lib: introduce ByteVector as a replacement for vector<uint8_t> and Spanner

After some benchmarking, it turns out that std::vector<uint8_t> is
about 160 times slower than malloc().  malloc() is faster than "new
uint8_t[]" too.  Get rid of std:;vector<uint8_t> and replace it with
a lightweight wrapper around malloc(), free(), and memcpy().

ByteVector has helpful methods for the common case of moving data to and
from ioctl calls that use a fixed-length header placed contiguously with a
variable-length input/output buffer.  Data bytes are shared between copied
ByteVector objects, allowing a large single buffer to be cheaply chopped
up into smaller objects without memory copies.  ByteVector implements the
more useful parts of the std::vector API, so it can replace std::vector
objects without needing an awkward adaptor class like Spanner.

Signed-off-by: Zygo Blaxell <bees@furryterror.org>

include/crucible/bytevector.h

@@ -0,0 +1,71 @@
+#ifndef _CRUCIBLE_BYTEVECTOR_H_
+#define _CRUCIBLE_BYTEVECTOR_H_
+
+#include <memory>
+
+#include <cstdint>
+#include <cstdlib>
+
+namespace crucible {
+	using namespace std;
+	// new[] is a little slower than malloc
+	// shared_ptr is about 2x slower than unique_ptr
+	// vector<uint8_t> is ~160x slower
+	// so we won't bother with unique_ptr because we can't do shared copies with it
+
+	class ByteVector {
+	public:
+		using Pointer = shared_ptr<uint8_t>;
+		using value_type = Pointer::element_type;
+		using iterator = value_type*;
+
+		ByteVector() = default;
+		ByteVector(size_t size);
+		ByteVector(const ByteVector &that, size_t start, size_t length);
+		ByteVector(iterator begin, iterator end, size_t min_size = 0);
+
+		ByteVector at(size_t start, size_t length) const;
+
+		value_type& at(size_t) const;
+		iterator begin() const;
+		void clear();
+		value_type* data() const;
+		bool empty() const;
+		iterator end() const;
+		value_type& operator[](size_t) const;
+		size_t size() const;
+		bool operator==(const ByteVector &that) const;
+
+		// this version of erase only works at the beginning or end of the buffer, else throws exception
+		void erase(iterator first);
+		void erase(iterator first, iterator last);
+
+		// An important use case is ioctls that have a fixed-size header struct
+		// followed by a buffer for further arguments.  These templates avoid
+		// doing reinterpret_casts every time.
+		template <class T> ByteVector(const T& object, size_t min_size);
+		template <class T> T* get() const;
+	private:
+		Pointer m_ptr;
+		size_t m_size = 0;
+	};
+
+	template <class T>
+	ByteVector::ByteVector(const T& object, size_t min_size)
+	{
+		const auto size = max(min_size, sizeof(T));
+		m_ptr = Pointer(static_cast<value_type*>(malloc(size)), free);
+		memcpy(m_ptr.get(), &object, sizeof(T));
+		m_size = size;
+	}
+
+	template <class T>
+	T*
+	ByteVector::get() const
+	{
+		return reinterpret_cast<T*>(data());
+	}
+
+}
+
+#endif // _CRUCIBLE_BYTEVECTOR_H_

lib/Makefile

@@ -4,6 +4,7 @@ default: libcrucible.a
 %.a: Makefile
 
 CRUCIBLE_OBJS = \
+	bytevector.o \
 	chatter.o \
 	city.o \
 	cleanup.o \

lib/bytevector.cc

@@ -0,0 +1,135 @@
+#include "crucible/bytevector.h"
+
+#include "crucible/error.h"
+
+namespace crucible {
+	using namespace std;
+
+	ByteVector::iterator
+	ByteVector::begin() const
+	{
+		return m_ptr.get();
+	}
+
+	ByteVector::iterator
+	ByteVector::end() const
+	{
+		return m_ptr.get() + m_size;
+	}
+
+	size_t
+	ByteVector::size() const
+	{
+		return m_size;
+	}
+
+	bool
+	ByteVector::empty() const
+	{
+		return !m_ptr || !m_size;
+	}
+
+	void
+	ByteVector::clear()
+	{
+		m_ptr.reset();
+		m_size = 0;
+	}
+
+	ByteVector::value_type&
+	ByteVector::operator[](size_t size) const
+	{
+		return m_ptr.get()[size];
+	}
+
+	ByteVector::ByteVector(const ByteVector &that, size_t start, size_t length)
+	{
+		THROW_CHECK0(out_of_range, that.m_ptr);
+		THROW_CHECK2(out_of_range, start, that.m_size, start < that.m_size);
+		THROW_CHECK2(out_of_range, start + length, that.m_size + length, start + length < that.m_size + length);
+		m_ptr = Pointer(that.m_ptr, that.m_ptr.get() + start);
+		m_size = length;
+	}
+
+	ByteVector
+	ByteVector::at(size_t start, size_t length) const
+	{
+		return ByteVector(*this, start, length);
+	}
+
+	ByteVector::value_type&
+	ByteVector::at(size_t size) const
+	{
+		THROW_CHECK0(out_of_range, m_ptr);
+		THROW_CHECK2(out_of_range, size, m_size, size < m_size);
+		return m_ptr.get()[size];
+	}
+
+	ByteVector::ByteVector(size_t size)
+	{
+		m_ptr = Pointer(static_cast<value_type*>(malloc(size)), free);
+		// bad_alloc doesn't fit THROW_CHECK's template
+		THROW_CHECK0(runtime_error, m_ptr);
+		m_size = size;
+	}
+
+	ByteVector::ByteVector(iterator begin, iterator end, size_t min_size)
+	{
+		const size_t size = end - begin;
+		const size_t alloc_size = max(size, min_size);
+		m_ptr = Pointer(static_cast<value_type*>(malloc(alloc_size)), free);
+		THROW_CHECK0(runtime_error, m_ptr);
+		m_size = alloc_size;
+		memcpy(m_ptr.get(), begin, size);
+	}
+
+	bool
+	ByteVector::operator==(const ByteVector &that) const
+	{
+		if (!m_ptr) {
+			return !that.m_ptr;
+		}
+		if (!that.m_ptr) {
+			return false;
+		}
+		if (m_size != that.m_size) {
+			return false;
+		}
+		if (m_ptr.get() == that.m_ptr.get()) {
+			return true;
+		}
+		return !memcmp(m_ptr.get(), that.m_ptr.get(), m_size);
+	}
+
+	void
+	ByteVector::erase(iterator begin, iterator end)
+	{
+		const size_t size = end - begin;
+		if (!size) return;
+		THROW_CHECK0(out_of_range, m_ptr);
+		const iterator my_begin = m_ptr.get();
+		const iterator my_end = my_begin + m_size;
+		THROW_CHECK4(out_of_range, my_begin, begin, my_end, end, my_begin == begin || my_end == end);
+		if (begin == my_begin) {
+			if (end == my_end) {
+				m_size = 0;
+				m_ptr.reset();
+				return;
+			}
+			m_ptr = Pointer(m_ptr, end);
+		}
+		m_size -= size;
+	}
+
+	void
+	ByteVector::erase(iterator begin)
+	{
+		erase(begin, begin + 1);
+	}
+
+	ByteVector::value_type*
+	ByteVector::data() const
+	{
+		return m_ptr.get();
+	}
+}