Files
nushell/crates/nu-command/tests/commands/math/mod.rs
Bahex 5478ec44bb to <format>: preserve round float numbers' type (#16016)
- fixes #16011

# Description
`Display` implementation for `f64` omits the decimal part for round
numbers, and by using it we did the same.
This affected:
- conversions to delimited formats: `csv`, `tsv`
- textual formats: `html`, `md`, `text`
- pretty printed `json` (`--raw` was unaffected)
- how single float values are displayed in the REPL

> [!TIP]
> This PR fixes our existing json pretty printing implementation.
> We can likely switch to using serde_json's impl using its
PrettyFormatter which allows arbitrary indent strings.

# User-Facing Changes
- Round trips through `csv`, `tsv`, and `json` preserve the type of
round floats.
- It's always clear whether a number is an integer or a float in the
REPL
  ```nushell
  4 / 2
  # => 2  # before: is this an int or a float?

  4 / 2
  # => 2.0  # after: clearly a float
  ``` 

# Tests + Formatting
Adjusted tests for the new behavior.

- 🟢 toolkit fmt
- 🟢 toolkit clippy
- 🟢 toolkit test
- 🟢 toolkit test stdlib

# After Submitting
N/A

---------

Co-authored-by: Bahex <17417311+Bahex@users.noreply.github.com>
2025-06-26 15:15:19 -05:00

500 lines
8.4 KiB
Rust

mod abs;
mod avg;
mod ceil;
mod floor;
mod log;
mod max;
mod median;
mod min;
mod mode;
mod product;
mod round;
mod sqrt;
mod stddev;
mod sum;
mod variance;
use nu_test_support::{nu, pipeline};
#[test]
fn one_arg() {
let actual = nu!(pipeline(
r#"
1
"#
));
assert_eq!(actual.out, "1");
}
#[test]
fn add() {
let actual = nu!(pipeline(
r#"
1 + 1
"#
));
assert_eq!(actual.out, "2");
}
#[test]
fn add_compound() {
let actual = nu!(pipeline(
r#"
1 + 2 + 2
"#
));
assert_eq!(actual.out, "5");
}
#[test]
fn precedence_of_operators() {
let actual = nu!(pipeline(
r#"
1 + 2 * 2
"#
));
assert_eq!(actual.out, "5");
}
#[test]
fn precedence_of_operators2() {
let actual = nu!(pipeline(
r#"
1 + 2 * 2 + 1
"#
));
assert_eq!(actual.out, "6");
}
#[test]
fn precedence_of_operators3() {
let actual = nu!(pipeline(
r#"
5 - 5 * 10 + 5
"#
));
assert_eq!(actual.out, "-40");
}
#[test]
fn precedence_of_operators4() {
let actual = nu!(pipeline(
r#"
5 - (5 * 10) + 5
"#
));
assert_eq!(actual.out, "-40");
}
#[test]
fn division_of_ints() {
let actual = nu!(pipeline(
r#"
4 / 2
"#
));
assert_eq!(actual.out, "2.0");
}
#[test]
fn division_of_ints2() {
let actual = nu!(pipeline(
r#"
1 / 4
"#
));
assert_eq!(actual.out, "0.25");
}
#[test]
fn error_zero_division_int_int() {
let actual = nu!(pipeline(
r#"
1 / 0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_division_float_int() {
let actual = nu!(pipeline(
r#"
1.0 / 0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_division_int_float() {
let actual = nu!(pipeline(
r#"
1 / 0.0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_division_float_float() {
let actual = nu!(pipeline(
r#"
1.0 / 0.0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn floor_division_of_ints() {
let actual = nu!(pipeline(
r#"
5 // 2
"#
));
assert_eq!(actual.out, "2");
}
#[test]
fn floor_division_of_ints2() {
let actual = nu!(pipeline(
r#"
-3 // 2
"#
));
assert_eq!(actual.out, "-2");
}
#[test]
fn floor_division_of_floats() {
let actual = nu!(pipeline(
r#"
-3.0 // 2.0
"#
));
assert_eq!(actual.out, "-2.0");
}
#[test]
fn error_zero_floor_division_int_int() {
let actual = nu!(pipeline(
r#"
1 // 0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_floor_division_float_int() {
let actual = nu!(pipeline(
r#"
1.0 // 0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_floor_division_int_float() {
let actual = nu!(pipeline(
r#"
1 // 0.0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn error_zero_floor_division_float_float() {
let actual = nu!(pipeline(
r#"
1.0 // 0.0
"#
));
assert!(actual.err.contains("division by zero"));
}
#[test]
fn proper_precedence_history() {
let actual = nu!(pipeline(
r#"
2 / 2 / 2 + 1
"#
));
assert_eq!(actual.out, "1.5");
}
#[test]
fn parens_precedence() {
let actual = nu!(pipeline(
r#"
4 * (6 - 3)
"#
));
assert_eq!(actual.out, "12");
}
#[test]
fn modulo() {
let actual = nu!(pipeline(
r#"
9 mod 2
"#
));
assert_eq!(actual.out, "1");
}
#[test]
fn floor_div_mod() {
let actual = nu!("let q = 8 // -3; let r = 8 mod -3; 8 == $q * -3 + $r");
assert_eq!(actual.out, "true");
let actual = nu!("let q = -8 // 3; let r = -8 mod 3; -8 == $q * 3 + $r");
assert_eq!(actual.out, "true");
}
#[test]
fn floor_div_mod_overflow() {
let actual = nu!(format!("{} // -1", i64::MIN));
assert!(actual.err.contains("overflow"));
let actual = nu!(format!("{} mod -1", i64::MIN));
assert!(actual.err.contains("overflow"));
}
#[test]
fn floor_div_mod_zero() {
let actual = nu!("1 // 0");
assert!(actual.err.contains("zero"));
let actual = nu!("1 mod 0");
assert!(actual.err.contains("zero"));
}
#[test]
fn floor_div_mod_large_num() {
let actual = nu!(format!("{} // {}", i64::MAX, i64::MAX / 2));
assert_eq!(actual.out, "2");
let actual = nu!(format!("{} mod {}", i64::MAX, i64::MAX / 2));
assert_eq!(actual.out, "1");
}
#[test]
fn unit_multiplication_math() {
let actual = nu!("1MB * 2");
assert_eq!(actual.out, "2.0 MB");
}
#[test]
fn unit_multiplication_float_math() {
let actual = nu!("1MB * 1.2");
assert_eq!(actual.out, "1.2 MB");
}
#[test]
fn unit_float_floor_division_math() {
let actual = nu!("1MB // 3.0");
assert_eq!(actual.out, "333.3 kB");
}
#[test]
fn unit_division_math() {
let actual = nu!("1MB / 4");
assert_eq!(actual.out, "250.0 kB");
}
#[test]
fn unit_float_division_math() {
let actual = nu!("1MB / 3.2");
assert_eq!(actual.out, "312.5 kB");
}
#[test]
fn duration_math() {
let actual = nu!(pipeline(
r#"
1wk + 1day
"#
));
assert_eq!(actual.out, "1wk 1day");
}
#[test]
fn duration_decimal_math() {
let actual = nu!(pipeline(
r#"
5.5day + 0.5day
"#
));
assert_eq!(actual.out, "6day");
}
#[test]
fn duration_math_with_nanoseconds() {
let actual = nu!(pipeline(
r#"
1wk + 10ns
"#
));
assert_eq!(actual.out, "1wk 10ns");
}
#[test]
fn duration_decimal_math_with_nanoseconds() {
let actual = nu!(pipeline(
r#"
1.5wk + 10ns
"#
));
assert_eq!(actual.out, "1wk 3day 12hr 10ns");
}
#[test]
fn duration_decimal_math_with_all_units() {
let actual = nu!(pipeline(
r#"
1wk + 3day + 8hr + 10min + 16sec + 121ms + 11us + 12ns
"#
));
assert_eq!(actual.out, "1wk 3day 8hr 10min 16sec 121ms 11µs 12ns");
}
#[test]
fn duration_decimal_dans_test() {
let actual = nu!(pipeline(
r#"
3.14sec
"#
));
assert_eq!(actual.out, "3sec 140ms");
}
#[test]
fn duration_math_with_negative() {
let actual = nu!(pipeline(
r#"
1day - 1wk
"#
));
assert_eq!(actual.out, "-6day");
}
#[test]
fn compound_comparison() {
let actual = nu!(pipeline(
r#"
4 > 3 and 2 > 1
"#
));
assert_eq!(actual.out, "true");
}
#[test]
fn compound_comparison2() {
let actual = nu!(pipeline(
r#"
4 < 3 or 2 > 1
"#
));
assert_eq!(actual.out, "true");
}
#[test]
fn compound_where() {
let actual = nu!(pipeline(
r#"
echo '[{"a": 1, "b": 1}, {"a": 2, "b": 1}, {"a": 2, "b": 2}]' | from json | where a == 2 and b == 1 | to json -r
"#
));
assert_eq!(actual.out, r#"[{"a":2,"b":1}]"#);
}
#[test]
fn compound_where_paren() {
let actual = nu!(pipeline(
r#"
echo '[{"a": 1, "b": 1}, {"a": 2, "b": 1}, {"a": 2, "b": 2}]' | from json | where ($it.a == 2 and $it.b == 1) or $it.b == 2 | to json -r
"#
));
assert_eq!(actual.out, r#"[{"a":2,"b":1},{"a":2,"b":2}]"#);
}
// TODO: these ++ tests are not really testing *math* functionality, maybe find another place for them
#[test]
fn concat_lists() {
let actual = nu!(pipeline(
r#"
[1 3] ++ [5 6] | to nuon
"#
));
assert_eq!(actual.out, "[1, 3, 5, 6]");
}
#[test]
fn concat_tables() {
let actual = nu!(pipeline(
r#"
[[a b]; [1 2]] ++ [[c d]; [10 11]] | to nuon
"#
));
assert_eq!(actual.out, "[{a: 1, b: 2}, {c: 10, d: 11}]");
}
#[test]
fn concat_strings() {
let actual = nu!(pipeline(
r#"
"foo" ++ "bar"
"#
));
assert_eq!(actual.out, "foobar");
}
#[test]
fn concat_binary_values() {
let actual = nu!(pipeline(
r#"
0x[01 02] ++ 0x[03 04] | to nuon
"#
));
assert_eq!(actual.out, "0x[01020304]");
}