mz_expr/scalar/func/impls/

float64.rs

// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.

use chrono::{DateTime, Utc};
use mz_expr_derive::sqlfunc;
use mz_ore::cast::TryCastFrom;
use mz_repr::adt::numeric::{self, Numeric, NumericMaxScale};
use mz_repr::adt::timestamp::CheckedTimestamp;
use mz_repr::{SqlScalarType, strconv};
use serde::{Deserialize, Serialize};

use crate::EvalError;
use crate::scalar::DomainLimit;

#[sqlfunc(
    sqlname = "-",
    preserves_uniqueness = false,
    inverse = NegFloat64,
    is_monotone = true
)]
fn neg_float64(a: f64) -> f64 {
    -a
}

#[sqlfunc(sqlname = "abs")]
fn abs_float64(a: f64) -> f64 {
    a.abs()
}

#[sqlfunc(sqlname = "roundf64")]
fn round_float64(a: f64) -> f64 {
    a.round_ties_even()
}

#[sqlfunc(sqlname = "truncf64")]
fn trunc_float64(a: f64) -> f64 {
    a.trunc()
}

#[sqlfunc(sqlname = "ceilf64")]
fn ceil_float64(a: f64) -> f64 {
    a.ceil()
}

#[sqlfunc(sqlname = "floorf64")]
fn floor_float64(a: f64) -> f64 {
    a.floor()
}

#[sqlfunc(
    sqlname = "double_to_smallint",
    preserves_uniqueness = false,
    inverse = super::CastInt16ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_int16(a: f64) -> Result<i16, EvalError> {
    let f = round_float64(a);
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= (i16::MIN as f64)) && (f < -(i16::MIN as f64)) {
        Ok(f as i16)
    } else {
        Err(EvalError::Int16OutOfRange(f.to_string().into()))
    }
}

#[sqlfunc(
    sqlname = "double_to_integer",
    preserves_uniqueness = false,
    inverse = super::CastInt32ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_int32(a: f64) -> Result<i32, EvalError> {
    let f = round_float64(a);
    // This condition is delicate because i32::MIN can be represented exactly by
    // an f64 but not i32::MAX. We follow PostgreSQL's approach here.
    //
    // See: https://github.com/postgres/postgres/blob/ca3b37487/src/include/c.h#L1074-L1096
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= (i32::MIN as f64)) && (f < -(i32::MIN as f64)) {
        Ok(f as i32)
    } else {
        Err(EvalError::Int32OutOfRange(f.to_string().into()))
    }
}

#[sqlfunc(
    sqlname = "f64toi64",
    preserves_uniqueness = false,
    inverse = super::CastInt64ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_int64(a: f64) -> Result<i64, EvalError> {
    let f = round_float64(a);
    // This condition is delicate because i64::MIN can be represented exactly by
    // an f64 but not i64::MAX. We follow PostgreSQL's approach here.
    //
    // See: https://github.com/postgres/postgres/blob/ca3b37487/src/include/c.h#L1074-L1096
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= (i64::MIN as f64)) && (f < -(i64::MIN as f64)) {
        Ok(f as i64)
    } else {
        Err(EvalError::Int64OutOfRange(f.to_string().into()))
    }
}

#[sqlfunc(
    sqlname = "double_to_real",
    preserves_uniqueness = false,
    inverse = super::CastFloat32ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_float32(a: f64) -> Result<f32, EvalError> {
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    let result = a as f32;
    if result.is_infinite() && !a.is_infinite() {
        Err(EvalError::FloatOverflow)
    } else if result == 0.0 && a != 0.0 {
        Err(EvalError::FloatUnderflow)
    } else {
        Ok(result)
    }
}

#[sqlfunc(
    sqlname = "double_to_text",
    preserves_uniqueness = false,
    inverse = super::CastStringToFloat64
)]
fn cast_float64_to_string(a: f64) -> String {
    let mut s = String::new();
    strconv::format_float64(&mut s, a);
    s
}

#[sqlfunc(
    sqlname = "double_to_uint2",
    preserves_uniqueness = false,
    inverse = super::CastUint16ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_uint16(a: f64) -> Result<u16, EvalError> {
    let f = round_float64(a);
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= 0.0) && (f <= (u16::MAX as f64)) {
        Ok(f as u16)
    } else {
        Err(EvalError::UInt16OutOfRange(f.to_string().into()))
    }
}

#[sqlfunc(
    sqlname = "double_to_uint4",
    preserves_uniqueness = false,
    inverse = super::CastUint32ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_uint32(a: f64) -> Result<u32, EvalError> {
    let f = round_float64(a);
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= 0.0) && (f <= (u32::MAX as f64)) {
        Ok(f as u32)
    } else {
        Err(EvalError::UInt32OutOfRange(f.to_string().into()))
    }
}

#[sqlfunc(
    sqlname = "double_to_uint8",
    preserves_uniqueness = false,
    inverse = super::CastUint64ToFloat64,
    is_monotone = true
)]
fn cast_float64_to_uint64(a: f64) -> Result<u64, EvalError> {
    let f = round_float64(a);
    // This condition is delicate because u64::MAX cannot be represented exactly by
    // an f64: `u64::MAX as f64` rounds up to 2^64. A `<=` bound would therefore let
    // 2^64 pass, and the subsequent `as u64` cast saturates it to u64::MAX. Comparing
    // with `<` keeps 2^64 (and larger) out of range, mirroring the f64 -> i64 cast.
    // TODO(benesch): remove potentially dangerous usage of `as`.
    #[allow(clippy::as_conversions)]
    if (f >= 0.0) && (f < (u64::MAX as f64)) {
        Ok(f as u64)
    } else {
        Err(EvalError::UInt64OutOfRange(f.to_string().into()))
    }
}

#[derive(
    Ord,
    PartialOrd,
    Clone,
    Debug,
    Eq,
    PartialEq,
    Serialize,
    Deserialize,
    Hash
)]
pub struct CastFloat64ToNumeric(pub Option<NumericMaxScale>);

#[sqlfunc(
    CastFloat64ToNumeric,
    sqlname = "double_to_numeric",
    inverse = super::CastNumericToFloat64,
    is_monotone = true,
    output_type_expr = SqlScalarType::Numeric { max_scale: self.0 }
        .nullable(input_type.nullable)
)]
fn cast_float64_to_numeric(&self, a: f64) -> Result<Numeric, EvalError> {
    if a.is_infinite() {
        return Err(EvalError::InfinityOutOfDomain(
            "casting double precision to numeric".into(),
        ));
    }
    let mut a = Numeric::from(a);
    if let Some(scale) = self.0 {
        if numeric::rescale(&mut a, scale.into_u8()).is_err() {
            return Err(EvalError::NumericFieldOverflow);
        }
    }
    match numeric::munge_numeric(&mut a) {
        Ok(_) => Ok(a),
        Err(_) => Err(EvalError::NumericFieldOverflow),
    }
}

#[sqlfunc(sqlname = "sqrtf64")]
fn sqrt_float64(a: f64) -> Result<f64, EvalError> {
    if a < 0.0 {
        return Err(EvalError::NegSqrt);
    }
    Ok(a.sqrt())
}

#[sqlfunc(sqlname = "cbrtf64")]
fn cbrt_float64(a: f64) -> f64 {
    a.cbrt()
}

#[sqlfunc]
fn cos(a: f64) -> Result<f64, EvalError> {
    if a.is_infinite() {
        return Err(EvalError::InfinityOutOfDomain("cos".into()));
    }
    Ok(a.cos())
}

#[sqlfunc]
fn acos(a: f64) -> Result<f64, EvalError> {
    if a < -1.0 || 1.0 < a {
        return Err(EvalError::OutOfDomain(
            DomainLimit::Inclusive(-1),
            DomainLimit::Inclusive(1),
            "acos".into(),
        ));
    }
    Ok(a.acos())
}

#[sqlfunc]
fn cosh(a: f64) -> f64 {
    a.cosh()
}

#[sqlfunc]
fn acosh(a: f64) -> Result<f64, EvalError> {
    if a < 1.0 {
        return Err(EvalError::OutOfDomain(
            DomainLimit::Inclusive(1),
            DomainLimit::None,
            "acosh".into(),
        ));
    }
    Ok(a.acosh())
}

#[sqlfunc]
fn sin(a: f64) -> Result<f64, EvalError> {
    if a.is_infinite() {
        return Err(EvalError::InfinityOutOfDomain("sin".into()));
    }
    Ok(a.sin())
}

#[sqlfunc]
fn asin(a: f64) -> Result<f64, EvalError> {
    if a < -1.0 || 1.0 < a {
        return Err(EvalError::OutOfDomain(
            DomainLimit::Inclusive(-1),
            DomainLimit::Inclusive(1),
            "asin".into(),
        ));
    }
    Ok(a.asin())
}

#[sqlfunc]
fn sinh(a: f64) -> f64 {
    a.sinh()
}

#[sqlfunc]
fn asinh(a: f64) -> f64 {
    a.asinh()
}

#[sqlfunc]
fn tan(a: f64) -> Result<f64, EvalError> {
    if a.is_infinite() {
        return Err(EvalError::InfinityOutOfDomain("tan".into()));
    }
    Ok(a.tan())
}

#[sqlfunc]
fn atan(a: f64) -> f64 {
    a.atan()
}

#[sqlfunc]
fn tanh(a: f64) -> f64 {
    a.tanh()
}

#[sqlfunc]
fn atanh(a: f64) -> Result<f64, EvalError> {
    if a < -1.0 || 1.0 < a {
        return Err(EvalError::OutOfDomain(
            DomainLimit::Inclusive(-1),
            DomainLimit::Inclusive(1),
            "atanh".into(),
        ));
    }
    Ok(a.atanh())
}

#[sqlfunc]
fn cot(a: f64) -> Result<f64, EvalError> {
    if a.is_infinite() {
        return Err(EvalError::InfinityOutOfDomain("cot".into()));
    }
    Ok(1.0 / a.tan())
}

#[sqlfunc]
fn radians(a: f64) -> f64 {
    a.to_radians()
}

#[sqlfunc]
fn degrees(a: f64) -> f64 {
    a.to_degrees()
}

#[sqlfunc(sqlname = "log10f64")]
fn log10(a: f64) -> Result<f64, EvalError> {
    if a.is_sign_negative() {
        return Err(EvalError::NegativeOutOfDomain("log10".into()));
    }
    if a == 0.0 {
        return Err(EvalError::ZeroOutOfDomain("log10".into()));
    }
    Ok(a.log10())
}

#[sqlfunc(sqlname = "lnf64")]
fn ln(a: f64) -> Result<f64, EvalError> {
    if a.is_sign_negative() {
        return Err(EvalError::NegativeOutOfDomain("ln".into()));
    }
    if a == 0.0 {
        return Err(EvalError::ZeroOutOfDomain("ln".into()));
    }
    Ok(a.ln())
}

#[sqlfunc(sqlname = "expf64")]
fn exp(a: f64) -> Result<f64, EvalError> {
    let r = a.exp();
    if r.is_infinite() {
        return Err(EvalError::FloatOverflow);
    }
    if r == 0.0 {
        return Err(EvalError::FloatUnderflow);
    }
    Ok(r)
}

#[sqlfunc(sqlname = "mz_sleep")]
fn sleep(a: f64) -> Result<Option<CheckedTimestamp<DateTime<Utc>>>, EvalError> {
    let duration = std::time::Duration::try_from_secs_f64(a).map_err(|_| {
        let mut val = String::new();
        strconv::format_float64(&mut val, a);
        EvalError::InvalidParameterValue(format!("cannot sleep for {val} seconds").into())
    })?;
    std::thread::sleep(duration);
    Ok(None)
}

#[sqlfunc(sqlname = "tots")]
fn to_timestamp(f: f64) -> Result<CheckedTimestamp<DateTime<Utc>>, EvalError> {
    const NANO_SECONDS_PER_SECOND: i64 = 1_000_000_000;
    if f.is_nan() {
        Err(EvalError::TimestampCannotBeNan)
    } else if f.is_infinite() {
        // TODO(jkosh44) implement infinite timestamps
        Err(EvalError::TimestampOutOfRange)
    } else {
        let mut secs = i64::try_cast_from(f.trunc()).ok_or(EvalError::TimestampOutOfRange)?;
        // NOTE(benesch): PostgreSQL has microsecond precision in its timestamps,
        // while chrono has nanosecond precision. While we normally accept
        // nanosecond precision, here we round to the nearest microsecond because
        // f64s lose quite a bit of accuracy in the nanosecond digits when dealing
        // with common Unix timestamp values (> 1 billion).
        let microsecs = (f.fract() * 1_000_000.0).round();
        let mut nanosecs =
            i64::try_cast_from(microsecs * 1_000.0).ok_or(EvalError::TimestampOutOfRange)?;
        if nanosecs < 0 {
            secs = secs.checked_sub(1).ok_or(EvalError::TimestampOutOfRange)?;
            nanosecs = NANO_SECONDS_PER_SECOND
                .checked_add(nanosecs)
                .ok_or(EvalError::TimestampOutOfRange)?;
        }
        // Ensure `nanosecs` is less than 1 second.
        secs = secs
            .checked_add(nanosecs / NANO_SECONDS_PER_SECOND)
            .ok_or(EvalError::TimestampOutOfRange)?;
        nanosecs %= NANO_SECONDS_PER_SECOND;
        let nanosecs = u32::try_from(nanosecs).map_err(|_| EvalError::TimestampOutOfRange)?;
        match DateTime::from_timestamp(secs, nanosecs) {
            Some(dt) => {
                CheckedTimestamp::from_timestamplike(dt).map_err(|_| EvalError::TimestampOutOfRange)
            }
            None => Err(EvalError::TimestampOutOfRange),
        }
    }
}