Files
Odin/core/strconv/strconv.odin

709 lines
14 KiB
Odin

package strconv
import "core:unicode/utf8"
parse_bool :: proc(s: string) -> (result: bool = false, ok: bool) {
switch s {
case "1", "t", "T", "true", "TRUE", "True":
return true, true;
case "0", "f", "F", "false", "FALSE", "False":
return false, true;
}
return;
}
_digit_value :: proc(r: rune) -> int {
ri := int(r);
v: int = 16;
switch r {
case '0'..='9': v = ri-'0';
case 'a'..='z': v = ri-'a'+10;
case 'A'..='Z': v = ri-'A'+10;
}
return v;
}
// Parses an integer value from a string, in the given base, without a prefix.
//
// Returns ok=false if no numeric value of the appropriate base could be found,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_i64_of_base("-1234eeee", 10);
// assert(n == -1234 && ok);
// ```
parse_i64_of_base :: proc(str: string, base: int) -> (value: i64, ok: bool) {
assert(base <= 16, "base must be 1-16");
s := str;
if s == "" {
return;
}
neg := false;
if len(s) > 1 {
switch s[0] {
case '-':
neg = true;
s = s[1:];
case '+':
s = s[1:];
}
}
i := 0;
for r in s {
if r == '_' {
i += 1;
continue;
}
v := i64(_digit_value(r));
if v >= i64(base) {
break;
}
value *= i64(base);
value += v;
i += 1;
}
s = s[i:];
if neg {
value = -value;
}
ok = len(s) == 0;
return;
}
// Parses a integer value from a string, in base 10, unless there's a prefix.
//
// Returns ok=false if a valid integer could not be found,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_i64_maybe_prefixed("1234");
// assert(n == 1234 && ok);
//
// n, ok = strconv.parse_i64_maybe_prefixed("0xeeee");
// assert(n == 0xeeee && ok);
// ```
parse_i64_maybe_prefixed :: proc(str: string) -> (value: i64, ok: bool) {
s := str;
if s == "" {
return;
}
neg := false;
if len(s) > 1 {
switch s[0] {
case '-':
neg = true;
s = s[1:];
case '+':
s = s[1:];
}
}
base: i64 = 10;
if len(s) > 2 && s[0] == '0' {
switch s[1] {
case 'b': base = 2; s = s[2:];
case 'o': base = 8; s = s[2:];
case 'd': base = 10; s = s[2:];
case 'z': base = 12; s = s[2:];
case 'x': base = 16; s = s[2:];
}
}
i := 0;
for r in s {
if r == '_' {
i += 1;
continue;
}
v := i64(_digit_value(r));
if v >= base {
break;
}
value *= base;
value += v;
i += 1;
}
s = s[i:];
if neg {
value = -value;
}
ok = len(s) == 0;
return;
}
parse_i64 :: proc{parse_i64_maybe_prefixed, parse_i64_of_base};
// Parses an unsigned integer value from a string, in the given base, and
// without a prefix.
//
// Returns ok=false if no numeric value of the appropriate base could be found,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_u64_of_base("1234eeee", 10);
// assert(n == 1234 && ok);
//
// n, ok = strconv.parse_u64_of_base("5678eeee", 16);
// assert(n == 0x5678eeee && ok);
// ```
parse_u64_of_base :: proc(str: string, base: int) -> (value: u64, ok: bool) {
assert(base <= 16, "base must be 1-16");
s := str;
if s == "" {
return;
}
if len(s) > 1 && s[0] == '+' {
s = s[1:];
}
i := 0;
for r in s {
if r == '_' {
i += 1;
continue;
}
v := u64(_digit_value(r));
if v >= u64(base) {
break;
}
value *= u64(base);
value += v;
i += 1;
}
s = s[i:];
ok = len(s) == 0;
return;
}
// Parses an unsigned integer value from a string in base 10, unless there's a prefix.
//
// Returns ok=false if a valid integer could not be found, if the value was negative,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_u64_maybe_prefixed("1234");
// assert(n == 1234 && ok);
//
// n, ok = strconv.parse_u64_maybe_prefixed("0xeeee");
// assert(n == 0xeeee && ok);
// ```
parse_u64_maybe_prefixed :: proc(str: string) -> (value: u64, ok: bool) {
s := str;
if s == "" {
return;
}
if len(s) > 1 && s[0] == '+' {
s = s[1:];
}
base := u64(10);
if len(s) > 2 && s[0] == '0' {
switch s[1] {
case 'b': base = 2; s = s[2:];
case 'o': base = 8; s = s[2:];
case 'd': base = 10; s = s[2:];
case 'z': base = 12; s = s[2:];
case 'x': base = 16; s = s[2:];
}
}
i := 0;
for r in s {
if r == '_' {
i += 1;
continue;
}
v := u64(_digit_value(r));
if v >= base {
break;
}
value *= base;
value += v;
i += 1;
}
s = s[i:];
ok = len(s) == 0;
return;
}
parse_u64 :: proc{parse_u64_maybe_prefixed, parse_u64_of_base};
// Parses an integer value from a string in the given base, or
// - if the string has a prefix (e.g: '0x') then that will determine the base;
// - otherwise, assumes base 10.
//
// Returns ok=false if no appropriate value could be found, or if the input string
// contained more than just the number.
//
// ```
// n, ok := strconv.parse_int("1234"); // without prefix, inferred base 10
// assert(n == 1234 && ok);
//
// n, ok = strconv.parse_int("ffff", 16); // without prefix, explicit base
// assert(n == 0xffff && ok);
//
// n, ok = strconv.parse_int("0xffff"); // with prefix and inferred base
// assert(n == 0xffff && ok);
// ```
parse_int :: proc(s: string, base := 0) -> (value: int, ok: bool) {
v: i64 = ---;
switch base {
case 0: v, ok = parse_i64_maybe_prefixed(s);
case: v, ok = parse_i64_of_base(s, base);
}
value = int(v);
return;
}
// Parses an unsigned integer value from a string in the given base, or
// - if the string has a prefix (e.g: '0x') then that will determine the base;
// - otherwise, assumes base 10.
//
// Returns ok=false if:
// - no appropriate value could be found; or
// - the value was negative.
// - the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_uint("1234"); // without prefix, inferred base 10
// assert(n == 1234 && ok);
//
// n, ok = strconv.parse_uint("ffff", 16); // without prefix, explicit base
// assert(n == 0xffff && ok);
//
// n, ok = strconv.parse_uint("0xffff"); // with prefix and inferred base
// assert(n == 0xffff && ok);
// ```
parse_uint :: proc(s: string, base := 0) -> (value: uint, ok: bool) {
v: u64 = ---;
switch base {
case 0: v, ok = parse_u64_maybe_prefixed(s);
case: v, ok = parse_u64_of_base(s, base);
}
value = uint(v);
return;
}
// Parses a 32-bit floating point number from a string.
//
// Returns ok=false if a base 10 float could not be found,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_f32("12.34eee");
// assert(n == 12.34 && ok);
//
// n, ok = strconv.parse_f32("12.34");
// assert(n == 12.34 && ok);
// ```
parse_f32 :: proc(s: string) -> (value: f32, ok: bool) {
v: f64 = ---;
v, ok = parse_f64(s);
return f32(v), ok;
}
// Parses a 64-bit floating point number from a string.
//
// Returns ok=false if a base 10 float could not be found,
// or if the input string contained more than just the number.
//
// ```
// n, ok := strconv.parse_f32("12.34eee");
// assert(n == 12.34 && ok);
//
// n, ok = strconv.parse_f32("12.34");
// assert(n == 12.34 && ok);
// ```
parse_f64 :: proc(str: string) -> (value: f64, ok: bool) {
s := str;
if s == "" {
return;
}
i := 0;
sign: f64 = 1;
switch s[i] {
case '-': i += 1; sign = -1;
case '+': i += 1;
}
for ; i < len(s); i += 1 {
r := rune(s[i]);
if r == '_' {
continue;
}
v := _digit_value(r);
if v >= 10 {
break;
}
value *= 10;
value += f64(v);
}
if i < len(s) && s[i] == '.' {
pow10: f64 = 10;
i += 1;
for ; i < len(s); i += 1 {
r := rune(s[i]);
if r == '_' {
continue;
}
v := _digit_value(r);
if v >= 10 {
break;
}
value += f64(v)/pow10;
pow10 *= 10;
}
}
frac := false;
scale: f64 = 1;
if i < len(s) && (s[i] == 'e' || s[i] == 'E') {
i += 1;
if i < len(s) {
switch s[i] {
case '-': i += 1; frac = true;
case '+': i += 1;
}
exp: u32 = 0;
for ; i < len(s); i += 1 {
r := rune(s[i]);
if r == '_' {
continue;
}
d := u32(_digit_value(r));
if d >= 10 {
break;
}
exp = exp * 10 + d;
}
if exp > 308 { exp = 308; }
for exp >= 50 { scale *= 1e50; exp -= 50; }
for exp >= 8 { scale *= 1e8; exp -= 8; }
for exp > 0 { scale *= 10; exp -= 1; }
}
}
s = s[i:];
if frac {
value = sign * (value/scale);
} else {
value = sign * (value*scale);
}
ok = len(s) == 0;
return;
}
append_bool :: proc(buf: []byte, b: bool) -> string {
n := 0;
if b {
n = copy(buf, "true");
} else {
n = copy(buf, "false");
}
return string(buf[:n]);
}
append_uint :: proc(buf: []byte, u: u64, base: int) -> string {
return append_bits(buf, u, base, false, 8*size_of(uint), digits, nil);
}
append_int :: proc(buf: []byte, i: i64, base: int) -> string {
return append_bits(buf, u64(i), base, true, 8*size_of(int), digits, nil);
}
itoa :: proc(buf: []byte, i: int) -> string {
return append_int(buf, i64(i), 10);
}
atoi :: proc(s: string) -> int {
v, _ := parse_int(s);
return v;
}
atof :: proc(s: string) -> f64 {
v, _ := parse_f64(s);
return v;
}
ftoa :: append_float;
append_float :: proc(buf: []byte, f: f64, fmt: byte, prec, bit_size: int) -> string {
return string(generic_ftoa(buf, f, fmt, prec, bit_size));
}
quote :: proc(buf: []byte, str: string) -> string {
write_byte :: proc(buf: []byte, i: ^int, bytes: ..byte) {
if i^ >= len(buf) {
return;
}
n := copy(buf[i^:], bytes[:]);
i^ += n;
}
if buf == nil {
return "";
}
c :: '"';
i := 0;
s := str;
write_byte(buf, &i, c);
for width := 0; len(s) > 0; s = s[width:] {
r := rune(s[0]);
width = 1;
if r >= utf8.RUNE_SELF {
r, width = utf8.decode_rune_in_string(s);
}
if width == 1 && r == utf8.RUNE_ERROR {
write_byte(buf, &i, '\\', 'x');
write_byte(buf, &i, digits[s[0]>>4]);
write_byte(buf, &i, digits[s[0]&0xf]);
}
if i < len(buf) {
x := quote_rune(buf[i:], r);
i += len(x);
}
}
write_byte(buf, &i, c);
return string(buf[:i]);
}
quote_rune :: proc(buf: []byte, r: rune) -> string {
write_byte :: proc(buf: []byte, i: ^int, bytes: ..byte) {
if i^ < len(buf) {
n := copy(buf[i^:], bytes[:]);
i^ += n;
}
}
write_string :: proc(buf: []byte, i: ^int, s: string) {
if i^ < len(buf) {
n := copy(buf[i^:], s);
i^ += n;
}
}
write_rune :: proc(buf: []byte, i: ^int, r: rune) {
if i^ < len(buf) {
b, w := utf8.encode_rune(r);
n := copy(buf[i^:], b[:w]);
i^ += n;
}
}
if buf == nil {
return "";
}
i := 0;
write_byte(buf, &i, '\'');
switch r {
case '\a': write_string(buf, &i, "\\a");
case '\b': write_string(buf, &i, "\\b");
case '\e': write_string(buf, &i, "\\e");
case '\f': write_string(buf, &i, "\\f");
case '\n': write_string(buf, &i, "\\n");
case '\r': write_string(buf, &i, "\\r");
case '\t': write_string(buf, &i, "\\t");
case '\v': write_string(buf, &i, "\\v");
case:
if r < 32 {
write_string(buf, &i, "\\x");
b: [2]byte;
s := append_bits(b[:], u64(r), 16, true, 64, digits, nil);
switch len(s) {
case 0: write_string(buf, &i, "00");
case 1: write_rune(buf, &i, '0');
case 2: write_string(buf, &i, s);
}
} else {
write_rune(buf, &i, r);
}
}
write_byte(buf, &i, '\'');
return string(buf[:i]);
}
unquote_char :: proc(str: string, quote: byte) -> (r: rune, multiple_bytes: bool, tail_string: string, success: bool) {
hex_to_int :: proc(c: byte) -> int {
switch c {
case '0'..='9': return int(c-'0');
case 'a'..='f': return int(c-'a')+10;
case 'A'..='F': return int(c-'A')+10;
}
return -1;
}
w: int;
if str[0] == quote && quote == '"' {
return;
} else if str[0] >= 0x80 {
r, w = utf8.decode_rune_in_string(str);
return r, true, str[w:], true;
} else if str[0] != '\\' {
return rune(str[0]), false, str[1:], true;
}
if len(str) <= 1 {
return;
}
s := str;
c := s[1];
s = s[2:];
switch c {
case:
return;
case 'a': r = '\a';
case 'b': r = '\b';
case 'f': r = '\f';
case 'n': r = '\n';
case 'r': r = '\r';
case 't': r = '\t';
case 'v': r = '\v';
case '\\': r = '\\';
case '"': r = '"';
case '\'': r = '\'';
case '0'..='7':
v := int(c-'0');
if len(s) < 2 {
return;
}
for i in 0..<len(s) {
d := int(s[i]-'0');
if d < 0 || d > 7 {
return;
}
v = (v<<3) | d;
}
s = s[2:];
if v > 0xff {
return;
}
r = rune(v);
case 'x', 'u', 'U':
count: int;
switch c {
case 'x': count = 2;
case 'u': count = 4;
case 'U': count = 8;
}
if len(s) < count {
return;
}
for i in 0..<count {
d := hex_to_int(s[i]);
if d < 0 {
return;
}
r = (r<<4) | rune(d);
}
s = s[count:];
if c == 'x' {
break;
}
if r > utf8.MAX_RUNE {
return;
}
multiple_bytes = true;
}
success = true;
tail_string = s;
return;
}
unquote_string :: proc(lit: string, allocator := context.allocator) -> (res: string, allocated, success: bool) {
contains_rune :: proc(s: string, r: rune) -> int {
for c, offset in s {
if c == r {
return offset;
}
}
return -1;
}
assert(len(lit) >= 2);
if lit[0] == '`' {
return lit[1:len(lit)-1], false, true;
}
s := lit;
quote := '"';
if s == `""` {
return "", false, true;
}
if contains_rune(s, '\n') >= 0 {
return s, false, false;
}
if contains_rune(s, '\\') < 0 && contains_rune(s, quote) < 0 {
if quote == '"' {
return s, false, true;
}
}
buf_len := 3*len(s) / 2;
buf := make([]byte, buf_len, allocator);
offset := 0;
for len(s) > 0 {
r, multiple_bytes, tail_string, ok := unquote_char(s, byte(quote));
if !ok {
delete(buf);
return s, false, false;
}
s = tail_string;
if r < 0x80 || !multiple_bytes {
buf[offset] = byte(r);
offset += 1;
} else {
b, w := utf8.encode_rune(r);
copy(buf[offset:], b[:w]);
offset += w;
}
}
new_string := string(buf[:offset]);
return new_string, true, true;
}