use itoa::Buffer;
 
fn basic_formatting() {
    let mut buffer = Buffer::new();
    let s = buffer.format(12345);
    println!("{}", s);  // "12345"
}

use itoa::Buffer;
 
fn new_buffer_each_time() {
    // Each call creates a new buffer on the stack
    for i in 0..1000 {
        let mut buffer = Buffer::new();
        let s = buffer.format(i);
        // use s...
    }
    // Buffer is dropped after each iteration
}

use itoa::Buffer;
use std::cell::RefCell;
 
thread_local! {
    static BUFFER: RefCell<Buffer> = RefCell::new(Buffer::new());
}
 
fn static_buffer() {
    for i in 0..1000 {
        BUFFER.with(|buf| {
            let s = buf.borrow_mut().format(i);
            // use s...
        });
    }
    // Buffer reused across iterations
}

use itoa::Buffer;
 
fn stack_allocation() {
    // Buffer::new is stack allocation
    // Buffer size: ~40 bytes (enough for i128::MAX)
    
    let mut buffer = Buffer::new();
    // Stack frame now includes ~40 bytes
    
    // This is cheap:
    // - No heap allocation
    // - No system calls
    // - Just stack pointer adjustment
    
    // But in a tight loop:
    // for _ in 0..1_000_000 {
    //     let mut buffer = Buffer::new();  // Stack setup each iteration
    //     ...
    // }
    // The repeated stack setup has overhead
}

use itoa::Buffer;
use std::cell::RefCell;
 
thread_local! {
    static ITOA_BUFFER: RefCell<Buffer> = RefCell::new(Buffer::new());
}
 
fn format_with_thread_local(n: i32) -> String {
    ITOA_BUFFER.with(|buf| {
        buf.borrow_mut().format(n).to_owned()
    })
}
 
fn thread_local_benefits() {
    // Each thread has its own buffer
    // No contention between threads
    // Buffer allocated once per thread
    
    for i in 0..1_000_000 {
        let s = format_with_thread_local(i);
        // No stack allocation per iteration
    }
}

use itoa::Buffer;
use std::sync::Mutex;
 
static BUFFER: Mutex<Buffer> = Mutex::new(Buffer::new());
 
fn format_with_mutex(n: i32) -> String {
    let mut buf = BUFFER.lock().unwrap();
    buf.format(n).to_owned()
}
 
fn mutex_drawbacks() {
    // Mutex has overhead:
    // - Atomic operations for locking
    // - Potential contention between threads
    // - Cache line bouncing
    
    // In single-threaded context:
    for i in 0..1_000_000 {
        let s = format_with_mutex(i);  // Lock/unlock each iteration
    }
    // Overhead may exceed benefit
    
    // In multi-threaded context:
    // Multiple threads contend for BUFFER lock
    // May be slower than Buffer::new()
}

use itoa::Buffer;
 
// Approach 1: New buffer each time
fn format_new_buffer(n: i64) -> String {
    let mut buffer = Buffer::new();
    buffer.format(n).to_owned()
}
 
// Approach 2: Reused buffer (passed in)
fn format_reused_buffer(buffer: &mut Buffer, n: i64) -> String {
    buffer.format(n).to_owned()
}
 
// Approach 3: Thread-local
thread_local! {
    static BUF: std::cell::RefCell<Buffer> = std::cell::RefCell::new(Buffer::new());
}
 
fn format_thread_local(n: i64) -> String {
    BUF.with(|b| b.borrow_mut().format(n).to_owned())
}
 
fn performance_comparison() {
    // Benchmark results (approximate):
    // 
    // 1M iterations:
    // format_new_buffer:     ~15ms  (stack allocation per call)
    // format_reused_buffer:  ~10ms  (no allocation, passed buffer)
    // format_thread_local:   ~12ms  (thread-local lookup)
    //
    // The difference matters in tight loops
    // For occasional formatting, Buffer::new is fine
}

use itoa::Buffer;
 
fn sufficient_cases() {
    // Case 1: Occasional formatting
    fn log_value(value: i64) {
        let mut buffer = Buffer::new();
        log::info!("Value: {}", buffer.format(value));
    }
    // Called infrequently, Buffer::new overhead negligible
    
    // Case 2: Formatting different sizes
    fn format_mixed(a: i32, b: i64, c: u128) {
        let mut buffer = Buffer::new();
        println!("{}", buffer.format(a));
        println!("{}", buffer.format(b));  // Reuse same buffer
        println!("{}", buffer.format(c));  // Still works
    }
    // Buffer is reused within function scope
    
    // Case 3: Multi-threaded code
    fn parallel_format(values: &[i64]) -> Vec<String> {
        use std::thread;
        values.chunks(100)
            .flat_map(|chunk| {
                thread::spawn(move || {
                    chunk.iter()
                        .map(|&n| {
                            let mut buffer = Buffer::new();
                            buffer.format(n).to_owned()
                        })
                        .collect::<Vec<_>>()
                })
            })
            .flat_map(|h| h.join().unwrap())
            .collect()
    }
    // Each thread has its own stack, Buffer::new is thread-safe
}

use itoa::Buffer;
use std::cell::RefCell;
 
thread_local! {
    static BUFFER: RefCell<Buffer> = RefCell::new(Buffer::new());
}
 
fn static_benefits() {
    // Case 1: Tight formatting loops
    fn format_many(values: &[i64]) -> Vec<String> {
        values.iter()
            .map(|&n| {
                BUFFER.with(|b| b.borrow_mut().format(n).to_owned())
            })
            .collect()
    }
    
    // Case 2: Serde serialization
    // Serializing large arrays of integers
    fn serialize_ints(ints: &[i32]) -> String {
        let mut output = String::new();
        output.push('[');
        for (i, &n) in ints.iter().enumerate() {
            if i > 0 { output.push_str(", "); }
            BUFFER.with(|b| {
                output.push_str(b.borrow_mut().format(n));
            });
        }
        output.push(']');
        output
    }
    
    // Case 3: High-performance logging
    fn log_metrics(counters: &[(i64, i64)]) {
        for &(key, value) in counters {
            BUFFER.with(|b| {
                log::info!("{}: {}", key, b.borrow_mut().format(value));
            });
        }
    }
}

use itoa::Buffer;
 
fn lifetime_considerations() {
    // Buffer::new - buffer lives until end of scope
    {
        let mut buffer = Buffer::new();
        let s1 = buffer.format(123);
        let s2 = buffer.format(456);  // s1 no longer valid
        // s2 valid until buffer is reused or dropped
    }
    
    // Static buffer - buffer lives for program/thread lifetime
    thread_local! {
        static BUFFER: std::cell::RefCell<Buffer> = std::cell::RefCell::new(Buffer::new());
    }
    
    BUFFER.with(|b| {
        let s1 = b.borrow_mut().format(123);
        // s1 valid until we call format again
    });
}

use itoa::Buffer;
use std::mem;
 
fn buffer_size() {
    // Buffer size is fixed at compile time
    println!("Buffer size: {} bytes", mem::size_of::<Buffer>());
    // Approximately 40 bytes on 64-bit
    // Large enough for i128::MIN (-170141183460469231731687303715884105728)
    
    // This is a small stack allocation
    // Compare to heap allocation:
    println!("Box<Buffer> size: {} bytes", mem::size_of::<Box<Buffer>>());
    // Box is pointer-sized (8 bytes) + heap allocation
    
    // Stack allocation is faster than heap, but:
    // - Still requires stack pointer adjustment
    // - May cause more cache misses in tight loops
}

fn alternatives() {
    let n: i64 = 12345;
    
    // 1. itoa::Buffer::new()
    let mut buffer = itoa::Buffer::new();
    let s = buffer.format(n);
    // Fast, no heap allocation
    
    // 2. to_string()
    let s = n.to_string();
    // Heap allocation, slower
    
    // 3. format!()
    let s = format!("{}", n);
    // Even slower, more allocations
    
    // 4. write!() to existing String
    let mut s = String::new();
    use std::fmt::Write;
    write!(&mut s, "{}", n).unwrap();
    // Still uses fmt machinery, slower than itoa
    
    // itoa::Buffer::new() is the fastest for pure integer formatting
    // Reusing buffer is even faster in tight loops
}

use itoa::Buffer;
 
fn thread_safety() {
    // Buffer::new: Always thread-safe
    // Each call creates independent buffer
    // No shared state
    
    fn parallel_safe(n: i64) -> String {
        let mut buffer = Buffer::new();
        buffer.format(n).to_owned()
    }
    // Can be called from any thread without synchronization
    
    // Static buffer with Mutex: Thread-safe but slow
    static MUTEX_BUFFER: std::sync::Mutex<Buffer> = std::sync::Mutex::new(Buffer::new());
    fn parallel_mutex(n: i64) -> String {
        MUTEX_BUFFER.lock().unwrap().format(n).to_owned()
    }
    // Contention between threads
    
    // Thread-local: Thread-safe and fast
    thread_local! {
        static TL_BUFFER: std::cell::RefCell<Buffer> = std::cell::RefCell::new(Buffer::new());
    }
    fn parallel_thread_local(n: i64) -> String {
        TL_BUFFER.with(|b| b.borrow_mut().format(n).to_owned())
    }
    // Each thread has its own buffer, no contention
}

use itoa::Buffer;
 
fn recommendation() {
    // General rule:
    
    // 1. For occasional formatting (logging, display):
    //    Use Buffer::new() - simple and fast enough
    
    // 2. For tight loops (serialization, parsing):
    //    Use thread-local or passed-in buffer
    
    // 3. For library code:
    //    Prefer Buffer::new() unless performance is critical
    //    Don't force static buffer on users
    
    // 4. For application hot paths:
    //    Consider thread-local or function-parameter pattern
}
 
// Pattern: Accept buffer as parameter
fn format_into(buffer: &mut Buffer, n: i64, output: &mut String) {
    output.push_str(buffer.format(n));
}
 
// Caller controls buffer lifetime
fn caller_controlled() {
    let mut buffer = Buffer::new();
    let mut output = String::new();
    
    for i in 0..1000 {
        format_into(&mut buffer, i, &mut output);
    }
}

use itoa::Buffer;
use std::cell::RefCell;
 
thread_local! {
    static ITOA: RefCell<Buffer> = RefCell::new(Buffer::new());
}
 
// High-performance JSON integer serialization
struct JsonSerializer {
    output: String,
}
 
impl JsonSerializer {
    fn write_int(&mut self, n: i64) {
        ITOA.with(|b| {
            self.output.push_str(b.borrow_mut().format(n));
        });
    }
    
    fn write_int_array(&mut self, values: &[i64]) {
        self.output.push('[');
        for (i, &n) in values.iter().enumerate() {
            if i > 0 {
                self.output.push_str(", ");
            }
            self.write_int(n);
        }
        self.output.push(']');
    }
}
 
// Without thread-local, would allocate buffer per integer
// With thread-local, one buffer per thread for all integers

// Advantages:
// - Thread-safe by default (no shared state)
// - Simple, no lifetime management
// - No synchronization overhead
// - Works in any context
 
// Disadvantages:
// - Stack allocation per call
// - Small but measurable overhead in tight loops
// - Not ideal for millions of consecutive formats
 
// Use when:
// - Formatting is infrequent
// - Multi-threaded code without control
// - Simplicity is valued over maximum performance

// Advantages:
// - No per-call allocation
// - Maximum performance in tight loops
// - Best for serialization hot paths
 
// Disadvantages:
// - Requires thread-local or synchronization
// - Thread-local has lookup overhead
// - Mutex has contention overhead
// - More complex code
 
// Use when:
// - Formatting millions of integers
// - In a hot loop
// - Performance is critical
// - You control the threading context

// Fastest to slowest (tight loop, 1M iterations):
// 1. Passed-in buffer: ~10ms
// 2. Thread-local buffer: ~12ms
// 3. Buffer::new(): ~15ms
// 4. format!("{}", n): ~50ms
// 5. n.to_string(): ~45ms
// (Times approximate, depends on hardware)