#string-parser #user-agent

ua-parser

Rust implementation of the User Agent String Parser project

2 unstable releases

0.2.0 Jun 22, 2024
0.1.0 Jun 17, 2024

#271 in HTTP client

Download history 146/week @ 2024-06-21 4/week @ 2024-06-28 3/week @ 2024-07-05 16/week @ 2024-07-26 2/week @ 2024-08-02 26/week @ 2024-09-13 63/week @ 2024-09-20 64/week @ 2024-09-27 11/week @ 2024-10-04

164 downloads per month

Apache-2.0

1.5MB
2K SLoC

Rust 1.5K SLoC // 0.1% comments JavaScript 393 SLoC // 0.0% comments C++ 91 SLoC Perl 29 SLoC // 0.2% comments

User Agent Parser

This module implements the browserscope / uap standard for rust, allowing the extraction of various metadata from user agents.

The browserscope standard is data-oriented, with regexes.yaml specifying the matching and extraction from user-agent strings. This library implements the maching protocols and provides various types to make loading the dataset easier, however it does not provide the data itself, to avoid dependencies on serialization libraries or constrain loading.

Dataset loading

The crate does not provide any sort of precompiled data file, or dedicated loader, however Regexes implements serde::Deserialize and can load a regexes.yaml file or any format-preserving conversion thereof (e.g. loading from json or cbor might be preferred if the application already depends on one of those):

# let ua_str = "";
let f = std::fs::File::open("regexes.yaml")?;
let regexes: ua_parser::Regexes = serde_yaml::from_reader(f)?;
let extractor = ua_parser::Extractor::try_from(regexes)?;

# Ok::<(), Box<dyn std::error::Error>>(())

All the data-description structures are also Plain Old Data, so they can be embedded in the application directly e.g. via a build script:

let parsers = vec![
    ua_parser::user_agent::Parser {
        regex: "foo".into(),
        family_replacement: Some("bar".into()),
        ..Default::default()
    }
];

Extraction

The crate provides the ability to either extract individual information sets (user agent — browser, OS, and device) or extract all three in a single call.

The three infosets are are independent and non-overlapping so while the full extractor may be convenient if only one is needed a complete extraction is unnecessary overhead, and the extractors themselves are somewhat costly to create and take up memory.

Complete Extractor

For the complete extractor, it is simply converted from the Regexes structure. The resulting Extractor embeds all three module-level extractors as attributes, and Extractor::extract-s into a 3-uple of ValueRefs.

Individual Extractors

The individual extractors are in the user_agent, [os], and device modules, the three modules follow the exact same model:

  • a Parser struct which specifies individual parser configurations, used as inputs to the Builder
  • a Builder, into which the relevant parsers can be push-ed
  • an Extractor created from the Builder, from which the user can extract a ValueRef
  • the ValueRef result of data extraction, which may borrow from (and is thus lifetime-bound to) the Parser substitution data and the user agent string it was extracted from
  • for convenience, an owned Value variant of the ValueRef
use ua_parser::os::{Builder, Parser, ValueRef};

let e = Builder::new()
    .push(Parser {
        regex: r"(Android)[ \-/](\d+)(?:\.(\d+)|)(?:[.\-]([a-z0-9]+)|)".into(),
        ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) Donut".into(),
        os_v1_replacement: Some("1".into()),
        os_v2_replacement: Some("2".into()),
        ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) Eclair".into(),
        os_v1_replacement: Some("2".into()),
        os_v2_replacement: Some("1".into()),
        ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) Froyo".into(),
        os_v1_replacement: Some("2".into()),
        os_v2_replacement: Some("2".into()),
        ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) Gingerbread".into(),
        os_v1_replacement: Some("2".into()),
        os_v2_replacement: Some("3".into()),
        ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) Honeycomb".into(),
        os_v1_replacement: Some("3".into()),
       ..Default::default()
    })?
    .push(Parser {
        regex: r"(Android) (\d+);".into(),
        ..Default::default()
    })?
    .build()?;

assert_eq!(
    e.extract("Android Donut"),
    Some(ValueRef {
        os: "Android".into(),
        major: Some("1".into()),
        minor: Some("2".into()),
        ..Default::default()
    }),
);
assert_eq!(
    e.extract("Android 15"),
    Some(ValueRef { os: "Android".into(), major: Some("15".into()), ..Default::default()}),
);
assert_eq!(
    e.extract("ZuneWP7"),
    None,
);
# Ok::<(), Box<dyn std::error::Error>>(())

Performances

The package has not been profiled or optimised yet, but it seems rather competitive with uap-cpp (tested on an M1 Pro MBP):

> ./UaParserBench ../uap-core/regexes.yaml benchmarks/useragents.txt 10
   25.13s user 0.07s system 99% cpu 25.279 total
> ./UaParserBench ../uap-core/regexes.yaml ../uap-python/samples/useragents.txt 100
  246.49s user 0.47s system 99% cpu 4:07.55 total
> target/release/examples/bench -r 10 ../uap-core/regexes.yaml ../uap-cpp/benchmarks/useragents.txt
   10.10s user 0.04s system 99% cpu 10.169 total

> target/release/examples/bench -r 100 ../uap-core/regexes.yaml ../uap-python/samples/useragents.txt
   98.46s user 0.04s system 99% cpu 1:38.73 total

Dependencies

~4–6MB
~109K SLoC