eDirStat
eDirStat is a modern, high-performance, cross-platform disk usage analyzer written in Rust. Inspired by legacy utilities like WinDirStat, it leverages an immediate-mode graphical interface egui to provide a real-time, interactive treemap visualization of your filesystem.
Unlike traditional analyzers that crawl sequentially, eDirStat is engineered from the ground up for modern multi-core systems. It couples a highly optimized, work-stealing multithreaded directory walker with a zero-copy arena data structure. This allows you to scan millions of files, locate space-wasting files using a treemap diagram (among other plots), identify duplicate files, and save or load system snapshots in milliseconds.
Up to 2.8x speedup vs WinDirStat
Up to 2.5x speedup vs WizTree
Up to 9.6x speedup vs QDirStat
📦 Distribution Model & Project Support
eDirStat is, and will always remain, entirely open-source and free. However, if you'd like to support continuous development, engineering, and maintenance costs, official pre-compiled binaries for Windows, macOS, and Linux are distributed exclusively as a paid download via Itch.io - xangelix.itch.io/edirstat.
- 100% Open Source: The complete codebase is free of charge and publicly available under the permissive MIT License. If you prefer to build the tool yourself, you are welcome to compile the source code on Windows, Linux, or macOS at no cost. Instructions are provided below.
- Package Managers: Package managers and their repository maintainers are welcome to bundle and distribute eDirStat through their respective channels at no cost, provided that it is distributed un-paid, "unofficial", and under the MIT License.
Purchasing precompiled packages directly funds the engineering efforts required to keep eDirStat fast, secure, and compatible with the latest operating systems.
📽️ Demo Video
https://github.com/user-attachments/assets/22075865-c258-4cee-bae4-247a48702f25
📸 Screenshots
🚀 Key Features
- ⚡ Work-Stealing Multi-threading: Powered by a lock-free task injector queue that keeps all CPU cores saturated during scanning-- inspired by
ripgrep.
- 🪟 NTFS MFT Scanner (Windows & Linux): Accesses raw NTFS volumes to parse the Master File Table directly, bypassing OS filesystem bottlenecks for near-instantaneous drive indexing (requires administrative/root privileges). Exported
$MFT files can also be parsed offline on any platform.
- 👥 7-Stage Deduplication Engine: Safely identifies byte-for-byte identical files using cryptographically secure BLAKE3 hashing. It is hardlink-aware to protect shared filesystem links and automatically filters out dataless cloud placeholders, special devices, and restricted entries.
- 🌍 Full Localization: Community-translated interface in 20 languages (English, Arabic, Bengali, Chinese (Simplified), Chinese (Traditional), Dutch, French, German, Hindi, Italian, Japanese, Korean, Polish, Portuguese (Brazil), Portuguese (Portugal), Russian, Spanish, Turkish, Ukrainian, and Vietnamese) with automatic system language detection and dedicated Google Noto font subsets.
- 🎨 Theme System: System (auto-detected), Dark, Light, and High Contrast themes.
- ⏹️ Scan Control: Cancel in-progress scans at any time, and restrict scans to a single filesystem (
-x).
- 🌐 Browser Snapshot Viewer: Explore saved
.edst snapshots in the browser via the wasm build on edirstat.com.
- 📦 Fast Compressed Snapshots: Writes structured tree snapshots to disk with Zstd compression in a compact columnar layout (v3) that decodes in milliseconds; legacy v2 snapshots remain readable. Cross-compatible on all little-endian platforms.
- 📊 Dynamic Treemap Visualization: Features a responsive layout canvas with smooth HSL gradient scaling based on file extensions, interactive subtree zoom navigation with directory breadcrumbs, and right-click folder focusing.
- ☁️ Cloud & Special File Badges: Instantly identifies cloud placeholders (
☁ iCloud, OneDrive, Dropbox dataless files), symbolic links (🔗), Unix devices (⚙ FIFOs, sockets, block/char devices), and permission restrictions (🔒) with localized tooltips across the tree, table, and details panels.
- 🗁 Native File Manager Reveal: Highlight and select files natively in macOS Finder, Windows File Explorer, and Linux file managers, or launch terminal sessions directly at any folder.
🔒 Privacy & Local-First Design
eDirStat is engineered as a private, local-first utility:
- 100% Local Filesystem Analysis: Scans, metadata traversal, and analytics are performed entirely in local memory on your CPU. No file names, contents, paths, or disk metrics are ever uploaded to external servers.
- Zero Telemetry: No analytics beacons, tracking cookies, advertising SDKs, or background crash reporters are included.
- App Sandbox Hardening: The official macOS App Store release runs strictly within Apple's App Sandbox without network entitlements.
- For complete legal details, see our Privacy Policy.
🚀 Installation & Build
Prerequisites
Ensure you have the Rust toolchain installed. See rust-toolchain.toml for the version currently required.
Build from Source
# Clone the repository
git clone https://github.com/xangelix/edirstat.git
cd edirstat
# Build the release executable
cargo build --release
The compiled binary will be located at target/release/edirstat.
Note: The desktop app builds on stable Rust. Only the wasm32 web frontend (scripts/build_web.sh) still needs the pinned nightly toolchain, for -Zbuild-std with atomics (wasm threads).
Release packages
The released packages (deb, rpm, pacman, the Windows MSI and zip, the macOS disk image) are built by seaworthy from seaworthy.toml: eDirStat is compiled once per CPU target (crates/edirstat_app, the targets cargo-multivers used before), and the edirstat launcher (launcher/) loads the best build for the machine in its own process, with nothing unpacked or written to disk at run time. .github/workflows/release.yml runs it on every release.
📖 Usage Guide
To run the GUI application from the command line:
./target/release/edirstat
You can also pass a directory path as a positional argument to automatically launch and begin scanning that folder on startup:
./target/release/edirstat /path/to/scan
Command-Line Reference
edirstat [path] # Scan a directory, or load a snapshot file
edirstat /path --to mysnapshot # Headless scan, saved to mysnapshot.edst.zst
edirstat /path --to snap --no-compression # Headless scan, uncompressed .edst output
edirstat /path --benchmark # Headless traversal timing report
edirstat /path -x # Restrict the scan to the same filesystem
Navigating the User Interface
- Scan a Directory:
Click the 📁 Scan Directory button in the top menu bar (or press Ctrl+O / ⌘O) to open the scan dialog, which lists detected drives and mount points for quick selection (or type a path directly). You can also drag and drop any directory directly onto the window. Scans can be cancelled at any time while running.
- Explore the Tree:
The left-hand panel displays a hierarchical directory explorer with informative file badges (
☁ cloud placeholder, 🔗 symlink, ⚙ special file, 🔒 permission denied). You can expand/collapse folders using the [+]/[-] toggles. Use the 🔍 Filter input bar (Ctrl+F / ⌘F) to narrow down the view to matching folders or files with case-sensitive and regular expression matching modes.
- Interact with the Treemap:
The central panel displays a visual representation of your disk space. Larger rectangles correspond to larger files or directories.
- Hovering: Move your cursor over a block to view its full path and size in a tooltip.
- Clicking: Click on a block to automatically select it in the directory tree on the left.
- Zooming & Breadcrumbs: Double-click or right-click any directory block (or choose Focus in Treemap) to zoom the treemap visualization into that subfolder. Use the interactive breadcrumb trail or the ⬆ Up button above the treemap to navigate back up through the folder hierarchy.
- Inspect File Extensions:
The right panel displays a sorted list of file extensions detected during the scan, complete with color-coded markers.
- Deduplicate Your Drive:
Switch to the 👥 Deduplicator tab to search for duplicate files on your scanned filesystem. Custom selection helpers allow you to automatically select duplicates while preserving the oldest, newest, or shortest-path file. Easily replace duplicate files with hardlinks or softlinks to reclaim disk space.
- Context Actions:
Right-click any item in the left-hand explorer or treemap to open a context menu:
- Focus in Treemap: Zooms the treemap visualization into the selected directory.
- Open in File Manager: Launches your operating system's default file browser (Explorer, Finder, or Files) and reveals/selects the highlighted item.
- Open Terminal Here: Launches your preferred terminal emulator navigated directly to the selected directory.
- Copies the name of the selected folder or file to the system clipboard.
Keyboard Shortcuts
| Action | Windows / Linux | macOS | Description |
|---|
| New Scan… | Ctrl+O | ⌘O | Open the scan target selection dialog |
| Rescan | Ctrl+R / F5 | ⌘R / F5 | Refresh and re-scan the current root folder |
| Save Snapshot… | Ctrl+S | ⌘S | Export the active scan to a compressed .edst.zst snapshot |
| Search / Filter | Ctrl+F | ⌘F | Focus the directory tree filter bar |
| Dismiss / Clear | Esc | Esc | Close active modal dialog or clear the search filter |
| Focus in Treemap | ⏎ Enter | ⏎ Enter | Zoom the treemap view into the selected folder |
| Go Up One Level | Alt+↑ | ⌥↑ | Navigate up one directory level in the treemap |
| Reset Treemap Zoom | Esc | Esc | Reset the treemap zoom back to the root directory |
| Toggle Left Panel | F9 | F9 | Show / collapse the directory tree explorer panel |
| Toggle Right Panel | F11 | F11 | Show / collapse the extension stats / details panel |
| Collapse All | Ctrl+Shift+C | ⇧⌘C | Collapse all expanded directory tree nodes |
| Move to Trash | Del | Del | Send selected file or folder to the system Recycle Bin / Trash |
| Delete (Permanent) | Shift+Del | ⇧ Del |
💾 Saving & Loading Snapshots
If you need to analyze a server or remote environment:
edirstat /path/to/my/dir --to mysnapshot- Transfer the
mysnapshot.edst file to another machine.
- Launch
edirstat and click 📖 Load Snapshot to open and navigate the tree with full interactivity, requiring no active filesystem connection.
⚙️ Architectural Design & Internals
1. Parallel Work-Stealing Walker (crates/edirstat/src/engine/traversal.rs)
The traversal engine avoids the performance bottlenecks of standard recursive single-threaded walkers. It utilizes crossbeam-deque for task scheduling inspired by ripgrep:
- Workers & Stealers: Each parallel thread operates on a local thread-safe FIFO task queue. When a thread runs out of directories to scan, it attempts to steal tasks from a global injector or peer worker queues.
- Cycle Detection: Avoids infinite directory loops (caused by recursive symbolic links) by checking filesystem identity descriptors (
dev/ino on Unix, and volume_serial_number/file_index on Windows) against an inherited stack of ancestors.
- Cooperative Cancellation: In-progress scans can be cancelled cleanly; workers poll a shared cancel token while scanning.
- Virtual Filesystem Guard: Scans rooted at
/ automatically skip virtual and mounted system directories (such as /proc, /sys, and /dev).
- Device Boundary Restrictions: Restricts the scan to the primary mount point or device boundary to prevent unintended traversal of system directories (e.g.,
/sys or /proc).
2. Lock-Free Snapshot Commit Loop (crates/edirstat/src/engine/coordinator.rs)
To prevent traversal worker threads from blocking the UI rendering cycle, edirstat decouples directory scanning from interface updates through an event-driven coordinator model:
- The Coordinator: Worker threads stream compressed structural events (
ScanEvent) over a lock-free channel to a dedicated background Coordinator thread.
- Dynamic ID Map: The Coordinator translates worker-local task identifiers to global array indexes in $O(1)$ amortized time.
- Atomic Snapshot Publishing: Instead of locking a mutable tree, the GUI accesses an immutable
FileArenaSnapshot read-only copy via arc_swap. The Coordinator issues updated snapshots to the GUI every 100–1000 milliseconds during an active scan (tiered by tree size).
3. Cache-Friendly Arena Representation (crates/edirstat-core/src/arena.rs)
To conserve memory and avoid pointer-chasing latency, the directory tree is flattened into a single contiguous array (arena):
[ Root Node ] ---> [ Child A ] ---> [ Child B ] ---> [ Child C ]
|
v
[ Sub-child 1 ]
- Index-Based References: Individual
FileNode blocks reference parent, child, and sibling nodes using raw u32 indices rather than heap-allocated pointers (Box or Rc).
- Plain Old Data (POD): The
FileNode struct is annotated with bytemuck::Pod and bytemuck::Zeroable, strictly aligned to 8-byte boundaries.
- Compact String Pool: Directory and file names are deduplicated and stored in a contiguous byte sequence (
StringPool). Nodes reference these names via a lightweight StringId wrapper.
4. Zstd-Compressed Snapshot Persistence (crates/edirstat-core/src/snapshot.rs)
The current (v3) .edst snapshot layout stores the arena as compact columnar data, optionally wrapped in a transparent Zstd container (.edst.zst):
+-------------------------------------------------------------+
| Header (72 Bytes, Little-Endian) |
| - Magic: "EDST" |
| - Version: u16 (v3; v2 legacy files still load) |
| - Uncompressed Size: u64 |
| - Node Count: u64 |
| - String Pool Offset & Length |
+-------------------------------------------------------------+
| Columnar Node Data (DFS pre-order) |
| - Control bytes (flags + timestamp shortcut bits) |
| - Varint columns: name ids, sizes, zigzag timestamp deltas |
| - Varint file/child counts for directories |
| - String Pool (varint-framed packed UTF-8) |
+-------------------------------------------------------------+
- Zstd Compression: Minimizes the disk storage footprint of snapshot files while maintaining high read/write speeds; the container is detected and unwrapped transparently on load.
--no-compression writes a raw .edst without the wrapper.
- Columnar Varint Encoding: Nodes are serialized field-by-field as LEB128 varint columns with zigzag delta-compressed timestamps. The control bytes bit-pack dataless cloud flags (
FLAG_DATALESS) and special file flags (FLAG_SPECIAL) in addition to directory and symlink states, making snapshots significantly smaller and richer to decode than the legacy flat-binary (v2) layout, which remains readable for backward compatibility.
- Cross-Platform Little-Endian: All multi-byte fields are serialized little-endian, keeping snapshots portable across little-endian platforms.
5. Multi-Stage Deduplication Engine (crates/edirstat-gui/src/stats/deduplicator.rs)
https://github.com/user-attachments/assets/a5743098-e88f-4fb7-bc0d-df073ed0615f
The deduplication module detects byte-for-byte identical files with minimal disk I/O. Candidate duplicate groups are identified and isolated through a 7-stage pipeline:
- Size Partitioning: Scanned files are grouped by identical byte counts. Special devices (FIFOs, sockets), dataless cloud placeholders, and permission-restricted files are automatically excluded, and singleton sizes are discarded immediately.
- Prefix Hashing: Worker threads read and hash the first 4KB of files to filter out non-matching formats.
- Midpoint Hashing: Computes a hash around the center of the remaining files to detect differences inside similar files.
- Suffix Hashing: Hashes the last 4KB of file data, which often contains unique trailing metadata.
- Multi-Range Hashing: Performs periodic block sampling (every 100MB) across large files to ensure long-distance uniformity without scanning entire gigabyte-scale structures.
- Full Cryptographic Hashing: Executes a full BLAKE3 cryptographic hash only over candidates that successfully cleared the previous five stages.
- Real-time Validation: Performs timestamp checking and
is_file() validation on disk immediately before grouping and hashing to protect against modifying files changed since snapshot generation.
The engine remains hardlink-aware, allowing it to accurately differentiate between physical duplicate copies and single-inode hardlinks, which consume no additional storage.
Benchmarks
Vs WinDirStat (v2.6.2)
A dense Windows primary drive.
NVMe PCIe Gen3 [ntfs]
Up to 2.8x speedup
video coming soon!
Vs WizTree (v4.31)
A dense Windows primary drive.
NVMe PCIe Gen3 [ntfs]
Up to 2.5x speedup
https://github.com/user-attachments/assets/2190e1d1-51ac-4b5e-89e6-80dfd1d9c8de
Vs QDirStat (v2.0.01)
To evaluate traversal performance, edirstat includes a custom comparison benchmark target comparing it against qdirstat-cache-writer (the official headless command-line crawler shipped with QDirStat for background scanning).
How It Works & Why It's Fair
- End-to-End Subprocess Spawning: Both tools are launched as independent external subprocesses (running the optimized release binary for
edirstat and the perl script execution for QDirStat). This captures full end-to-end CLI execution time, including binary loading, runtime initialization (Perl interpreter boot vs. Rust startup), option parsing, and traversal startup.
- Warm Cache Inode Priming: The benchmark performs 2 warm-up runs for each target directory to prime the OS directory entry page caches. This eliminates disk I/O bottlenecks and isolates CPU/algorithm execution efficiency (multi-threaded, work-stealing Rust vs. single-threaded Perl).
- Statistical Averaging: Measurements are collected across 5 consecutive sample runs to compute a robust median and average traversal duration.
Results
Across a diverse suite of storage devices and directory layouts, eDirStat consistently exceeds the performance of QDirStat in duplicate, delivering up to a 9.6x speedup (ranging from 6.0x to 9.6x) in scan time!
Whether crawling highly nested code repositories on high-speed NVMe drives, game installations on SATA SSDs, or deep directory trees on enterprise HDDs, edirstat's parallel, work-stealing multithreading model, allows it to remain the fastest disk usage analyzer available.
Key Highlights:
- Up to 9.60x faster than the
QDirStat backend writer, especially on SSDs.
- Achieves a 6.04x speedup on mechanical HDDs even when processing massive, deeply nested directory paths.
- Smoothly scales directory traversal workload across all available CPU threads.
System Details
AMD Ryzen 9 9950X3D (32)
Linux 6.18.34-1-cachyos-lts
/home/tux/Documents/git
Samsung 990 Pro NVMe SSD (Gen 4) [btrfs]
Dense, an enormous amount of small files and directories.
Running benches/compare.rs (target/release/deps/compare-95ade4249a41f004)
==================================================
eDirStat vs QDirStat Benchmark
==================================================
Target Directory : /home/tux/Documents/git
CPU Cores Available: 32
==================================================
Performing 2 warm-up runs...
Performing 5 sample runs...
Run 1/5... edirstat: 856.10ms, qdirstat: 6.70s
Run 2/5... edirstat: 869.77ms, qdirstat: 7.11s
Run 3/5... edirstat: 881.21ms, qdirstat: 6.89s
Run 4/5... edirstat: 865.67ms, qdirstat: 6.91s
Run 5/5... edirstat: 851.16ms, qdirstat: 7.31s
================ RESULTS SUMMARY ================
eDirStat (Rust, parallel):
Min : 851.16ms
Max : 881.21ms
Median: 865.67ms
Mean : 864.78ms
QDirStat (Perl writer):
Min : 6.70s
Max : 7.31s
Median: 6.91s
Mean : 6.98s
Speedup (QDirStat / eDirStat): 7.99x
==================================================
/run/media/tux/F1/Games/PC/SteamLibrary/steamapps/common
Samsung SSD 870 QVO 8TB [btrfs]
Game files, a mix of large and small files on a SATA SSD.
Running benches/compare.rs (target/release/deps/compare-95ade4249a41f004)
==================================================
eDirStat vs QDirStat Benchmark
==================================================
Target Directory : /run/media/tux/F1/Games/PC/SteamLibrary/steamapps/common
CPU Cores Available: 32
==================================================
Performing 2 warm-up runs...
Performing 5 sample runs...
Run 1/5... edirstat: 471.00ms, qdirstat: 4.24s
Run 2/5... edirstat: 478.63ms, qdirstat: 5.12s
Run 3/5... edirstat: 467.80ms, qdirstat: 4.54s
Run 4/5... edirstat: 475.84ms, qdirstat: 4.27s
Run 5/5... edirstat: 451.38ms, qdirstat: 8.45s
================ RESULTS SUMMARY ================
eDirStat (Rust, parallel):
Min : 451.38ms
Max : 478.63ms
Median: 471.00ms
Mean : 468.93ms
QDirStat (Perl writer):
Min : 4.24s
Max : 8.45s
Median: 4.54s
Mean : 5.32s
Speedup (QDirStat / eDirStat): 9.64x
==================================================
/run/media/tux/D1
Seagate Exos X18 18TB HDD [btrfs]
Large files, but fewer, on an HDD. Less directory nesting.
Running benches/compare.rs (target/release/deps/compare-95ade4249a41f004)
==================================================
eDirStat vs QDirStat Benchmark
==================================================
Target Directory : /run/media/tux/D1
CPU Cores Available: 32
==================================================
Performing 2 warm-up runs...
Performing 5 sample runs...
Run 1/5... edirstat: 8.01ms, qdirstat: 33.43ms
Run 2/5... edirstat: 5.34ms, qdirstat: 33.03ms
Run 3/5... edirstat: 5.48ms, qdirstat: 32.67ms
Run 4/5... edirstat: 4.59ms, qdirstat: 33.06ms
Run 5/5... edirstat: 5.81ms, qdirstat: 33.49ms
================ RESULTS SUMMARY ================
eDirStat (Rust, parallel):
Min : 4.59ms
Max : 8.01ms
Median: 5.48ms
Mean : 5.85ms
QDirStat (Perl writer):
Min : 32.67ms
Max : 33.49ms
Median: 33.06ms
Mean : 33.14ms
Speedup (QDirStat / eDirStat): 6.03x
==================================================
/run/media/tux/B4
Toshiba MG09SACA16EA 16TB HDD [btrfs]
An enormous amount of tiny files with deep directory nesting, on an HDD.
Running benches/compare.rs (target/release/deps/compare-95ade4249a41f004)
==================================================
eDirStat vs QDirStat Benchmark
==================================================
Target Directory : /run/media/tux/B4
CPU Cores Available: 32
==================================================
Performing 2 warm-up runs...
Performing 5 sample runs...
Run 1/5... edirstat: 515.15ms, qdirstat: 3.54s
Run 2/5... edirstat: 520.41ms, qdirstat: 3.51s
Run 3/5... edirstat: 537.08ms, qdirstat: 3.50s
Run 4/5... edirstat: 580.76ms, qdirstat: 4.24s
Run 5/5... edirstat: 612.70ms, qdirstat: 4.05s
================ RESULTS SUMMARY ================
eDirStat (Rust, parallel):
Min : 515.15ms
Max : 612.70ms
Median: 537.08ms
Mean : 553.22ms
QDirStat (Perl writer):
Min : 3.50s
Max : 4.24s
Median: 3.54s
Mean : 3.77s
Speedup (QDirStat / eDirStat): 6.60x
==================================================
Benchmark Disclaimer & Configuration:
- Comparisons were conducted against WinDirStat v2.6.2, WizTree v4.31, and QDirStat v2.0.01 under controlled testing conditions with primed system caches.
- eDirStat is an independent open-source utility and is not associated with, sponsored by, or endorsed by the trademark holders of those projects.
- Performance measurements depend heavily on hardware setup, filesystem fragmentation, operating system scheduling, and disk caching behavior; individual test results may vary.
💖 Contributors
- @Lej77 — NTFS
$MFT fixes and improvements, including Master File Table scanning of NTFS drives on Linux (#14) and MFT extension record support (#15).
- @AlexanderSchuetz97 — many feature requests, German translation review, and extensive bug testing.
- @hollmmes — Turkish (
tr-TR) localization support (#17).
📝 License
This project is licensed under the MIT License.
Disclaimer: WinDirStat, WizTree, and QDirStat are trademarks of their respective owners. eDirStat is an independent open-source project and is not affiliated with, sponsored by, or endorsed by the trademark holders.
