seq_runtime/

io.rs

//! I/O Operations for Seq
//!
//! These functions are exported with C ABI for LLVM codegen to call.
//!
//! # Safety Contract
//!
//! **IMPORTANT:** These functions are designed to be called ONLY by compiler-generated code,
//! not by end users or arbitrary C code. The compiler is responsible for:
//!
//! - Ensuring stack has correct types (verified by type checker)
//! - Passing valid, null-terminated C strings to `push_string`
//! - Never calling these functions directly from user code
//!
//! # String Handling
//!
//! String literals from the compiler must be valid UTF-8 C strings (null-terminated).
//! Currently, each string literal is allocated as an owned `String`. See
//! `docs/STRING_INTERNING_DESIGN.md` for discussion of future optimizations
//! (interning, static references, etc.).

use crate::stack::{Stack, pop, push};
use crate::value::Value;
use std::ffi::CStr;
use std::io;
use std::sync::LazyLock;

/// Coroutine-aware stdout mutex.
/// Uses may::sync::Mutex which yields the coroutine when contended instead of blocking the OS thread.
/// By serializing access to stdout, we prevent RefCell borrow panics that occur when multiple
/// coroutines on the same thread try to access stdout's internal RefCell concurrently.
static STDOUT_MUTEX: LazyLock<may::sync::Mutex<()>> = LazyLock::new(|| may::sync::Mutex::new(()));

/// Valid exit code range for Unix compatibility
const EXIT_CODE_MIN: i64 = 0;
const EXIT_CODE_MAX: i64 = 255;

/// Write a string to stdout followed by a newline
///
/// Stack effect: ( str -- )
///
/// # Safety
/// Stack must have a String value on top
///
/// # Concurrency
/// Uses may::sync::Mutex to serialize stdout writes from multiple strands.
/// When the mutex is contended, the strand yields to the scheduler (doesn't block the OS thread).
/// This prevents RefCell borrow panics when multiple strands write concurrently.
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_write_line(stack: Stack) -> Stack {
    assert!(!stack.is_null(), "write_line: stack is empty");

    let (rest, value) = unsafe { pop(stack) };

    match value {
        Value::String(s) => {
            // Acquire coroutine-aware mutex (yields if contended, doesn't block)
            // This serializes access to stdout
            let _guard = STDOUT_MUTEX.lock().unwrap();

            // Write directly to fd 1 using libc to avoid Rust's std::io::stdout() RefCell.
            // Rust's standard I/O uses RefCell which panics on concurrent access from
            // multiple coroutines on the same thread.
            let str_slice = s.as_str();
            let newline = b"\n";
            unsafe {
                libc::write(
                    1,
                    str_slice.as_ptr() as *const libc::c_void,
                    str_slice.len(),
                );
                libc::write(1, newline.as_ptr() as *const libc::c_void, newline.len());
            }

            rest
        }
        _ => panic!("write_line: expected String on stack, got {:?}", value),
    }
}

/// Read a line from stdin
///
/// Returns the line including trailing newline.
/// Returns empty string "" at EOF.
/// Use `string-chomp` to remove trailing newlines if needed.
///
/// # Line Ending Normalization
///
/// Line endings are normalized to `\n` regardless of platform. Windows-style
/// `\r\n` endings are converted to `\n`. This ensures consistent behavior
/// across different operating systems.
///
/// Stack effect: ( -- str )
///
/// # Safety
/// Always safe to call
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_read_line(stack: Stack) -> Stack {
    use std::io::BufRead;

    let stdin = io::stdin();
    let mut line = String::new();

    stdin
        .lock()
        .read_line(&mut line)
        .expect("read_line: failed to read from stdin (I/O error or EOF)");

    // Normalize line endings: \r\n -> \n
    if line.ends_with("\r\n") {
        line.pop(); // remove \n
        line.pop(); // remove \r
        line.push('\n'); // add back \n
    }

    unsafe { push(stack, Value::String(line.into())) }
}

/// Read a line from stdin with explicit EOF detection
///
/// Returns the line and a status flag:
/// - ( line 1 ) on success (line includes trailing newline)
/// - ( "" 0 ) at EOF
///
/// Stack effect: ( -- String Int )
///
/// The `+` suffix indicates this returns a result pattern (value + status).
///
/// # Line Ending Normalization
///
/// Line endings are normalized to `\n` regardless of platform. Windows-style
/// `\r\n` endings are converted to `\n`. This ensures consistent behavior
/// across different operating systems.
///
/// # Safety
/// Always safe to call
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_read_line_plus(stack: Stack) -> Stack {
    use std::io::BufRead;

    let stdin = io::stdin();
    let mut line = String::new();

    let bytes_read = stdin
        .lock()
        .read_line(&mut line)
        .expect("read_line_safe: failed to read from stdin");

    // Normalize line endings: \r\n -> \n
    if line.ends_with("\r\n") {
        line.pop(); // remove \n
        line.pop(); // remove \r
        line.push('\n'); // add back \n
    }

    // bytes_read == 0 means EOF
    let status = if bytes_read > 0 { 1i64 } else { 0i64 };

    let stack = unsafe { push(stack, Value::String(line.into())) };
    unsafe { push(stack, Value::Int(status)) }
}

/// Maximum bytes allowed for a single read_n call (10MB)
/// This prevents accidental or malicious massive memory allocations.
/// LSP messages are typically < 1MB, so 10MB provides generous headroom.
const READ_N_MAX_BYTES: i64 = 10 * 1024 * 1024;

/// Validates and extracts the byte count from a Value for read_n.
/// Returns Ok(usize) on success, Err(message) on validation failure.
fn validate_read_n_count(value: &Value) -> Result<usize, String> {
    match value {
        Value::Int(n) if *n < 0 => Err(format!(
            "read_n: byte count must be non-negative, got {}",
            n
        )),
        Value::Int(n) if *n > READ_N_MAX_BYTES => Err(format!(
            "read_n: byte count {} exceeds maximum allowed ({})",
            n, READ_N_MAX_BYTES
        )),
        Value::Int(n) => Ok(*n as usize),
        _ => Err(format!("read_n: expected Int on stack, got {:?}", value)),
    }
}

/// Read exactly N bytes from stdin
///
/// Returns the bytes read and a status flag:
/// - ( string 1 ) on success (read all N bytes)
/// - ( string 0 ) at EOF or partial read (string may be shorter than N)
///
/// Stack effect: ( Int -- String Int )
///
/// Like `io.read-line+`, this returns a result pattern (value + status) to allow
/// explicit EOF detection. The function name omits the `+` suffix for brevity
/// since byte-count reads are inherently status-oriented.
///
/// This is used for protocols like LSP where message bodies are byte-counted
/// and don't have trailing newlines.
///
/// # UTF-8 Handling
/// The bytes are interpreted as UTF-8. Invalid UTF-8 sequences are replaced
/// with the Unicode replacement character (U+FFFD). This is appropriate for
/// text-based protocols like LSP but may not be suitable for binary data.
///
/// # Safety
/// Stack must have an Int on top. The integer must be non-negative and
/// not exceed READ_N_MAX_BYTES (10MB).
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_read_n(stack: Stack) -> Stack {
    use std::io::Read;

    assert!(!stack.is_null(), "read_n: stack is empty");

    let (stack, value) = unsafe { pop(stack) };
    let n = validate_read_n_count(&value).unwrap_or_else(|e| panic!("{}", e));

    let stdin = io::stdin();
    let mut buffer = vec![0u8; n];
    let mut total_read = 0;

    {
        let mut handle = stdin.lock();
        while total_read < n {
            match handle.read(&mut buffer[total_read..]) {
                Ok(0) => break, // EOF
                Ok(bytes_read) => total_read += bytes_read,
                Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
                Err(e) => panic!("read_n: failed to read from stdin: {}", e),
            }
        }
    }

    // Truncate to actual bytes read
    buffer.truncate(total_read);

    // Convert to String (assuming UTF-8)
    let s = String::from_utf8_lossy(&buffer).into_owned();

    // Status: 1 if we read all N bytes, 0 otherwise
    let status = if total_read == n { 1i64 } else { 0i64 };

    let stack = unsafe { push(stack, Value::String(s.into())) };
    unsafe { push(stack, Value::Int(status)) }
}

/// Convert an integer to a string
///
/// Stack effect: ( Int -- String )
///
/// # Safety
/// Stack must have an Int value on top
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_int_to_string(stack: Stack) -> Stack {
    assert!(!stack.is_null(), "int_to_string: stack is empty");

    let (rest, value) = unsafe { pop(stack) };

    match value {
        Value::Int(n) => unsafe { push(rest, Value::String(n.to_string().into())) },
        _ => panic!("int_to_string: expected Int on stack, got {:?}", value),
    }
}

/// Push a C string literal onto the stack (for compiler-generated code)
///
/// Stack effect: ( -- str )
///
/// # Safety
/// The c_str pointer must be valid and null-terminated
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_push_string(stack: Stack, c_str: *const i8) -> Stack {
    assert!(!c_str.is_null(), "push_string: null string pointer");

    let s = unsafe {
        CStr::from_ptr(c_str)
            .to_str()
            .expect("push_string: invalid UTF-8 in string literal")
            .to_owned()
    };

    unsafe { push(stack, Value::String(s.into())) }
}

/// Push a SeqString value onto the stack
///
/// This is used when we already have a SeqString (e.g., from closures).
/// Unlike push_string which takes a C string, this takes a SeqString by value.
///
/// Stack effect: ( -- String )
///
/// # Safety
/// The SeqString must be valid. This is only called from LLVM-generated code, not actual C code.
#[allow(improper_ctypes_definitions)]
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_push_seqstring(
    stack: Stack,
    seq_str: crate::seqstring::SeqString,
) -> Stack {
    unsafe { push(stack, Value::String(seq_str)) }
}

/// Exit the program with a status code
///
/// Stack effect: ( exit_code -- )
///
/// # Safety
/// Stack must have an Int on top. Never returns.
#[unsafe(no_mangle)]
pub unsafe extern "C" fn patch_seq_exit_op(stack: Stack) -> ! {
    assert!(!stack.is_null(), "exit_op: stack is empty");

    let (_rest, value) = unsafe { pop(stack) };

    match value {
        Value::Int(code) => {
            // Explicitly validate exit code is in Unix-compatible range
            if !(EXIT_CODE_MIN..=EXIT_CODE_MAX).contains(&code) {
                panic!(
                    "exit_op: exit code must be in range {}-{}, got {}",
                    EXIT_CODE_MIN, EXIT_CODE_MAX, code
                );
            }
            std::process::exit(code as i32);
        }
        _ => panic!("exit_op: expected Int on stack, got {:?}", value),
    }
}

// Public re-exports with short names for internal use
pub use patch_seq_exit_op as exit_op;
pub use patch_seq_int_to_string as int_to_string;
pub use patch_seq_push_seqstring as push_seqstring;
pub use patch_seq_push_string as push_string;
pub use patch_seq_read_line as read_line;
pub use patch_seq_read_line_plus as read_line_plus;
pub use patch_seq_read_n as read_n;
pub use patch_seq_write_line as write_line;

#[cfg(test)]
mod tests {
    use super::*;
    use crate::value::Value;
    use std::ffi::CString;

    #[test]
    fn test_write_line() {
        unsafe {
            let stack = crate::stack::alloc_test_stack();
            let stack = push(stack, Value::String("Hello, World!".into()));
            let _stack = write_line(stack);
        }
    }

    #[test]
    fn test_push_string() {
        unsafe {
            let stack = crate::stack::alloc_test_stack();
            let test_str = CString::new("Test").unwrap();
            let stack = push_string(stack, test_str.as_ptr());

            let (_stack, value) = pop(stack);
            assert_eq!(value, Value::String("Test".into()));
        }
    }

    #[test]
    fn test_empty_string() {
        unsafe {
            // Empty string should be handled correctly
            let stack = crate::stack::alloc_test_stack();
            let empty_str = CString::new("").unwrap();
            let stack = push_string(stack, empty_str.as_ptr());

            let (_stack, value) = pop(stack);
            assert_eq!(value, Value::String("".into()));

            // Write empty string should work without panic
            let stack = push(stack, Value::String("".into()));
            let _stack = write_line(stack);
        }
    }

    #[test]
    fn test_unicode_strings() {
        unsafe {
            // Test that Unicode strings are handled correctly
            let stack = crate::stack::alloc_test_stack();
            let unicode_str = CString::new("Hello, 世界! 🌍").unwrap();
            let stack = push_string(stack, unicode_str.as_ptr());

            let (_stack, value) = pop(stack);
            assert_eq!(value, Value::String("Hello, 世界! 🌍".into()));
        }
    }

    // =========================================================================
    // read_n validation tests
    // =========================================================================

    #[test]
    fn test_read_n_valid_input() {
        assert_eq!(super::validate_read_n_count(&Value::Int(0)), Ok(0));
        assert_eq!(super::validate_read_n_count(&Value::Int(100)), Ok(100));
        assert_eq!(
            super::validate_read_n_count(&Value::Int(1024 * 1024)), // 1MB
            Ok(1024 * 1024)
        );
    }

    #[test]
    fn test_read_n_negative_input() {
        let result = super::validate_read_n_count(&Value::Int(-1));
        assert!(result.is_err());
        assert!(result.unwrap_err().contains("must be non-negative"));
    }

    #[test]
    fn test_read_n_large_negative_input() {
        let result = super::validate_read_n_count(&Value::Int(i64::MIN));
        assert!(result.is_err());
        assert!(result.unwrap_err().contains("must be non-negative"));
    }

    #[test]
    fn test_read_n_exceeds_max_bytes() {
        let result = super::validate_read_n_count(&Value::Int(super::READ_N_MAX_BYTES + 1));
        assert!(result.is_err());
        assert!(result.unwrap_err().contains("exceeds maximum allowed"));
    }

    #[test]
    fn test_read_n_at_max_bytes_ok() {
        // Exactly at the limit should be OK
        let result = super::validate_read_n_count(&Value::Int(super::READ_N_MAX_BYTES));
        assert_eq!(result, Ok(super::READ_N_MAX_BYTES as usize));
    }

    #[test]
    fn test_read_n_wrong_type_string() {
        let result = super::validate_read_n_count(&Value::String("not an int".into()));
        assert!(result.is_err());
        assert!(result.unwrap_err().contains("expected Int"));
    }

    #[test]
    fn test_read_n_wrong_type_bool() {
        let result = super::validate_read_n_count(&Value::Bool(true));
        assert!(result.is_err());
        assert!(result.unwrap_err().contains("expected Int"));
    }
}