zig/std/io.zig
Andrew Kelley 78d4fb20c4 inline parameters
This replaces the current generic syntax for functions and replaces
it with the concept of inline parameters.

This paves the way for the "all structs anonymous" proposal.

Closes #151.
2016-07-25 22:55:15 -07:00

407 lines
11 KiB
Zig

const linux = @import("linux.zig");
const errno = @import("errno.zig");
const math = @import("math.zig");
pub const stdin_fileno = 0;
pub const stdout_fileno = 1;
pub const stderr_fileno = 2;
pub var stdin = InStream {
.fd = stdin_fileno,
};
pub var stdout = OutStream {
.fd = stdout_fileno,
.buffer = undefined,
.index = 0,
};
pub var stderr = OutStream {
.fd = stderr_fileno,
.buffer = undefined,
.index = 0,
};
/// The function received invalid input at runtime. An Invalid error means a
/// bug in the program that called the function.
pub error Invalid;
/// When an Unexpected error occurs, code that emitted the error likely needs
/// a patch to recognize the unexpected case so that it can handle it and emit
/// a more specific error.
pub error Unexpected;
pub error DiskQuota;
pub error FileTooBig;
pub error SigInterrupt;
pub error Io;
pub error NoSpaceLeft;
pub error BadPerm;
pub error PipeFail;
pub error BadFd;
pub error IsDir;
pub error NotDir;
pub error SymLinkLoop;
pub error ProcessFdQuotaExceeded;
pub error SystemFdQuotaExceeded;
pub error NameTooLong;
pub error NoDevice;
pub error PathNotFound;
pub error NoMem;
const buffer_size = 4 * 1024;
const max_u64_base10_digits = 20;
const max_f64_digits = 65;
pub const OpenRead = 0b0001;
pub const OpenWrite = 0b0010;
pub const OpenCreate = 0b0100;
pub const OpenTruncate = 0b1000;
pub struct OutStream {
fd: isize,
buffer: [buffer_size]u8,
index: isize,
pub fn write(os: &OutStream, bytes: []const u8) -> %isize {
var src_bytes_left = bytes.len;
var src_index: @typeof(bytes.len) = 0;
const dest_space_left = os.buffer.len - os.index;
while (src_bytes_left > 0) {
const copy_amt = math.min(isize, dest_space_left, src_bytes_left);
@memcpy(&os.buffer[os.index], &bytes[src_index], copy_amt);
os.index += copy_amt;
if (os.index == os.buffer.len) {
%return os.flush();
}
src_bytes_left -= copy_amt;
}
return bytes.len;
}
/// Prints a byte buffer, flushes the buffer, then returns the number of
/// bytes printed. The "f" is for "flush".
pub fn printf(os: &OutStream, str: []const u8) -> %isize {
const byte_count = %return os.write(str);
%return os.flush();
return byte_count;
}
pub fn print_u64(os: &OutStream, x: u64) -> %isize {
if (os.index + max_u64_base10_digits >= os.buffer.len) {
%return os.flush();
}
const amt_printed = buf_print_u64(os.buffer[os.index...], x);
os.index += amt_printed;
return amt_printed;
}
pub fn print_i64(os: &OutStream, x: i64) -> %isize {
if (os.index + max_u64_base10_digits >= os.buffer.len) {
%return os.flush();
}
const amt_printed = buf_print_i64(os.buffer[os.index...], x);
os.index += amt_printed;
return amt_printed;
}
pub fn print_f64(os: &OutStream, x: f64) -> %isize {
if (os.index + max_f64_digits >= os.buffer.len) {
%return os.flush();
}
const amt_printed = buf_print_f64(os.buffer[os.index...], x, 4);
os.index += amt_printed;
return amt_printed;
}
pub fn flush(os: &OutStream) -> %void {
const write_ret = linux.write(os.fd, &os.buffer[0], os.index);
const write_err = linux.get_errno(write_ret);
if (write_err > 0) {
return switch (write_err) {
errno.EINVAL => unreachable{},
errno.EDQUOT => error.DiskQuota,
errno.EFBIG => error.FileTooBig,
errno.EINTR => error.SigInterrupt,
errno.EIO => error.Io,
errno.ENOSPC => error.NoSpaceLeft,
errno.EPERM => error.BadPerm,
errno.EPIPE => error.PipeFail,
else => error.Unexpected,
}
}
os.index = 0;
}
pub fn close(os: &OutStream) -> %void {
const closed = linux.close(os.fd);
if (closed < 0) {
return switch (-closed) {
errno.EIO => error.Io,
errno.EBADF => error.BadFd,
errno.EINTR => error.SigInterrupt,
else => error.Unexpected,
}
}
}
}
pub struct InStream {
fd: isize,
pub fn open(path: []u8) -> %InStream {
const fd = linux.open(path, linux.O_LARGEFILE|linux.O_RDONLY, 0);
if (fd < 0) {
return switch (-fd) {
errno.EFAULT => unreachable{},
errno.EINVAL => unreachable{},
errno.EACCES => error.BadPerm,
errno.EFBIG, errno.EOVERFLOW => error.FileTooBig,
errno.EINTR => error.SigInterrupt,
errno.EISDIR => error.IsDir,
errno.ELOOP => error.SymLinkLoop,
errno.EMFILE => error.ProcessFdQuotaExceeded,
errno.ENAMETOOLONG => error.NameTooLong,
errno.ENFILE => error.SystemFdQuotaExceeded,
errno.ENODEV => error.NoDevice,
errno.ENOENT => error.PathNotFound,
errno.ENOMEM => error.NoMem,
errno.ENOSPC => error.NoSpaceLeft,
errno.ENOTDIR => error.NotDir,
errno.EPERM => error.BadPerm,
else => error.Unexpected,
}
}
return InStream { .fd = fd, };
}
pub fn read(is: &InStream, buf: []u8) -> %isize {
const amt_read = linux.read(is.fd, &buf[0], buf.len);
if (amt_read < 0) {
return switch (-amt_read) {
errno.EINVAL => unreachable{},
errno.EFAULT => unreachable{},
errno.EBADF => error.BadFd,
errno.EINTR => error.SigInterrupt,
errno.EIO => error.Io,
else => error.Unexpected,
}
}
return amt_read;
}
pub fn close(is: &InStream) -> %void {
const closed = linux.close(is.fd);
if (closed < 0) {
return switch (-closed) {
errno.EIO => error.Io,
errno.EBADF => error.BadFd,
errno.EINTR => error.SigInterrupt,
else => error.Unexpected,
}
}
}
}
pub fn parse_unsigned(inline T: type, buf: []u8, radix: u8) -> %T {
var x: T = 0;
for (buf) |c| {
const digit = %return char_to_digit(c, radix);
x = %return math.mul_overflow(T, x, radix);
x = %return math.add_overflow(T, x, digit);
}
return x;
}
pub error InvalidChar;
fn char_to_digit(c: u8, radix: u8) -> %u8 {
const value = if ('0' <= c && c <= '9') {
c - '0'
} else if ('A' <= c && c <= 'Z') {
c - 'A' + 10
} else if ('a' <= c && c <= 'z') {
c - 'a' + 10
} else {
return error.InvalidChar;
};
return if (value >= radix) error.InvalidChar else value;
}
pub fn buf_print_signed(inline T: type, out_buf: []u8, x: T) -> isize {
const uint = @int_type(false, T.bit_count, false);
if (x < 0) {
out_buf[0] = '-';
return 1 + buf_print_unsigned(uint, out_buf[1...], uint(-(x + 1)) + 1);
} else {
return buf_print_unsigned(uint, out_buf, uint(x));
}
}
pub fn buf_print_i64(out_buf: []u8, x: i64) -> isize {
buf_print_signed(i64, out_buf, x)
}
pub fn buf_print_unsigned(inline T: type, out_buf: []u8, x: T) -> isize {
var buf: [max_u64_base10_digits]u8 = undefined;
var a = x;
var index: isize = buf.len;
while (true) {
const digit = a % 10;
index -= 1;
buf[index] = '0' + u8(digit);
a /= 10;
if (a == 0)
break;
}
const len = buf.len - index;
@memcpy(&out_buf[0], &buf[index], len);
return len;
}
pub fn buf_print_u64(out_buf: []u8, x: u64) -> isize {
buf_print_unsigned(u64, out_buf, x)
}
pub fn buf_print_f64(out_buf: []u8, x: f64, decimals: isize) -> isize {
const numExpBits = 11;
const numRawSigBits = 52; // not including implicit 1 bit
const expBias = 1023;
var decs = decimals;
if (decs >= max_u64_base10_digits) {
decs = max_u64_base10_digits - 1;
}
if (x == math.f64_get_pos_inf()) {
const buf2 = "+Inf";
@memcpy(&out_buf[0], &buf2[0], buf2.len);
return 4;
} else if (x == math.f64_get_neg_inf()) {
const buf2 = "-Inf";
@memcpy(&out_buf[0], &buf2[0], buf2.len);
return 4;
} else if (math.f64_is_nan(x)) {
const buf2 = "NaN";
@memcpy(&out_buf[0], &buf2[0], buf2.len);
return 3;
}
var buf: [max_f64_digits]u8 = undefined;
var len: isize = 0;
// 1 sign bit
// 11 exponent bits
// 52 significand bits (+ 1 implicit always non-zero bit)
const bits = math.f64_to_bits(x);
if (bits & (1 << 63) != 0) {
buf[0] = '-';
len += 1;
}
const rexponent: i64 = i64((bits >> numRawSigBits) & ((1 << numExpBits) - 1));
const exponent = rexponent - expBias - numRawSigBits;
if (rexponent == 0) {
buf[len] = '0';
len += 1;
@memcpy(&out_buf[0], &buf[0], len);
return len;
}
const sig = (bits & ((1 << numRawSigBits) - 1)) | (1 << numRawSigBits);
if (exponent >= 0) {
// number is an integer
if (exponent >= 64 - 53) {
// use XeX form
// TODO support printing large floats
//len += buf_print_u64(buf[len...], sig << 10);
const str = "LARGEF64";
@memcpy(&buf[len], &str[0], str.len);
len += str.len;
} else {
// use typical form
len += buf_print_u64(buf[len...], sig << u64(exponent));
buf[len] = '.';
len += 1;
var i: isize = 0;
while (i < decs) {
buf[len] = '0';
len += 1;
i += 1;
}
}
} else {
// number is not an integer
// print out whole part
len += buf_print_u64(buf[len...], sig >> u64(-exponent));
buf[len] = '.';
len += 1;
// print out fractional part
// dec_num holds: fractional part * 10 ^ decs
var dec_num: u64 = 0;
var a: isize = 1;
var i: isize = 0;
while (i < decs + 5) {
a *= 10;
i += 1;
}
// create a mask: 1's for the fractional part, 0's for whole part
var masked_sig = sig & ((1 << u64(-exponent)) - 1);
i = -1;
while (i >= exponent) {
var bit_set = ((1 << u64(i-exponent)) & masked_sig) != 0;
if (bit_set) {
dec_num += usize(a) >> usize(-i);
}
i -= 1;
}
dec_num /= 100000;
len += decs;
i = len - 1;
while (i >= len - decs) {
buf[i] = '0' + u8(dec_num % 10);
dec_num /= 10;
i -= 1;
}
}
@memcpy(&out_buf[0], &buf[0], len);
len
}
#attribute("test")
fn parse_u64_digit_too_big() {
parse_unsigned(u64, "123a", 10) %% |err| {
if (err == error.InvalidChar) return;
unreachable{};
};
unreachable{};
}