Wax is a Rust library that provides opinionated and portable globs that can be matched against file paths and directory trees. Globs use a familiar syntax and support expressive features with semantics that emphasize component boundaries.

Match a path against a glob:

use wax::{Glob, Program};

let glob = Glob::new("*.png").unwrap();
assert!(glob.is_match("logo.png"));

Match a path against a glob with matched text (captures):

use wax::{CandidatePath, Glob, Program};

let glob = Glob::new("**/{*.{go,rs}}").unwrap();

let path = CandidatePath::from("src/main.go");
let matched = glob.matched(&path).unwrap();

assert_eq!("main.go", matched.get(2).unwrap());

Match a directory tree against a glob:

use wax::Glob;

let glob = Glob::new("**/*.{md,txt}").unwrap();
for entry in glob.walk("doc") {
    let entry = entry.unwrap();
    // ...
}

Match a directory tree against a glob with negations:

use wax::walk::{FileIterator, LinkBehavior};
use wax::Glob;

let glob = Glob::new("**/*.{md,txt}").unwrap();
for entry in glob
    .walk_with_behavior("doc", LinkBehavior::ReadTarget)
    .not("**/secret/**")
    .unwrap()
{
    let entry = entry.unwrap();
    // ...
}

Match a path against multiple globs:

use wax::{Glob, Program};

let any = wax::any([
    "src/**/*.rs",
    "tests/**/*.rs",
    "doc/**/*.md",
    "pkg/**/PKGBUILD",
]).unwrap();
assert!(any.is_match("src/token/mod.rs"));

See more details below.

Globs are encoded as UTF-8 strings called glob expressions that resemble Unix paths consisting of nominal components delimited by separators. The most fundamental type in the Wax API is Glob, which is constructed from a glob expression via inherent functions or standard conversion traits. Data is borrowed where possible in most APIs, but can be copied into owned instances using an into_owned method with most types.

use wax::Glob;

let glob = Glob::new("site/img/logo.svg").unwrap();

Not only are APIs designed for portability, but so too are glob expressions. Regardless of platform or operating system, globs support the same features and use the same syntax. Glob expressions are distinct from paths, which differ in syntax and features on each platform.

In glob expressions, forward slash / is the only path component separator and back slashes \ are forbidden (back slash is used for escape sequences, but the literal sequence \\ is not supported). This means that it is impossible to represent \ in nominal path components, but this character is generally forbidden as such and its disuse avoids confusion.

Globs enforce various rules regarding meta-characters, patterns, and component boundaries that reject nonsense expressions. While these rules can sometimes make glob expressions a bit more difficult to compose, they also make glob expressions more consistent, easier to reason about, and less prone to errors.

Globs resemble Unix paths, but additionally support patterns that can be matched against paths and directory trees. Patterns use a syntax that resembles globbing in Unix shells and tools like git, though there are some important differences.

use wax::Glob;

let glob = Glob::new("**/*.{go,rs}").unwrap();
assert!(glob.is_match("src/lib.rs"));

Patterns form captures that can be used to extract matched text (as seen in many regular expression engines). In the above example, there are three patterns that can be queried for matched text: **/, *, and {go,rs}. Every glob expression has an implicit capture for the complete matched text.

Globs use a consistent and opinionated format and patterns are not configurable; the semantics of a particular glob are always the same. For example, * never matches across component boundaries. Components are an important part of paths and file system trees, and only the tree wildcard ** (see below) implicitly matches across them.

Wildcards match some amount of arbitrary text in paths and are the most fundamental pattern provided by globs (and likely the most familiar).

The zero-or-more wildcards * and $ match zero or more of any character within a component (never path separators). Zero-or-more wildcards cannot be adjacent to other zero-or-more wildcards. The * wildcard is eager and will match the longest possible text while the $ wildcard is lazy and will match the shortest possible text. When followed by a literal, * stops at the last occurrence of that literal while $ stops at the first occurence.

The exactly-one wildcard ? matches any single character within a component (never path separators). Exactly-one wildcards do not group automatically, so a pattern of contiguous wildcards such as ??? form distinct captures for each ? wildcard. An alternation can be used to group exactly-one wildcards into a single capture, such as {???}.

The tree wildcard ** matches any characters across zero or more components. This is the only pattern that implicitly matches across arbitrary component boundaries; all other patterns do not implicitly match across component boundaries. When a tree wildcard participates in a match and does not terminate the pattern, its captured text includes the trailing separator. If a tree wildcard does not participate in a match, then its captured text is an empty string.

Tree wildcards must be delimited by forward slashes or terminations (the beginning and/or end of an expression). Tree wildcards and path separators are distinct and any adjacent forward slashes that form a tree wildcard are parsed together. Rooting forward slashes in tree wildcards are meaningful and the glob expressions **/*.txt and /**/*.txt differ in that the former is relative (has no root) and the latter has a root.

If a glob expression consists solely of a tree wildcard, then it matches any and all paths and the complete contents of any and all directory trees, including the root.

Character classes match any single character from a group of literals and ranges within a component (never path separators). Classes are delimited by square brackets [...]. Individual character literals are specified as is, such as [ab] to match either a or b. Character ranges are formed from two characters separated by a hyphen, such as [x-z] to match x, y, or z. Character classes match characters exactly and are always case-sensitive, so the expressions [ab] and {a,b} are not necessarily the same.

Any number of character literals and ranges can be used within a single character class. For example, [qa-cX-Z] matches any of q, a, b, c, X, Y, or Z.

Character classes may be negated by including an exclamation mark ! at the beginning of the class pattern. For example, [!a] matches any character except for a. These are the only patterns that support negation.

It is possible to escape meta-characters like *, $, etc., using character classes though globs also support escaping via a backslash \. To match the control characters [, ], and - within a character class, they must be escaped via a backslash, such as [a\-] to match a or -.

Character classes have notable platform-specific behavior, because they match arbitrary characters in native paths but never match path separators. This means that if a character class consists of only path separators on a given platform, then the character class is considered empty and matches nothing. For example, in the expression a[/]b the character class [/] matches nothing on Unix and Windows. Such character classes are not rejected, because the role of arbitrary characters depends on the platform. In practice, this is rarely a concern, but such patterns should be avoided.

Character classes have limited utility on their own, but compose well with repetitions.

Alternations match an arbitrary sequence of one or more comma separated sub-globs delimited by curly braces {...,...}. For example, {a?c,x?z,foo} matches any of the alternative globs a?c, x?z, or foo. Alternations may be arbitrarily nested and composed with repetitions.

Alternations form a single capture group regardless of the contents of their sub-globs. This capture is formed from the complete match of the sub-glob, so if the alternation {a?c,x?z} matches the path abc, then the captured text will be abc (not b). Alternations can be used to group captures using a single sub-glob, such as {*.{go,rs}} to capture an entire file name with a particular extension or {???} to group a sequence of exactly-one wildcards.

Alternations must consider adjacency rules and neighboring patterns. For example, *{a,b*} is allowed but *{a,*b} is not. Additionally, they may not contain a sub-glob consisting of a singular tree wildcard ** and cannot root a glob expression as this could cause the expression to match or walk overlapping trees.

Repetitions match a sub-glob a specified number of times. Repetitions are delimited by angle brackets with a separating colon <...:...> where a sub-glob precedes the colon and an optional bounds specification follows it. For example, <a*/:0,> matches the sub-glob a*/ zero or more times. Though not implicit like tree wildcards, repetitions can match across component boundaries (and can themselves include tree wildcards). Repetitions may be arbitrarily nested and composed with alternations.

Bound specifications are formed from inclusive lower and upper bounds separated by a comma ,, such as :1,4 to match between one and four times. The upper bound is optional and may be omitted. For example, :1, matches one or more times (note the trailing comma ,). A singular bound is convergent, so :3 matches exactly three times (both the lower and upper bounds are three). If no lower or upper bound is specified, then the sub-glob matches one or more times, so <a:> and <a:1,> are equivalent. Similarly, if the colon : is also omitted, then the sub-glob matches zero or more times, so <a> and <a:0,> are equivalent.

Repetitions form a singular capture group regardless of the contents of their sub-glob. The capture is formed from the complete match of the sub-glob. If the repetition <abc/> matches abc/abc/, then the captured text will be abc/abc/.

Repetitions compose well with character classes. Most often, a glob expression like {????} is sufficient, but the more specific expression <[0-9]:4> further constrains the matched characters to digits, for example. Repetitions may also be more terse, such as <?:8>. Furthermore, repetitions can form tree expressions that further constrain components, such as <[!.]*/>[!.]* to match paths that contain no leading dots . in any component.

Repetitions must consider adjacency rules and neighboring patterns. For example, a/<b/**:1,> is allowed but <a/**:1,>/b is not. Additionally, they may not contain a sub-glob consisting of a singular separator /, a singular zero-or-more wildcard * or $, nor a singular tree wildcard **. Repetitions with a lower bound of zero may not root a glob expression, as this could cause the expression to match or walk overlapping trees.

Glob patterns can be combined and matched together using the any combinator. any accepts an IntoIterator of Patterns, such as compiled Programs like Glob or pattern text like str slices. The output is an Any, which implements Program and efficiently matches any of its input patterns.

use wax::{Glob, Program};

let any = wax::any(["**/*.txt", "src/**/*.rs"]).unwrap();
assert!(any.is_match("src/lib.rs"));

Any and the any combinator can be used anywhere a Pattern or Program can.

use wax::walk::FileIterator;
use wax::Glob;

let glob = Glob::new("**/*.{md,rs,toml,txt,yaml,yml}").unwrap();
for entry in glob
    .walk("projects")
    .not(wax::any([
        "**/{.git,.github,target}/**",
        "**/{lib,main,mod}.rs",
    ]))
    .unwrap()
{
    let entry = entry.unwrap();
    // ...
}

Unlike Glob, an Any cannot be matched against a directory tree (as with Glob::walk). However, Any supports features that alternations do not and Globs cannot represent, such as overlapping trees.

Flags toggle the matching behavior of globs. Importantly, flags are a part of a glob expression rather than an API. Behaviors are toggled immediately following flags in the order in which they appear in glob expressions. Flags are delimited by parenthesis with a leading question mark (?...) and may appear anywhere within a glob expression so long as they do not split tree wildcards (e.g., a/*(?i)* is not allowed). Each flag is represented by a single character and can be negated by preceding the corresponding character with a minus -. Flags are toggled in the order in which they appear within (?...).

The only supported flag is the case-insensitivty flag i. By default, glob expressions use the same case sensitivity as the target platforms's file system APIs (case-sensitive on Unix and case-insensitive on Windows), but i can be used to toggle this explicitly as needed. For example, (?-i)photos/**/*.(?i){jpg,jpeg} matches file paths beneath a photos directory with a case-sensitive base and a case-insensitive extension jpg or jpeg.

Wax considers literals, their configured case sensitivity, and the case sensitivity of the target platform's file system APIs when partitioning glob expressions with Glob::partition. Partitioning is unaffected in glob expressions with no flags.

The GlobError type represents error conditions that can occur when building a pattern or walking a directory tree. GlobError and its sub-errors implement the standard Error and Display traits via thiserror.

Wax optionally integrates with the miette crate, which can be used to capture and display diagnostics. This can be useful for reporting errors to users that provide glob expressions. When enabled, error types implement the Diagnostic trait.

Error: wax::glob::adjacent_zero_or_more

  x malformed glob expression: adjacent zero-or-more wildcards `*` or `$`
   ,----
 1 | doc/**/*{.md,.tex,*.txt}
   :        |^^^^^^^^|^^^^^^^
   :        |        | `-- here
   :        |        `-- in this alternation
   :        `-- here
   `----

Wax also provides inspection APIs that allow code to query glob metadata, such as captures and variance.

use wax::Glob;

let glob = Glob::new("videos/**/{*.{mp4,webm}}").unwrap();
assert_eq!(2, glob.captures().count());

Wax provides some optional integrations and features that can be toggled via the Cargo features described below.

Features can be configured in a crate's Cargo.toml manifest.

[dependency.wax]
version = "^0.x.0"
default-features = false
features = [
    "miette",
    "walk"
]

Any components not recognized as separators nor patterns are interpreted as literals. In combination with strict rules, this means some platform-specific path features cannot be used directly in globs. This limitation is by design and additional code may be necessary to bridge this gap for some use cases.

Globs support no notion of a current or parent directory. The path components . and .. are interpreted as literals and only match paths with the corresponding components (even on Unix and Windows). For example, the glob src/../*.rs matches the path src/../lib.rs but does not match the semantically equivalent path lib.rs.

Parent directory components have unclear meaning and far less utility when they follow patterns in a glob. However, such components are intuitive and are often important for escaping a working directory when they precede variant patterns (i.e., as a prefix). For example, the glob ../src/**/*.rs has more obvious intended meaning than the glob src/**/../*.rs. As seen above though, the first glob would only match the literal path component .. and not paths that replace this with a parent directory.

Glob::partition can be used to isolate semantic components that precede patterns and apply semantic path operations to them (namely ..). Glob::partition partitions a glob into an invariant PathBuf prefix and a variant Glob postfix. Here, invariant means that the partition contains no glob patterns that resolve differently than an equivalent native path using the target platform's file system APIs. The prefix can be used as needed in combination with the glob.

use dunce; // Avoids UNC paths on Windows.
use std::path::Path;
use wax::{Glob, Program};

let path: &Path = /* ... */ // Candidate path.

let directory = Path::new("."); // Working directory.
let (prefix, glob) = Glob::new("../../src/**").unwrap().partition();
let prefix = dunce::canonicalize(directory.join(&prefix)).unwrap();
if dunce::canonicalize(path)
    .unwrap()
    .strip_prefix(&prefix)
    .map(|path| glob.is_match(path))
    .unwrap_or(false)
{
    // ...
}

Additionally, Glob::has_semantic_literals can be used to detect literal components in a glob that have special semantics on the target platform. When the miette feature is enabled, such literals are reported as warnings.

use wax::Glob;

let glob = Glob::new("../**/src/**/main.rs").unwrap();
assert!(glob.has_semantic_literals());

While globs can be rooted, they cannot include schemes nor Windows path prefixes. For example, the Windows UNC share path \\server\share\src cannot be represented directly as a glob.

This can be limiting, but the design of Wax explicitly forbids this: Windows prefixes and other volume components are not portable. Instead, when this is needed, an additional native path or working directory must be used, such as the --tree option provided by Nym. In most contexts, globs are applied relative to some such working directory.

Globs are strictly nominal and do not support any non-nominal constraints. It is not possible to directly filter or otherwise select paths or files based on additional metadata (such as a modification timestamp) in a glob expression. However, it is possible for user code to query any such metadata for a matching path or effeciently apply such filtering when matching directory trees using FileIterator::filter_tree.

For such additional features, including metadata filters and transformations using matched text, see Nym.

Globs operate exclusively on UTF-8 encoded text. However, this encoding is not used for paths on all platforms. Wax uses the CandidatePath type to re-encode native paths via lossy conversions that use Unicode replacement codepoints whenever a part of a path cannot be represented as valid UTF-8. In practice, most paths can be losslessly encoded in UTF-8, but this means that Wax cannot match nor capture some literal byte strings.

At the time of writing, Wax is experimental and unstable. It is possible that glob expression syntax and semantics may change between versions in the 0.y.z series without warning nor deprecation.