httpdir-tui

A very simple terminal browser for HTTP directory indexes, with background downloads that can be resumed.

Many mirrors and software archives publish their files as plain "index of" pages (Apache mod_autoindex, nginx autoindex, h5ai, miniserve, ...). httpdir-tui lets you walk through them like a file manager, sort them, and download the files you need, without leaving the terminal or opening a browser. It is built on ratatui and on the httpdirectory crate, which does the actual parsing of the index pages.

Features

• Browse a remote directory index: type, date, name and size of every entry. Any index format understood by httpdirectory works.
• Sort the current listing by name, date or size, ascending or descending, with an arrow in the column header. The order is kept while you navigate.
• Refresh the current listing without leaving it.
• Go back instantly: previously visited directories are kept in memory, so going back never hits the network.
• Download in the background while you keep browsing. A panel shows the most recent transfers with their progress.
• Resume interrupted downloads: after a failure or a quit, pressing the download key again (on the same file) continues where the transfer stopped, as long as the server can prove the file has not changed (see below).
• Stay responsive whatever the speed of the server: the interface never waits for the network.

Requirements

Rust 1.88 or later (required by httpdirectory).

Install

Use cargo install httpdir-tui --locked to install httpdir-tui

Run

httpdir-tui https://cloud.debian.org/images/cloud/

The URL is required. Downloaded files are written to the directory you launched the program from.

Keys

Key	Action
Up / Down	Move the selection by one (wraps around at the
	ends)
PageUp / PageDown	Move the selection by ten
Enter	Open the selected directory (or ..)
Backspace	Go back to the previous directory (no network call)
n / d / s	Sort by name / date / size; each press flips the
	direction (the first one sorts descending)
r	Reload the current directory
g	Download the selected file
q / Esc	Quit

Downloads

• A file is first written to <name>.part and renamed to <name> only once it is complete: a file with its final name is always a whole file.
• An existing file is never overwritten (checked when the transfer starts).
• At most two transfers run at the same time; the others wait their turn, to stay polite with the remote server.
• Names announced by the remote page are untrusted data: a name that is not a plain file name (such as ../../.bashrc) is refused and reported in the panel.
• Large files are not cut off: only the connection and stalled reads time out, never the total duration of a transfer.

Resuming

While a transfer runs, <name>.part.meta records which version of the remote file the data comes from (its ETag, or its Last-Modified). If the transfer fails, or if you quit, both files are kept. The next download of the same file asks the server for the missing part only, and the server continues only if the file is still the same one; otherwise it sends the whole file and the download starts over. A resume therefore never glues the beginning of one version to the end of another.

If the server provides no usable validator, or does not support range requests, the file is simply downloaded again from the start. To force a fresh start, delete the .part file.

Architecture

• src/fetch.rs: all directory listing requests, each in its own tokio::spawn task. The UI never awaits an HTTP request; it reads results from a channel.
• src/download.rs: background downloads, with their own HTTP client (the one inside httpdirectory has a global timeout and is not exposed). Progress is shared through atomics read by the UI; the channel only carries the final outcome.
• src/app.rs: application state: current listing, history stack, selection, sort order, downloads. It is mutated in one place only.
• src/ui.rs: ratatui rendering (URL header, entry table, downloads panel, help bar). It only reads the state.
• src/cli.rs: command line (a single, required URL), parsed by clap.
• src/main.rs: terminal setup, event loop and wiring of the modules above. The screen is redrawn only when something changed (and on every 50 ms tick while a download runs, since its progress is not announced).

Deliberately not done

• No prefetching of the tree: each level is loaded on demand, which limits the number of requests sent to the server.
• Going back (Backspace) never re-requests a directory: it pops a local history stack. Entering a directory a second time does fetch it again.
• No automatic retries: a failed transfer waits for you to press g again, rather than hammering a struggling server.
• No opening of files: Enter only descends into directories.

Known limitations

• Quitting while a download is running abandons it. The .part and .part.meta files stay on disk so that it can be resumed next time.
• Downloads always go to the launch directory; there is no destination setting.