Files
c3c/lib/std/io_formatter_private.c3
Christoffer Lerno ab78663f3c Add usz and isz.
2022-10-10 15:44:03 +02:00

573 lines
14 KiB
C

module std::io;
private fn void! Formatter.left_adjust(Formatter* this, usz len)
{
if (!this.flags.left) return;
for (usz l = len; l < this.width; l++) this.out(' ')?;
}
private fn void! Formatter.right_adjust(Formatter* this, usz len)
{
if (this.flags.left) return;
for (usz l = len; l < this.width; l++) this.out(' ')?;
}
private fn NtoaType int_from_variant(variant arg, bool *is_neg)
{
*is_neg = false;
$if (NtoaType.typeid == uint128.typeid):
switch (arg)
{
case int128:
int128 val = *arg;
return (*is_neg = val < 0) ? (~(NtoaType)val) + 1 : val;
case uint128:
return *arg;
}
$endif;
if (arg.type.kindof == TypeKind.POINTER)
{
return (NtoaType)(uptr)*(void**)arg.ptr;
}
switch (arg)
{
case bool:
return (NtoaType)*arg;
case ichar:
int val = *arg;
return (*is_neg = val < 0) ? (~(NtoaType)val) + 1 : (NtoaType)val;
case short:
int val = *arg;
return (*is_neg = val < 0) ? (~(NtoaType)val) + 1 : (NtoaType)val;
case int:
int val = *arg;
return (*is_neg = val < 0) ? (~(NtoaType)val) + 1 : (NtoaType)val;
case long:
long val = *arg;
return (*is_neg = val < 0) ? (~(NtoaType)val) + 1 : (NtoaType)val;
case char:
return *arg;
case ushort:
return *arg;
case uint:
return *arg;
case ulong:
return *arg;
case float:
float f = *arg;
return (NtoaType)((*is_neg = f < 0) ? -f : f);
case double:
double d = *arg;
return (NtoaType)((*is_neg = d < 0) ? -d : d);
default:
return 0;
}
}
private fn FloatType float_from_variant(variant arg)
{
$if (env::I128_SUPPORT):
switch (arg)
{
case int128:
return *arg;
case uint128:
return *arg;
}
$endif;
$if (env::F128_SUPPORT):
if (arg.type == float128.typeid) return *((float128*)arg.ptr);
$endif;
$if (env::F16_SUPPORT):
if (arg.type == float16.typeid) return *((float16*)arg.ptr);
$endif;
if (arg.type.kindof == TypeKind.POINTER)
{
return (FloatType)(uptr)(void*)arg.ptr;
}
switch (arg)
{
case bool:
return (FloatType)*arg;
case ichar:
return *arg;
case short:
return *arg;
case int:
return *arg;
case long:
return *arg;
case char:
return *arg;
case ushort:
return *arg;
case uint:
return *arg;
case ulong:
return *arg;
case float:
return (FloatType)*arg;
case double:
return (FloatType)*arg;
default:
return 0;
}
}
/**
* Read a simple integer value, typically for formatting.
*
* @param [inout] len_ptr "the length remaining."
* @param [in] buf "the buf to read from."
* @param maxlen "the maximum len that can be read."
* @return "The result of the atoi."
**/
private fn uint simple_atoi(char* buf, usz maxlen, usz* len_ptr) @inline
{
uint i = 0;
usz len = *len_ptr;
while (len < maxlen)
{
char c = buf[len];
if (c < '0' || c > '9') break;
i = i * 10 + c - '0';
len++;
}
*len_ptr = len;
return i;
}
private fn void! Formatter.out_substr(Formatter *this, char[] str)
{
usz l = conv::utf8_codepoints(str);
uint prec = this.prec;
if (this.flags.precision && l < prec) l = prec;
this.right_adjust(' ')?;
usz index = 0;
usz chars = str.len;
char* ptr = str.ptr;
while (index < chars)
{
char c = ptr[index];
// Break if we have precision set and we ran out...
if (c & 0xC0 != 0x80 && this.flags.precision && !prec--) break;
this.out(c)?;
index++;
}
return this.left_adjust(l);
}
union ConvUnion
{
ulong u;
double f;
}
private fn void! Formatter.etoa(Formatter* this, FloatType value)
{
// check for NaN and special values
if (value != value || value < FloatType.min || value > FloatType.max)
{
return this.ftoa(value);
}
// determine the sign
bool negative = value < 0;
if (negative) value = -value;
// default precision
if (!this.flags.precision)
{
this.prec = PRINTF_DEFAULT_FLOAT_PRECISION;
}
// determine the decimal exponent
// based on the algorithm by David Gay (https://www.ampl.com/netlib/fp/dtoa.c)
ConvUnion conv;
conv.f = (double)value;
int exp2 = (int)(conv.u >> 52 & 0x7FF) - 1023; // effectively log2
conv.u = (conv.u & (1u64 << 52 - 1)) | (1023u64 << 52); // drop the exponent so conv.F is now in [1,2)
// now approximate log10 from the log2 integer part and an expansion of ln around 1.5
int expval = (int)(0.1760912590558 + exp2 * 0.301029995663981 + (conv.f - 1.5) * 0.289529654602168);
// now we want to compute 10^expval but we want to be sure it won't overflow
exp2 = (int)(expval * 3.321928094887362 + 0.5);
double z = expval * 2.302585092994046 - exp2 * 0.6931471805599453;
double z2 = z * z;
conv.u = (ulong)(exp2 + 1023) << 52;
// compute exp(z) using continued fractions, see https://en.wikipedia.org/wiki/Exponential_function#Continued_fractions_for_ex
conv.f *= 1 + 2 * z / (2 - z + (z2 / (6 + (z2 / (10 + z2 / 14)))));
// correct for rounding errors
if (value < conv.f)
{
expval--;
conv.f /= 10;
}
// the exponent format is "%+03d" and largest value is "307", so set aside 4-5 characters
uint minwidth = ((expval < 100) && (expval > -100)) ? 4 : 5;
// in "%g" mode, "prec" is the number of *significant figures* not decimals
if (this.flags.adapt_exp)
{
// do we want to fall-back to "%f" mode?
if (value >= 1e-4 && value < 1e6)
{
this.prec = this.prec > expval ? this.prec - expval - 1 : 0;
this.flags.precision = true; // make sure ftoa respects precision
// no characters in exponent
minwidth = 0;
expval = 0;
}
else
{
// we use one sigfig for the whole part
if (this.prec > 0 && this.flags.precision) this.prec--;
}
}
// Adjust width
uint fwidth = this.width > minwidth ? this.width - minwidth : 0;
// if we're padding on the right, DON'T pad the floating part
if (this.flags.left && minwidth) fwidth = 0;
// rescale the float value
if (expval) value /= conv.f;
// output the floating part
usz start_idx = this.idx;
PrintFlags old = this.flags;
this.flags.adapt_exp = false;
this.width = fwidth;
this.ftoa(negative ? -value : value)?;
this.flags = old;
// output the exponent part
if (minwidth)
{
// output the exponential symbol
this.out(this.flags.uppercase ? 'E' : 'e')?;
// output the exponent value
this.flags = { .zeropad = true, .plus = true };
this.width = minwidth - 1;
this.prec = 0;
this.ntoa((NtoaType)(expval < 0 ? -expval : expval), expval < 0, 10)?;
this.flags = old;
// might need to right-pad spaces
this.left_adjust(this.idx - start_idx)?;
}
}
// internal ftoa for fixed decimal floating point
private fn void! Formatter.ftoa(Formatter* this, FloatType value)
{
char[PRINTF_FTOA_BUFFER_SIZE] buf = void;
usz len = 0;
const FloatType[] POW10 = { 1, 10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000, 1000000000 };
FloatType diff = 0.0;
// powers of 10
// test for special values
if (value != value)
{
return this.out_reverse("nan");
}
if (value < -FloatType.max)
{
return this.out_reverse("fni-");
}
if (value > FloatType.max)
{
if (this.flags.plus)
{
return this.out_reverse("fni+");
}
return this.out_reverse("fni");
}
// test for very large values
// standard printf behavior is to print EVERY whole number digit -- which could be 100s of characters overflowing your buffers == bad
if (value > PRINTF_MAX_FLOAT || value < -PRINTF_MAX_FLOAT)
{
return this.etoa(value);
}
// test for negative
bool negative = value < 0;
if (negative) value = 0 - value;
// set default precision, if not set explicitly
if (!this.flags.precision) this.prec = PRINTF_DEFAULT_FLOAT_PRECISION;
// limit precision to 9, cause a prec >= 10 can lead to overflow errors
while (this.prec > 9)
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '0';
this.prec--;
}
// Safe due to 1e9 limit.
int whole = (int)value;
FloatType tmp = (value - whole) * POW10[this.prec];
ulong frac = (ulong)tmp;
diff = tmp - frac;
switch (true)
{
case diff > 0.5:
++frac;
// handle rollover, e.g. case 0.99 with prec 1 is 1.0
if (frac >= POW10[this.prec])
{
frac = 0;
++whole;
}
case diff < 0.5:
break;
case !frac && (frac & 1):
// if halfway, round up if odd OR if last digit is 0
++frac;
}
if (!this.prec)
{
diff = value - (FloatType)whole;
if ((!(diff < 0.5) || diff > 0.5) && (whole & 1))
{
// exactly 0.5 and ODD, then round up
// 1.5 -> 2, but 2.5 -> 2
++whole;
}
}
else
{
uint count = this.prec;
// now do fractional part, as an unsigned number
do
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
--count;
buf[len++] = (char)(48 + (frac % 10));
}
while (frac /= 10);
// add extra 0s
while (count-- > 0)
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '0';
}
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
// add decimal
buf[len++] = '.';
}
// do whole part, number is reversed
do
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = (char)(48 + (whole % 10));
}
while (whole /= 10);
// pad leading zeros
if (!this.flags.left && this.flags.zeropad)
{
if (this.width && (negative || this.flags.plus || this.flags.space)) this.width--;
while (len < this.width)
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '0';
}
}
char next = {|
if (negative) return '-';
if (this.flags.plus) return '+';
if (this.flags.space) return ' ';
return 0;
|};
if (next)
{
if (len >= PRINTF_FTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = next;
}
return this.out_reverse(buf[:len]);
}
private fn void! Formatter.ntoa(Formatter* this, NtoaType value, bool negative, uint base)
{
char[PRINTF_NTOA_BUFFER_SIZE] buf = void;
usz len = 0;
// no hash for 0 values
if (!value) this.flags.hash = false;
// write if precision != 0 or value is != 0
if (!this.flags.precision || value)
{
char past_10 = (this.flags.uppercase ? 'A' : 'a') - 10;
do
{
if (len >= PRINTF_NTOA_BUFFER_SIZE) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
char digit = (char)(value % base);
buf[len++] = digit + (digit < 10 ? '0' : past_10);
value /= base;
}
while (value);
}
return this.ntoa_format(buf[:PRINTF_NTOA_BUFFER_SIZE], len, negative, base);
}
private fn void! Formatter.ntoa_format(Formatter* this, char[] buf, usz len, bool negative, uint base)
{
// pad leading zeros
if (!this.flags.left)
{
if (this.width && this.flags.zeropad && (negative || this.flags.plus || this.flags.space)) this.width--;
while (len < this.prec)
{
if (len >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '0';
}
while (this.flags.zeropad && len < this.width)
{
if (len >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '0';
}
}
// handle hash
if (this.flags.hash && base != 10)
{
if (!this.flags.precision && len && len == this.prec && len == this.width)
{
len--;
if (len) len--;
}
if (base != 10)
{
if (len + 1 >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
switch (base)
{
case 16:
buf[len++] = this.flags.uppercase ? 'X' : 'x';
case 8:
buf[len++] = this.flags.uppercase ? 'O' : 'o';
case 2:
buf[len++] = this.flags.uppercase ? 'B' : 'b';
default:
unreachable();
}
buf[len++] = '0';
}
}
switch (true)
{
case negative:
if (len >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '-';
case this.flags.plus:
if (len >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = '+';
case this.flags.space:
if (len >= buf.len) return PrintFault.INTERNAL_BUFFER_EXCEEDED!;
buf[len++] = ' ';
}
if (!len) return;
return this.out_reverse(buf[:len]);
}
$if (env::I128_SUPPORT):
define NtoaType = uint128;
$else:
define NtoaType = ulong;
$endif;
private fn void! Formatter.ntoa_variant(Formatter* this, variant arg, uint base)
{
bool is_neg;
NtoaType val = int_from_variant(arg, &is_neg);
return this.ntoa(val, is_neg, base) @inline;
}
private fn void! Formatter.out_char(Formatter* this, variant arg)
{
uint l = 1;
// pre padding
this.right_adjust(l)?;
// char output
Char32 c = types::variant_to_int(arg, uint) ?? 0xFFFD;
switch (true)
{
case c < 0x7f:
this.out((char)c)?;
case c < 0x7ff:
this.out((char)(0xC0 | c >> 6))?;
this.out((char)(0x80 | (c & 0x3F)))?;
case c < 0xffff:
this.out((char)(0xE0 | c >> 12))?;
this.out((char)(0x80 | (c >> 6 & 0x3F)))?;
this.out((char)(0x80 | (c & 0x3F)))?;
default:
this.out((char)(0xF0 | c >> 18))?;
this.out((char)(0x80 | (c >> 12 & 0x3F)))?;
this.out((char)(0x80 | (c >> 6 & 0x3F)))?;
this.out((char)(0x80 | (c & 0x3F)))?;
}
return this.left_adjust(l);
}
private fn void! Formatter.out_reverse(Formatter* this, char[] buf)
{
usz buffer_start_idx = this.idx;
usz len = buf.len;
// pad spaces up to given width
if (!this.flags.left && !this.flags.zeropad)
{
for (usz i = len; i < this.width; i++)
{
this.out(' ')?;
}
}
// reverse string
while (len) this.out(buf[--len])?;
// append pad spaces up to given width
return this.left_adjust(this.idx - buffer_start_idx);
}
private fn void! printf_advance_format(usz format_len, usz *index_ptr) @inline
{
usz val = ++(*index_ptr);
if (val >= format_len) return FormattingFault.UNTERMINATED_FORMAT!;
}
private fn variant! next_variant(variant* args_ptr, usz args_len, usz* arg_index_ptr) @inline
{
if (*arg_index_ptr >= args_len) return FormattingFault.MISSING_ARG!;
return args_ptr[(*arg_index_ptr)++];
}
private fn int! printf_parse_format_field(variant* args_ptr, usz args_len, usz* args_index_ptr, char* format_ptr, usz format_len, usz* index_ptr) @inline
{
char c = format_ptr[*index_ptr];
if (c >= '0' && c <= '9') return simple_atoi(format_ptr, format_len, index_ptr);
if (c != '*') return 0;
printf_advance_format(format_len, index_ptr)?;
variant val = next_variant(args_ptr, args_len, args_index_ptr)?;
if (!val.type.kindof.is_int()) return FormattingFault.INVALID_WIDTH_ARG!;
uint! intval = types::variant_to_int(val, int);
if (catch intval) return FormattingFault.INVALID_WIDTH_ARG!;
return intval;
}