF-8 characters as necessary to read a single code point.
+ */
 template <>
-struct fn_<u8_type>
+struct skip_fn_<u8_type>
 {
     template <ranges::input_iterator I>
-    constexpr inline cp_type operator()(I & it) const noexcept
+    constexpr inline void operator()(I & it) const noexcept
     {
-        cp_type cp;
-
-        if (it[0] < 0x80) [[likely]]
+        if (*it < 0x80) [[likely]]
         {
-            cp = *it++;
+            ranges::advance(it, 1);
         }
-        else if ((it[0] & 0xe0) == 0xc0)
+        else if ((*it & 0xe0) == 0xc0)
         {
-            cp = ((long)(it[0] & 0x1f) << 6) | ((long)(it[1] & 0x3f) << 0);
-            it += 2;
+            ranges::advance(it, 2);
         }
-        else if ((it[0] & 0xf0) == 0xe0)
+        else if ((*it & 0xf0) == 0xe0)
         {
-            cp = ((long)(it[0] & 0x0f) << 12) | ((long)(it[1] & 0x3f) << 6)
-                 | ((long)(it[2] & 0x3f) << 0);
-            it += 3;
+            ranges::advance(it, 3);
         }
-        else if ((it[0] & 0xf8) == 0xf0 && (it[0] <= 0xf4))
+        else if ((*it & 0xf8) == 0xf0 && (*it <= 0xf4))
         {
-            cp = ((long)(it[0] & 0x07) << 18) | ((long)(it[1] & 0x3f) << 12)
-                 | ((long)(it[2] & 0x3f) << 6) | ((long)(it[3] & 0x3f) << 0);
-            it += 4;
+            ranges::advance(it, 4);
         }
         else [[unlikely]]
         {
-            cp = -1;
-            ++it;
+            ranges::advance(it, 1);
         }
+    }
+};
 
-        return cp;
+namespace utf8
+{
+
+/**
+ * CodePoint -> UTF8 encoding view
+ *
+ * Reads codepoints and emits UTF8. Effectively used for conversion from numpy to qdb.
+ */
+template <typename Rng>
+requires(concepts::forward_range_t<Rng, detail::code_point>) class encode_view_
+    : public ranges::view_facade<encode_view_<Rng>, ranges::finite>
+{
+    friend ranges::range_access;
+    using iterator_t = ranges::iterator_t<Rng>;
+    using value_type = u8_type;
+
+    /**
+     * Use a separate cursor struct, so we can implement the range as a forward range (i.e.
+     * make the iterators weakly comparable).
+     */
+    template <typename IterT, typename SentT>
+    struct cursor
+    {
+    private:
+        IterT iter_;
+        SentT last_;
+        next_chars<value_type> next_;
+        encode_fn_<value_type> encode_;
+        count_fn_<value_type> count_;
+
+    public:
+        cursor() = default;
+        cursor(IterT iter, SentT last)
+            : iter_{iter}
+            , last_{last}
+        {
+            next();
+        };
+
+        inline value_type read() const noexcept
+        {
+            return next_.top();
+        }
+
+        constexpr inline void next() noexcept
+        {
+            if (next_.empty() == false) [[likely]]
+            {
+                next_.pop();
+            }
+
+            if (iter_ != last_ && next_.empty() == true) [[likely]]
+            {
+                next_ = encode_(*(iter_++));
+            }
+        };
+
+        constexpr inline bool equal(cursor<IterT, SentT> const & rhs) const noexcept
+        {
+            return iter_ == rhs.iter_ && next_.empty() == rhs.next_.empty();
+        }
+
+        constexpr inline bool equal(ranges::default_sentinel_t) const noexcept
+        {
+            return iter_ == last_ && next_.empty();
+        }
+
+        inline ranges::iter_difference_t<IterT> distance_to(ranges::default_sentinel_t) const
+        {
+            // This is a hack, but we initialize the return value to `1` if we already read
+            // the next token in our constructor.
+            ranges::iter_difference_t<IterT> ret(next_.size());
+
+            // And finally, let's parse all the remaining tokens
+            for (IterT cur = iter_; cur != last_; ++cur)
+            {
+                ret += count_(*cur);
+            }
+
+            return ret;
+        }
+    };
+
+public:
+    encode_view_() = default;
+    explicit encode_view_(Rng && rng) noexcept
+        : rng_{rng}
+    {}
+
+    inline decltype(auto) begin_cursor() const
+    {
+        auto beg = ranges::begin(rng_);
+        auto end = ranges::end(rng_);
+        return cursor<decltype(beg), decltype(end)>{beg, end};
     }
+
+public:
+private:
+    Rng rng_;
 };
 
 /**
- * UTF8->UTF32 decoding view
+ * UTF8->CodePoint decoding view
  *
  * Reads UTF8 and decodes it into codepoints (i.e. UTF32/UCS4). Effectively
  * used for conversion from qdb strings to numpy.
  */
 
-template <typename InCharT, typename Rng>
-    requires(ranges::sized_range<Rng>)
-class view_ : public ranges::view_facade<view_<InCharT, Rng>, ranges::finite>
+template <typename Rng>
+requires(ranges::forward_range<Rng>) class decode_view_
+    : public ranges::view_facade<decode_view_<Rng>, ranges::finite>
 {
     friend ranges::range_access;
     using iterator_t = ranges::iterator_t<Rng>;
-    using value_type = cp_type;
+    using value_type = detail::code_point;
+
+    /**
+     * Use a separate cursor struct, so we can implement the range as a forward range (i.e.
+     * make the iterators weakly comparable).
+     */
+    template <typename IterT, typename SentT>
+    struct cursor
+    {
+    private:
+        IterT iter_;
+        SentT last_;
+        decode_fn_<u8_type> decode_{};
+        skip_fn_<u8_type> skip_{};
+
+    public:
+        cursor() = default;
+        cursor(IterT iter, SentT last)
+            : iter_{iter}
+            , last_{last} {};
+
+        inline value_type read() const noexcept
+        {
+            // XXX(leon)
+            // This is somewhat inefficient, as we're peeking ahead trying to find the "next"
+            // codepoint, which may actually iterate forward more than one step.
+            //
+            // Then, when next() is requested, we _actually_ move ahead, and have to do the same
+            // (similar) checks again.
+            //
+            // However, after too much edge cases and debugging, I conclude that in order to be
+            // "really" compatible with ranges, you *must* separate these two, there's no way around
+            // it.
+            return decode_(iter_);
+        };
+
+        constexpr inline void next() noexcept
+        {
+            skip_(iter_);
+        };
+
+        constexpr inline bool equal(cursor<IterT, SentT> const & rhs) const noexcept
+        {
+            return iter_ == rhs.iter_;
+        }
+
+        constexpr inline bool equal(ranges::default_sentinel_t) const noexcept
+        {
+            return iter_ == last_;
+        }
+
+        inline ranges::iter_difference_t<iterator_t> distance_to(ranges::default_sentinel_t) const
+        {
+            return _distance_to(last_);
+        }
+
+    private:
+        template <typename T>
+        inline ranges::iter_difference_t<iterator_t> _distance_to(T const & other) const
+        {
+            ranges::iter_difference_t<iterator_t> ret{0};
+
+            for (IterT cur = iter_; cur != last_; skip_(cur))
+            {
+                ++ret;
+            }
+
+            return ret;
+        }
+    };
 
     // Lifecycle
 public:
-    view_() = default;
-    explicit view_(Rng && rng) noexcept
-        : rng_{rng}
-        , iter_{ranges::begin(rng)} {};
+    decode_view_() = default;
+    explicit decode_view_(Rng && rng) noexcept
+        : rng_{rng} {};
 
-    constexpr inline ranges::range_size_t<Rng> size() const noexcept
+    inline decltype(auto) begin_cursor() const
     {
-        return ranges::size(rng_);
+        auto beg = ranges::begin(rng_);
+        auto end = ranges::end(rng_);
+
+        return cursor<decltype(beg), decltype(end)>{beg, end};
     }
 
-    constexpr inline bool equal(ranges::default_sentinel_t) const noexcept
+private:
+    Rng rng_;
+};
+
+}; // namespace utf8
+
+}; // namespace detail
+
+namespace utf8
+{
+
+struct encode_view_base_fn
+{
+    template <typename Rng>
+    constexpr detail::utf8::encode_view_<Rng> operator()(Rng && rng) const
     {
-        return iter_ == ranges::end(rng_);
+        return detail::utf8::encode_view_<Rng>{std::forward<Rng>(rng)};
     }
+};
 
-public:
-    // Actual iterator / decoding logic
-    constexpr inline value_type read() const noexcept
+struct encode_view_bind_fn
+{
+    constexpr auto operator()() const
     {
-        // Because our iterator advances 1..4 positions for every read(), we cannot
-        // easily separate read() and next() without taking a performance hit.
-        //
-        // As such, we'll const_cast ourselves to make our internal position
-        // writable.
-        view_ * this_ = const_cast<view_ *>(this);
-        return this_->read();
-    };
+        return ranges::make_view_closure(encode_view_base_fn{});
+    }
+};
+
+struct encode_view_fn
+    : encode_view_base_fn
+    , encode_view_bind_fn
+{
+    using encode_view_base_fn::operator();
+    using encode_view_bind_fn::operator();
+};
 
-    constexpr inline value_type read() noexcept
+RANGES_INLINE_VARIABLE(encode_view_fn, encode_view)
+
+struct decode_view_base_fn
+{
+    /**
+     * Default case: our input range already has a 'perfect' u8_type (unsigned char)
+     */
+    template <typename Rng>
+    requires(concepts::range_t<Rng, u8_type>) constexpr detail::utf8::decode_view_<Rng> operator()(
+        Rng && rng) const
     {
-        return decode_(iter_);
-    };
+        return detail::utf8::decode_view_<Rng>{std::forward<Rng>(rng)};
+    }
+
+    /**
+     * Non-default case, which happens when e.g. using a std::string as a range: the underlying type is
+     * not exactly u8_type, but we can easily convert it to the appropriate type.
+     *
+     * This avoids a case where signed chars were interpreted as unsigned chars and all kinds of
+     * shenanigans, which is fixed here.
+     */
+    template <typename Rng>
+    requires(
+        !concepts::range_t<Rng,
+            u8_type> && std::is_nothrow_convertible_v<ranges::range_value_t<Rng>, u8_type>) constexpr decltype(auto)
+    operator()(Rng && rng) const
+    {
+        auto xform_fn = [](auto x) -> u8_type { return static_cast<u8_type>(x); };
 
-    constexpr inline void next() noexcept {};
+        // Transform and delegate to the 'regular' operator()
+        return operator()(ranges::views::transform(rng, xform_fn));
+    }
+};
 
-    // General plumbing
+struct decode_view_bind_fn
+{
+    constexpr auto operator()() const
+    {
+        return ranges::make_view_closure(decode_view_base_fn{});
+    }
+};
 
-private:
-    Rng rng_;
-    iterator_t iter_;
-    fn_<InCharT> decode_{};
+struct decode_view_fn
+    : decode_view_base_fn
+    , decode_view_bind_fn
+{
+    using decode_view_base_fn::operator();
+    using decode_view_bind_fn::operator();
 };
 
-template <typename InCharT, typename Rng>
-inline view_<InCharT, Rng> view(Rng && rng)
+RANGES_INLINE_VARIABLE(decode_view_fn, decode_view)
+
+}; // namespace utf8
+
+namespace utf32
+{
+
+/**
+ * Returns view that reads codepoints and emits UTF-32 characters.
+ */
+inline decltype(auto) encode_view()
 {
-    return view_<InCharT, Rng>{std::forward<Rng>(rng)};
+    auto xform_fn = [](detail::code_point x) -> u32_type { return static_cast<u32_type>(x.get()); };
+    return ranges::views::transform(xform_fn);
 }
 
+/**
+ * Returns view that reads codepoints and emits UTF-32 characters.
+ */
 template <class Rng>
-inline decltype(auto) utf8_view(Rng && rng)
+requires(concepts::input_range_t<Rng, detail::code_point>) inline decltype(auto) encode_view(Rng && rng)
+{
+    return rng | encode_view();
+}
+
+/**
+ * Returns view that reads UTF-32 characters and emits codepoints.
+ */
+inline decltype(auto) decode_view()
+{
+    auto xform_fn = [](u32_type x) -> detail::code_point { return detail::code_point{x}; };
+    return ranges::views::transform(xform_fn);
+}
+
+/**
+ * Returns view that reads UTF-32 characters and emits codepoints.
+ */
+template <typename Rng>
+inline decltype(auto) decode_view(Rng && rng)
 {
-    return view<u8_type>(std::forward<Rng>(rng));
+    return rng | decode_view();
 }
 
-}; // namespace decode
+}; // namespace utf32
 
 }; // namespace qdb::convert::unicode