weaseldb/src/format.hpp

#pragma once

#include <concepts>
#include <cstdarg>
#include <cstdio>
#include <cstring>
#include <string_view>
#include <type_traits>

#include "arena_allocator.hpp"

/**
 * @brief Runtime printf-style formatting with arena allocation optimization.
 *
 * This function provides familiar printf-style formatting with intelligent
 * optimization for arena allocation. It attempts single-pass formatting by
 * speculatively using available arena space, falling back to two-pass
 * formatting only when necessary.
 *
 * The function uses an optimized allocation strategy:
 * 1. **Single-pass attempt**: Try to format directly into available arena space
 * 2. **Fallback to two-pass**: If formatting doesn't fit, measure required size
 *    and allocate exactly what's needed
 *
 * ## Supported Format Specifiers:
 * All standard printf format specifiers are supported:
 * - **Integers**: %d, %i, %u, %x, %X, %o, %ld, %lld, etc.
 * - **Floating point**: %f, %e, %E, %g, %G, %.2f, etc.
 * - **Strings**: %s, %.*s, etc.
 * - **Characters**: %c
 * - **Pointers**: %p
 * - **Width/precision**: %10d, %-10s, %.2f, %*.*s, etc.
 *
 * ## Performance Characteristics:
 * - **Optimistic single-pass**: Often avoids the cost of measuring format size
 * - **Arena allocation**: Uses fast arena allocation (~1ns vs ~20-270ns for
 * malloc)
 * - **Memory efficient**: Returns unused space to arena via realloc()
 * - **Fallback safety**: Two-pass approach handles any format that doesn't fit
 *
 * @param arena Arena allocator for memory management
 * @param fmt Printf-style format string
 * @param ... Variable arguments matching format specifiers
 * @return std::string_view pointing to arena-allocated formatted string
 * @note Aborts program on formatting errors (never returns invalid data)
 * @note GCC format attribute enables compile-time format string validation
 *
 * ## Usage Examples:
 * ```cpp
 * ArenaAllocator arena(1024);
 *
 * // Basic formatting
 * auto msg = format(arena, "Hello %s!", "World");
 * // msg == "Hello World!"
 *
 * // Numeric formatting with precision
 * auto value = format(arena, "Pi: %.3f", 3.14159);
 * // value == "Pi: 3.142"
 *
 * // Mixed types with width/alignment
 * auto table = format(arena, "%-10s %5d %8.2f", "Item", 42, 99.95);
 * // table == "Item           42    99.95"
 *
 * // Error messages
 * auto error = format(arena, "Error %d: %s (line %d)", 404, "Not found", 123);
 * // error == "Error 404: Not found (line 123)"
 * ```
 *
 * ## When to Use:
 * - **Printf familiarity**: When you prefer printf-style format strings
 * - **Runtime flexibility**: Format strings from variables, config, or user
 * input
 * - **Complex formatting**: Precision, width, alignment, padding
 * - **Debugging**: Quick formatted output for logging/debugging
 * - **Mixed precision**: Different numeric precision requirements
 *
 * ## When to Use static_format() Instead:
 * - **Hot paths**: Performance-critical code where every nanosecond counts
 * - **Simple concatenation**: Basic string + number + string combinations
 * - **Compile-time optimization**: When all types/values known at compile time
 * - **Template contexts**: Where compile-time buffer sizing is beneficial
 *
 * ## Optimization Details:
 * The function uses `ArenaAllocator::allocate_remaining_space()` to claim all
 * available arena space and attempt formatting. If successful, it shrinks the
 * allocation to the actual size used. If formatting fails (doesn't fit), it
 * falls back to the traditional two-pass approach: measure size, allocate
 * exactly, then format.
 *
 * This strategy optimizes for the common case where available arena space is
 * sufficient, while maintaining correctness for all cases.
 */
std::string_view format(ArenaAllocator &arena, const char *fmt, ...)
    __attribute__((format(printf, 2, 3)));

namespace detail {

template <int kLen> struct StringTerm {
  explicit constexpr StringTerm(const char *s) : s(s) {}
  static constexpr int kMaxLength = kLen;
  void write(char *&buf) const {
    std::memcpy(buf, s, kLen);
    buf += kLen;
  }

private:
  const char *s;
};
template <int kLen>
constexpr StringTerm<kLen - 1> term(const char (&array)[kLen]) {
  return StringTerm<kLen - 1>{array};
}

template <class IntType> constexpr int decimal_length(IntType x) {
  static_assert(std::is_integral_v<IntType>,
                "decimal_length requires integral type");

  if constexpr (std::is_signed_v<IntType>) {
    // Handle negative values by using unsigned equivalent
    using Unsigned = std::make_unsigned_t<IntType>;
    // Safe conversion: cast to unsigned first, then negate in unsigned
    // arithmetic
    auto abs_x = x < 0 ? -static_cast<Unsigned>(x) : static_cast<Unsigned>(x);
    int result = 0;
    do {
      ++result;
      abs_x /= 10;
    } while (abs_x);
    return result;
  } else {
    int result = 0;
    do {
      ++result;
      x /= 10;
    } while (x);
    return result;
  }
}

template <std::integral IntType> struct IntTerm {
  static constexpr bool kSigned = std::is_signed_v<IntType>;
  using Unsigned = std::make_unsigned_t<IntType>;

  explicit constexpr IntTerm(IntType v) : v(v) {}

  static constexpr int kMaxLength =
      decimal_length(Unsigned(-1)) + (kSigned ? 1 : 0);

  void write(char *&buf) const {
    char itoa_buf[kMaxLength];
    auto x = static_cast<Unsigned>(v);

    if constexpr (kSigned) {
      if (v < 0) {
        *buf++ = '-';
        x = -static_cast<Unsigned>(v);
      }
    }

    int i = kMaxLength;
    do {
      itoa_buf[--i] = static_cast<char>('0' + (x % 10));
      x /= 10;
    } while (x);

    while (i < kMaxLength) {
      *buf++ = itoa_buf[i++];
    }
  }

private:
  IntType v;
};

template <std::integral IntType> constexpr IntTerm<IntType> term(IntType s) {
  return IntTerm<IntType>{s};
}

struct DoubleTerm {
  explicit constexpr DoubleTerm(double s) : s(s) {}
  static constexpr int kMaxLength = 24;
  void write(char *&buf) const;

private:
  double s;
};

// Variable template for compile-time max length access
template <typename T>
inline constexpr int max_decimal_length_v = decltype(term(T{}))::kMaxLength;

inline constexpr DoubleTerm term(double s) { return DoubleTerm(s); }

} // namespace detail

/**
 * @brief Compile-time optimized formatting for high-performance code paths.
 *
 * This function provides ultra-fast string formatting by calculating buffer
 * sizes at compile time and using specialized term handlers for each type.
 * It's designed for performance-critical code where formatting overhead
 * matters.
 *
 * Unlike the runtime `format()` function, `static_format()` processes all
 * arguments at compile time to determine exact memory requirements and uses
 * optimized term writers for maximum speed.
 *
 * ## Supported Types:
 * - **String literals**: C-style string literals and arrays
 * - **Integers**: All integral types (int, int64_t, uint32_t, etc.)
 * - **Floating point**: double (uses high-precision Grisu2 algorithm)
 * - **Custom types**: Via specialization of `detail::term()`
 *
 * ## Performance Characteristics:
 * - **Compile-time buffer sizing**: Buffer size calculated at compile time (no
 * runtime measurement)
 * - **Optimized arena allocation**: Uses pre-calculated exact buffer sizes with
 * arena allocator
 * - **Specialized type handling**: Fast paths for common types via template
 * specialization
 * - **Memory efficient**: Uses arena.realloc() to return unused space to the
 * arena
 *
 * @tparam Ts Types of the arguments to format (auto-deduced)
 * @param arena Arena allocator for memory management
 * @param ts Arguments to format - can be string literals, integers, doubles
 * @return std::string_view pointing to arena-allocated formatted string
 *
 * ## Usage Examples:
 * ```cpp
 * ArenaAllocator arena(1024);
 *
 * // String concatenation
 * auto result1 = static_format(arena, "Hello ", "World", "!");
 * // result1 == "Hello World!"
 *
 * // Mixed types
 * auto result2 = static_format(arena, "Count: ", 42, ", Rate: ", 3.14);
 * // result2 == "Count: 42, Rate: 3.14"
 *
 * // Error messages
 * auto error = static_format(arena, "Error ", 404, ": ", "Not found");
 * // error == "Error 404: Not found"
 * ```
 *
 * ## When to Use:
 * - **Hot paths**: Performance-critical code where formatting speed matters
 * - **Known types**: When argument types are known at compile time
 * - **Simple formatting**: Concatenation and basic type conversion
 * - **Template code**: Where compile-time optimization is beneficial
 *
 * ## When to Use format() Instead:
 * - **Printf-style formatting**: When you need format specifiers like "%d",
 * "%.2f"
 * - **Runtime flexibility**: When format strings come from variables/config
 * - **Complex formatting**: When you need padding, precision, etc.
 * - **Convenience**: For quick debugging or non-critical paths
 *
 * @note All arguments are passed by forwarding reference for optimal
 * performance
 * @note Memory is arena-allocated and automatically sized to exact requirements
 * @note Compile-time errors occur if unsupported types are used
 * @note This function is constexpr-friendly and optimizes well in release
 * builds
 */
template <class... Ts>
std::string_view static_format(ArenaAllocator &arena, Ts &&...ts) {
  constexpr int upper_bound = (decltype(detail::term(ts))::kMaxLength + ...);
  char *result = arena.allocate<char>(upper_bound);
  char *buf = result;
  (detail::term(ts).write(buf), ...);
  const int size = static_cast<int>(buf - result);
  return std::string_view(
      arena.realloc(result, upper_bound, upper_bound - size),
      static_cast<std::size_t>(size));
}