Configurable, precise and fast rust string parser to a Duration

Released API Docs | Changelog


GitHub branch checks state Crates.io docs.rs MSRV

Table of Contents

Overview

fundu provides a flexible and fast parser to convert rust strings into a [std::time::Duration] and for negative durations into a [time::Duration]. Some examples for valid input strings with the standard feature:

For examples of the custom feature see Customization section. A quick summary of features provided by this crate:

fundu aims for good performance and being a lightweight crate. It is purely built on top of the rust stdlib, and there are no additional dependencies required in the standard configuration. The accepted number format is per default the scientific floating point format and compatible with [f64::from_str]. However, the number format and other aspects can be customized up to formats like systemd time spans. See also the examples Examples section and the examples folder. For a direct comparison of fundu vs the rust native methods Duration::(try_)from_secs_f64 see Comparison.

For further details see the Documentation!

Installation

Add this to Cargo.toml for fundu with the standard feature.

toml [dependencies] fundu = "0.5.1"

fundu is split into two main features, standard (providing DurationParser and parse_duration) and custom (providing the CustomDurationParser). The first is described here in in detail, the latter adds fully customizable identifiers for time units. Most of the time only one of the parsers is needed. To include only the CustomDurationParser add the following to Cargo.toml:

toml [dependencies] fundu = { version = "0.5.1", default-features = false, features = ["custom"] }

Activating the negative feature allows parsing negative numbers to negative [time::Duration]s.

Examples

If only the default parser is required once, then the parse_duration method can be used.

```rust use fundu::parse_duration; use std::time::Duration;

let input = "1.0e2s"; asserteq!(parseduration(input).unwrap(), Duration::new(100, 0)); ```

When a customization of the accepted TimeUnits is required, then [DurationParser::with_time_units] can be used.

```rust use fundu::DurationParser; use fundu::TimeUnit::*; use std::time::Duration;

let input = "3m"; asserteq!( DurationParser::withtime_units(&[Minute]).parse(input).unwrap(), Duration::new(180, 0) ); ```

When no time units are configured, seconds is assumed.

```rust use fundu::DurationParser; use std::time::Duration;

let input = "1.0e2"; asserteq!( DurationParser::withouttime_units().parse(input).unwrap(), Duration::new(100, 0) ); ```

However, setting the default time unit to something different than seconds can be achieved with

```rust use fundu::{DurationParser, TimeUnit::*}; use std::time::Duration;

asserteq!( DurationParser::withouttimeunits() .defaultunit(MilliSecond) .parse("1000") .unwrap(), Duration::new(1, 0) ); ```

Note the following will return an error because y (Years) is not in the default set of TimeUnits.

```rust use fundu::DurationParser;

asserteq!( DurationParser::new().parse("3y").unwraperr().to_string(), "Time unit error: Invalid time unit: 'y' at column 1" ); ```

The parser is reusable and the set of time units is fully customizable

```rust use fundu::{DurationParser, TimeUnit::*}; use std::time::Duration;

let parser = DurationParser::withtimeunits(&[NanoSecond, Minute, Hour]); for (input, expected) in &[ ("9e3ns", Duration::new(0, 9000)), ("10m", Duration::new(600, 0)), ("1.1h", Duration::new(3960, 0)), ("7", Duration::new(7, 0)), ] { assert_eq!(parser.parse(input).unwrap(), *expected); } ```

The identifiers for time units can be fully customized with any number of valid utf-8 sequences if the custom feature is activated:

```rust use fundu::{CustomDurationParser, TimeUnit::*}; use std::time::Duration;

let parser = CustomDurationParser::withtimeunits( &[ (MilliSecond, &["χιλιοστό του δευτερολέπτου"]), (Second, &["s", "secs"]), (Hour, &["⏳"]) ] ); for (input, expected) in &[ (".3χιλιοστό του δευτερολέπτου", Duration::new(0, 300000)), ("1e3secs", Duration::new(1000, 0)), ("1.1⏳", Duration::new(3960, 0)), ] { asserteq!(parser.parse(input).unwrap(), *expected); } ```

It's also possible to parse multiple durations at once with parse_multiple. The different durations can be separated by an optional delimiter (a closure matching a u8) defined with parse_multiple. If the delimiter is not encountered, a number also indicates a new duration.

```rust use std::time::Duration;

use fundu::DurationParser;

let mut parser = DurationParser::new(); parser.parse_multiple(Some(|byte| matches!(byte, b' ' | b'\t')));

asserteq!(parser.parse("1.5h 2e+2ns"), Ok(Duration::new(5400, 200))); asserteq!(parser.parse("55s500ms"), Ok(Duration::new(55, 500000000))); asserteq!(parser.parse("1\t1"), Ok(Duration::new(2, 0))); asserteq!(parser.parse("1. .1"), Ok(Duration::new(1, 100000000))); asserteq!(parser.parse("2h"), Ok(Duration::new(2 * 60 * 60, 0))); asserteq!( parser.parse("300ms20s 5d"), Ok(Duration::new(5 * 60 * 60 * 24 + 20, 300000000)) ); ```

See also the examples folder for common recipes and integration with other crates. Run an example with

shell cargo run --example $FILE_NAME_WITHOUT_FILETYPE_SUFFIX

like the systemd time span parser example

```shell

For some of the examples a help is available. To pass arguments to the example itself separate the arguments for cargo and the example with --

$ cargo run --example systemd --features custom --no-default-features -- --help ...

To actually run the example execute

$ cargo run --example systemd --features custom --no-default-features '300ms20s 5day' Original: 300ms20s 5day μs: 432020300000 Human: 5d 20s 300ms ```

Time units

Second is the default time unit (if not specified otherwise for example with [DurationParser::default_unit]) which is applied when no time unit was encountered in the input string. The table below gives an overview of the constructor methods and which time units are available. If a custom set of time units is required, DurationParser::with_time_units can be used.

TimeUnit | Default identifier | Calculation | Default time unit ---:| ---:| ---:|:---: Nanosecond | ns | 1e-9s | ☑ Microsecond | Ms | 1e-6s | ☑ Millisecond | ms | 1e-3s | ☑ Second | s | SI definition | ☑ Minute | m | 60s | ☑ Hour | h | 60m | ☑ Day | d | 24h | ☑ Week | w | 7d | ☑ Month | M | Year / 12 | ☐ Year | y | 365.25d | ☐

Note that Months and Years are not included in the default set of time units. The current implementation uses an approximate calculation of Months and Years in seconds and if they are included in the final configuration, the Julian year based calculation is used. (See table above)

With the CustomDurationParser from the custom feature, the identifiers for time units can be fully customized.

Customization

Unlike other crates, fundu does not try to establish a standard for time units and their identifiers or a specific number format. So, a lot of these aspects can be adjusted with ease when initializing or building the parser. Here's an incomplete example for possible customizations of the number format:

```rust use std::time::Duration;

use fundu::TimeUnit::*; use fundu::{DurationParser, ParseError};

let parser = DurationParser::builder() // Use a custom set of time units. For demonstration purposes just NanoSecond = ns .customtimeunits(&[NanoSecond]) // Allow some whitespace characters as delimiter between the number and the time unit .allowdelimiter(|byte| matches!(byte, b'\t' | b'\n' | b'\r' | b' ')) // Makes the number optional. If no number was encountered 1 is assumed .numberisoptional() // Disable parsing the fractional part of the number => 1.0 will return an error .disablefraction() // Disable parsing the exponent => 1e0 will return an error .disable_exponent() // Finally, build a reusable DurationParser .build();

// Some valid input for (input, expected) in &[ ("ns", Duration::new(0, 1)), ("1000\t\n\r ns", Duration::new(0, 1000)), ] { assert_eq!(parser.parse(input).unwrap(), *expected); }

// Some invalid input for (input, expected) in &[ ( "1.0ns", ParseError::Syntax(1, "No fraction allowed".tostring()), ), ( "1e9ns", ParseError::Syntax(1, "No exponent allowed".tostring()), ), ] { asserteq!(parser.parse(input).unwraperr(), *expected); } ```

Here's an example for fully-customizable time units which uses the CustomDurationParser from the custom feature:

```rust use std::time::Duration;

use fundu::TimeUnit::*; use fundu::{CustomDurationParser, Multiplier};

let mut parser = CustomDurationParser::withtimeunits(&[ (Second, &["s", "secs", "seconds"]), (Minute, &["min"]), (Hour, &["ώρα"]), ]);

// Let's define a custom time unit fortnight == 2 weeks which isn't part of the basic // [TimeUnit]s: parser.customtimeunit(Week, Multiplier(2, 0), &["f", "fortnight", "fortnights"]);

asserteq!(parser.parse("42e-1ώρα").unwrap(), Duration::new(15120, 0)); asserteq!( parser.parse("1fortnight").unwrap(), Duration::new(60 * 60 * 24 * 7 * 2, 0) ); ```

Benchmarks

To run the benchmarks on your machine, clone the repository

shell git clone https://github.com/Joining7943/fundu.git cd fundu

and then run all benchmarks with

shell cargo bench --all-features

The iai-callgrind (feature = with-iai) and flamegraph (feature = with-flamegraph) benchmarks can only be run on unix. Use the --features option of cargo to run the benchmarks for specific features:

shell cargo bench --features standard,custom,negative

The above won't run the flamegraph and iai-callgrind benchmarks.

Benchmarks can be further filtered for example with

shell cargo bench --bench benchmarks_standard cargo bench --bench benchmarks_standard -- 'parsing speed' cargo bench --features custom --no-default-features --bench benchmarks_custom

For more infos, see the help with

shell cargo bench --help # The cargo help for bench cargo bench --bench benchmarks_standard -- --help # The criterion help

To get a rough idea about the parsing times, here the average parsing speed of some inputs on a comparatively slow machine (Quad core 3000Mhz, 8GB DDR3, Linux)

Input | avg parsing time | ~ samples / s --- | ---:| ---: 1 | 37.925 ns | 26_367_831.245 123456789.123456789 | 73.162 ns | 13_668_297.750 format!("{}.{}e-1022", "1".repeat(1022), "1".repeat(1022)) | 551.59 ns | 1_812_940.771

For comparison, fundu's precision and additional features only add a very low performance overhead for small and some mixed input and performs better than the reference function from the stdlib as the input gets larger (the reference function is Duration::from_secs_f64(input.parse().unwrap())):

Input | avg parsing time | ~ samples / s --- | ---:| ---: 1 | 25.630 ns | 39_016_777.214 123456789.123456789 | 45.007 ns | 22_218_765.969 format!("{}.{}e-1022", "1".repeat(1022), "1".repeat(1022)) | 1.7457 µs | 572_836.111

The initialization for fixed size time unit sets with DurationParser::new, DurationParser::with_all_time_units takes around 1-2 ns and is negligibly small. The initialization time for custom sets with DurationParser::with_time_units has a maximum of around 10 ns.

Comparison fundu vs Duration::from_secs_f64

Here's a short incomplete overview of differences and advantages of fundu over using Duration::from_secs_f64(input.parse().unwrap()) (and Duration::try_from_secs_f64(input.parse().unwrap()))

Input | Result fundu | Result Duration::(try_)from_secs_f64 ---:| --- | --- 01271480964981728917.1 | Duration::new(1_271_480_964_981_728_917, 100_000_000) | Duration::new(1_271_480_964_981_729_024, 0) 1.11111111111e10 | Duration::new(11_111_111_111, 100_000_000) | Duration::new(11_111_111_111, 100_000_381) 1ns | Duration::new(0, 1) | cannot parse time units 1000 | When changing the default unit to MilliSecond -> Duration::new(1, 0) | is always seconds based 1e20 | Duration::MAX | panics or returns an error due to: can not convert float seconds to Duration: value is either too big or NaN infinity | Duration::MAX | panics or returns an error due to: can not convert float seconds to Duration: value is either too big or NaN

fundu has a small impact on performance when the input is small but performs better for large input (See performance). Depending on the input data and if you need to parse a massive amount of inputs and don't need the full precision or any of fundu's features, you may prefer using the native methods from the rust stdlib.

Platform support

Since fundu is purely built on top of the rust stdlib without platform specific code, this library should be compatible with all platforms. Please open an issue if you find any unsupported platforms which rust itself supports.

See also the CI

TODO

See also Changelog

License

MIT license (LICENSE or http://opensource.org/licenses/MIT)