Skip to content

Commit

Permalink
Automatic exponential formatting in Debug
Browse files Browse the repository at this point in the history
* {:.PREC?} already had legitimately useful behavior (recursive formatting of structs using
  fixed precision for floats) and I suspect that changes to the output there would be unwelcome.

  (besides, precision introduces sinister edge cases where a number can be rounded up to one
  of the thresholds)

  Thus, the new behavior of Debug is, "dynamically switch to exponential, but only if there's
  no precision."

* This could not be implemented in terms of float_to_decimal_common without repeating the branch
  on precision, so 'float_to_general_debug' is a new function.  The name is '_debug' instead of
  '_common' because the considerations in the previous bullet make this logic pretty specific
  to Debug.

* 'float_to_decimal_common' is now only used by Display, so I inlined the min_precision argument
  and renamed the function accordingly.
  • Loading branch information
ExpHP committed Jun 20, 2021
1 parent 312b894 commit 8731d4d
Show file tree
Hide file tree
Showing 4 changed files with 75 additions and 6 deletions.
53 changes: 49 additions & 4 deletions library/core/src/fmt/float.rs
Expand Up @@ -2,6 +2,26 @@ use crate::fmt::{Debug, Display, Formatter, LowerExp, Result, UpperExp};
use crate::mem::MaybeUninit;
use crate::num::flt2dec;

#[doc(hidden)]
trait GeneralFormat: PartialOrd {
/// Determines if a value should use exponential based on its magnitude, given the precondition
/// that it will not be rounded any further before it is displayed.
fn already_rounded_value_should_use_exponential(&self) -> bool;
}

macro_rules! impl_general_format {
($($t:ident)*) => {
$(impl GeneralFormat for $t {
fn already_rounded_value_should_use_exponential(&self) -> bool {
let abs = $t::abs_private(*self);
(abs != 0.0 && abs < 1e-4) || abs >= 1e+16
}
})*
}
}

impl_general_format! { f32 f64 }

// Don't inline this so callers don't use the stack space this function
// requires unless they have to.
#[inline(never)]
Expand Down Expand Up @@ -53,8 +73,7 @@ where
fmt.pad_formatted_parts(&formatted)
}

// Common code of floating point Debug and Display.
fn float_to_decimal_common<T>(fmt: &mut Formatter<'_>, num: &T, min_precision: usize) -> Result
fn float_to_decimal_display<T>(fmt: &mut Formatter<'_>, num: &T) -> Result
where
T: flt2dec::DecodableFloat,
{
Expand All @@ -67,6 +86,7 @@ where
if let Some(precision) = fmt.precision {
float_to_decimal_common_exact(fmt, num, sign, precision)
} else {
let min_precision = 0;
float_to_decimal_common_shortest(fmt, num, sign, min_precision)
}
}
Expand Down Expand Up @@ -144,19 +164,44 @@ where
}
}

fn float_to_general_debug<T>(fmt: &mut Formatter<'_>, num: &T) -> Result
where
T: flt2dec::DecodableFloat + GeneralFormat,
{
let force_sign = fmt.sign_plus();
let sign = match force_sign {
false => flt2dec::Sign::Minus,
true => flt2dec::Sign::MinusPlus,
};

if let Some(precision) = fmt.precision {
// this behavior of {:.PREC?} predates exponential formatting for {:?}
float_to_decimal_common_exact(fmt, num, sign, precision)
} else {
// since there is no precision, there will be no rounding
if num.already_rounded_value_should_use_exponential() {
let upper = false;
float_to_exponential_common_shortest(fmt, num, sign, upper)
} else {
let min_precision = 1;
float_to_decimal_common_shortest(fmt, num, sign, min_precision)
}
}
}

macro_rules! floating {
($ty:ident) => {
#[stable(feature = "rust1", since = "1.0.0")]
impl Debug for $ty {
fn fmt(&self, fmt: &mut Formatter<'_>) -> Result {
float_to_decimal_common(fmt, self, 1)
float_to_general_debug(fmt, self)
}
}

#[stable(feature = "rust1", since = "1.0.0")]
impl Display for $ty {
fn fmt(&self, fmt: &mut Formatter<'_>) -> Result {
float_to_decimal_common(fmt, self, 0)
float_to_decimal_display(fmt, self)
}
}

Expand Down
2 changes: 1 addition & 1 deletion library/core/src/num/f32.rs
Expand Up @@ -448,7 +448,7 @@ impl f32 {
// private use internally.
#[inline]
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
const fn abs_private(self) -> f32 {
pub(crate) const fn abs_private(self) -> f32 {
f32::from_bits(self.to_bits() & 0x7fff_ffff)
}

Expand Down
2 changes: 1 addition & 1 deletion library/core/src/num/f64.rs
Expand Up @@ -447,7 +447,7 @@ impl f64 {
// private use internally.
#[inline]
#[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
const fn abs_private(self) -> f64 {
pub(crate) const fn abs_private(self) -> f64 {
f64::from_bits(self.to_bits() & 0x7fff_ffff_ffff_ffff)
}

Expand Down
24 changes: 24 additions & 0 deletions library/core/tests/fmt/float.rs
Expand Up @@ -12,6 +12,16 @@ fn test_format_f64() {
assert_eq!("1.23456789E3", format!("{:E}", 1234.56789f64));
assert_eq!("0.0", format!("{:?}", 0.0f64));
assert_eq!("1.01", format!("{:?}", 1.01f64));

let high_cutoff = 1e16_f64;
assert_eq!("1e16", format!("{:?}", high_cutoff));
assert_eq!("-1e16", format!("{:?}", -high_cutoff));
assert!(!is_exponential(&format!("{:?}", high_cutoff * (1.0 - 2.0 * f64::EPSILON))));
assert_eq!("-3.0", format!("{:?}", -3f64));
assert_eq!("0.0001", format!("{:?}", 0.0001f64));
assert_eq!("9e-5", format!("{:?}", 0.00009f64));
assert_eq!("1234567.9", format!("{:.1?}", 1234567.89f64));
assert_eq!("1234.6", format!("{:.1?}", 1234.56789f64));
}

#[test]
Expand All @@ -28,4 +38,18 @@ fn test_format_f32() {
assert_eq!("1.2345679E3", format!("{:E}", 1234.56789f32));
assert_eq!("0.0", format!("{:?}", 0.0f32));
assert_eq!("1.01", format!("{:?}", 1.01f32));

let high_cutoff = 1e16_f32;
assert_eq!("1e16", format!("{:?}", high_cutoff));
assert_eq!("-1e16", format!("{:?}", -high_cutoff));
assert!(!is_exponential(&format!("{:?}", high_cutoff * (1.0 - 2.0 * f32::EPSILON))));
assert_eq!("-3.0", format!("{:?}", -3f32));
assert_eq!("0.0001", format!("{:?}", 0.0001f32));
assert_eq!("9e-5", format!("{:?}", 0.00009f32));
assert_eq!("1234567.9", format!("{:.1?}", 1234567.89f32));
assert_eq!("1234.6", format!("{:.1?}", 1234.56789f32));
}

fn is_exponential(s: &str) -> bool {
s.contains("e") || s.contains("E")
}

0 comments on commit 8731d4d

Please sign in to comment.