Actively maintained fork of unrar / muja/unrar.rs. This fork continues active development with performance improvements, extraction progress callbacks, and updates to the latest UnRAR source releases.

Add to your Cargo.toml:

[dependencies]
unrar-ng = "0.7"

Then use unrar_ng::Archive; in your code.

Breaking change in 0.7

The library targets were renamed: unrar → unrar_ng, unrar_sys → unrar_ng_sys. Code written against 0.6.x using use unrar::Archive; no longer compiles by default. Two migration paths:

1. Recommended (clean) — update both Cargo.toml and source:

[dependencies]
unrar-ng = "0.7"

use unrar_ng::Archive;

2. Minimal-change — keep use unrar::Archive; source by aliasing the dep:

[dependencies]
unrar = { package = "unrar-ng", version = "0.7" }
# Only if you also depend on the FFI crate directly:
unrar_sys = { package = "unrar-ng-sys", version = "0.7" }

Cargo's dep-rename mechanism makes the consumer-side extern crate / use name follow the dep key, regardless of the dependency's [lib] name, so existing use unrar::Archive; / use unrar_sys::*; lines continue to work.

High-level wrapper around the unrar C library provided by rarlab.

This library can only extract and list archives, it cannot create them.

The primary motivation for this fork is batch extraction performance.

The original crate uses the UnRAR DLL's per-file API (RARReadHeaderEx + RARProcessFile), which internally calls SearchBlock(HEAD_FILE) on every iteration. This causes redundant traversal of archive block headers, resulting in severe performance degradation when extracting archives with many small files.

unrar-ng adds a new batch extraction API (extract_all) that uses the same efficient traversal loop as the native unrar CLI, completely eliminating this overhead.

End-to-end extraction benchmark comparing the original unrar crate against unrar-ng. Test file: kernel-v7.0.rar (~94,000 files), created from the Linux kernel v7.0 source tree (downloaded from https://github.com/torvalds/linux/archive/refs/tags/v7.0.zip). The native unrar CLI and unzip are included as reference baselines.

unzip on the same source tree (unzip v7.0.zip) takes 7.17 s for reference.

Key takeaways:

• ~10.3x faster than the original unrar crate (70.69 s → 6.88 s).
• On par with the native unrar CLI (6.88 s vs 7.15 s).
• Comparable to unzip extracting the same content (6.88 s vs 7.17 s).

For a detailed technical analysis, see Batch Extraction Performance Optimization.

use unrar_ng::Archive;

let archive = Archive::new("large_archive.rar")
    .open_for_processing()
    .expect("Failed to open archive");

archive.extract_all("./output")
    .expect("Failed to extract");

Please look inside the examples directory to see how to use this library. Specifically the lister example is well documented and advanced!

Basic example to list archive entries:

use unrar_ng::Archive;

fn main() {
    for entry in Archive::new("archive.rar").open_for_listing().unwrap() {
        println!("{}", entry.unwrap());
    }
}

Run this example: cargo run --example basic_list path/to/archive.rar. You can create an archive by using the rar CLI: rar a archive.rar .

The primary type in this crate is [Archive] which denotes an archive on the file system. Archive itself makes no FS operations, unless one of the open methods are called, returning an [OpenArchive].

The [Archive] struct provides two major classes of methods:

1. methods that do not touch the FS. These are opinionated utility methods that are based on RAR path conventions out in the wild. Most commonly, multipart files usually have extensions such as .part08.rar or .r08.rar. Since extracting must start at the first part, it may be helpful to figure that out using, for instance, archive.as_first_part()
2. methods that open the underlying path in the specified mode (possible modes are [List], [ListSplit] and [Process]). These methods have the word open in them, are fallible operations, return [OpenArchive] inside a Result and are as follows:
   • open_for_listing and open_for_listing_split: list the archive entries (skipping over content/payload)
   • open_for_processing: process archive entries as well as content/payload
   • break_open: read archive even if an error is returned, if possible. The [OpenMode] must be provided explicitly.

An archive is opened in one of these three modes: [List], [ListSplit] or [Process]. This library does not provide random access into archives. Instead, files inside the archive can only be processed as a stream, unidirectionally, front to back, alternating between ReadHeader and ProcessFile operations (as dictated by the underlying C++ library).

That is the idea behind cursors:

Via cursors, the archive keeps track what operation is permitted next:

• [CursorBeforeHeader] -> ReadHeader
• [CursorBeforeFile] -> ProcessFile

The library enforces this by making use of the typestate pattern. An archive, once opened, starts in the CursorBeforeHeader state and, thus, must have its read_header method called, which returns a new OpenArchive instance in the CursorBeforeFile state that only exposes methods that internally map to the ProcessFile operation. Which methods are accessible in each step depends on the archive's current state and the mode it was opened in.

Here is an overview of what methods are exposed for the OpenMode/Cursor combinations:

Archives opened in [List] or [ListSplit] mode also implement [Iterator] whereas archives in [Process] mode do not (though this may change in future releases). That is because the first two will read and return headers while being forced to skip over the payload whereas the latter has more sophisticated processing possibilities that's not easy to convey using an [Iterator].

For more sophisticated examples, please look inside the examples/ folder.

Here's what a function that returns the first content of a file could look like:

fn first_file_content<P: AsRef<Path>>(path: P) -> UnrarResult<Vec<u8>> {
    let archive = Archive::new(&path).open_for_processing()?; // cursor: before header
    let archive = archive.read_header()?.expect("empty archive"); // cursor: before file
    dbg!(&archive.entry().filename);
    let (data, _rest) = archive.read()?; // cursor: before header
    Ok(data)
}
# use std::path::Path;
# use unrar_ng::{Archive, UnrarResult};
#
# let data = first_file_content("data/version.rar").unwrap();
# assert_eq!(std::str::from_utf8(&data), Ok("unrar-0.4.0"));

• Multipart files
• Listing archives
• Extracting them
• Reading them into memory (without extracting)
• Testing them
• Encrypted archives with password
• Linked statically against the unrar source.
• Build unrar C++ code from source
• Basic functionality that operates on filenames / paths (without reading archives)
• Documentation / RustDoc
• Test Suite
• utilizes type system to enforce correct usage
• Well-designed errors (planned)
• TBD

As this library is only a wrapper, these following features are not easily feasible and as such not planned:

• Creating archives
• Random access into arbitrary archive entries
• Pure Rust implementation
• Processing archives from a file descriptor / fs::File handle
• Processing archives from a byte stream

Feel free to contribute! If you detect a bug, open an issue.

Pull requests are also welcome!

If you need help using the library, feel free to create a new discussion or open an issue.

The parts authored by this library's contributors are licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

The embedded C/C++ library uses its own license. For more informations, see its license file.

unrar-ng is a fork of the original unrar crate at muja/unrar.rs. Huge thanks to @muja and the original contributors — unrar-ng builds directly on their work.