#spinner #read-line #async #async-io #read-input #user-input #terminal

r3bl_terminal_async

Async non-blocking read_line implemenation with multiline editor, with concurrent display output from tasks, and colorful animated spinners

7 releases

0.5.3 May 22, 2024
0.5.2 May 7, 2024
0.5.1 Apr 28, 2024
0.4.0 Apr 21, 2024
0.3.1 Apr 18, 2024

#231 in Command-line interface

Download history 514/week @ 2024-04-16 195/week @ 2024-04-23 27/week @ 2024-04-30 177/week @ 2024-05-07 2/week @ 2024-05-14 143/week @ 2024-05-21 7/week @ 2024-05-28

329 downloads per month

Apache-2.0

11MB
27K SLoC

r3bl_terminal_async

The r3bl_terminal_async library lets your CLI program be asynchronous and interactive without blocking the main thread. Your spawned tasks can use it to concurrently write to the display output, pause and resume it. You can also display of colorful animated spinners ⌛🌈 for long running tasks. With it, you can create beautiful, powerful, and interactive REPLs (read execute print loops) with ease.

Why use this crate

  1. Because read_line() is blocking. And there is no way to terminate an OS thread that is blocking in Rust. To do this you have to exit the process (who's thread is blocked in read_line()).

    • There is no way to get read_line() unblocked once it is blocked.
    • You can use process::exit() or panic!() to kill the entire process. This is not appealing.
    • Even if that task is wrapped in a thread::spawn() or thread::spawn_blocking(), it isn't possible to cancel or abort that thread, without cooperatively asking it to exit. To see what this type of code looks like, take a look at this.
  2. Another annoyance is that when a thread is blocked in read_line(), and you have to display output to stdout concurrently, this poses some challenges.

    • This is because the caret is moved by read_line() and it blocks.
    • When another thread / task writes to stdout concurrently, it assumes that the caret is at row 0 of a new line.
    • This results in output that doesn't look good.

Here is a video of the terminal_async and spinner examples in this crate, in action:

terminal_async_video

Features

  1. Read user input from the terminal line by line, while your program concurrently writes lines to the same terminal. One Readline instance can be used to spawn many async stdout writers ([SharedWriter]) that can write to the terminal concurrently. For most users the TerminalAsync struct is the simplest way to use this crate. You rarely have to access the underlying Readline or SharedWriter directly. But you can if you need to. SharedWriter can be cloned and is thread-safe. However, there is only one instance of Readline per TerminalAsync instance.

  2. Generate a spinner (indeterminate progress indicator). This spinner works concurrently with the rest of your program. When the Spinner is active it automatically pauses output from all the SharedWriter instances that are associated with one Readline instance. Typically a spawned task clones its own SharedWriter to generate its output. This is useful when you want to show a spinner while waiting for a long-running task to complete. Please look at the example to see this in action, by running cargo run --example terminal_async. Then type starttask1, press Enter. Then type spinner, press Enter.

  3. Use tokio tracing with support for concurrent stout writes. If you choose to log to stdout then the concurrent version (SharedWriter) from this crate will be used. This ensures that the concurrent output is supported even for your tracing logs to stdout.

  4. You can also plug in your own terminal, like stdout, or stderr, or any other terminal that implements SendRawTerminal trait for more details.

This crate can detect when your terminal is not in interactive mode. Eg: when you pipe the output of your program to another program. In this case, the readline feature is disabled. Both the TerminalAsync and Spinner support this functionality. So if you run the examples in this crate, and pipe something into them, they won't do anything. Here's an example:

# This will work.
cargo run --examples terminal_async

# This won't do anything. Just exits with no error.
echo "hello" | cargo run --examples terminal_async

To learn more about how this crate itself was built, please checkout the Build with Naz video series on developerlife.com YT channel:

Input Editing Behavior

While entering text, the user can edit and navigate through the current input line with the following key bindings:

  • Works on all platforms supported by crossterm.
  • Full Unicode Support (Including Grapheme Clusters).
  • Multiline Editing.
  • In-memory History.
  • Left, Right: Move cursor left/right.
  • Up, Down: Scroll through input history.
  • Ctrl-W: Erase the input from the cursor to the previous whitespace.
  • Ctrl-U: Erase the input before the cursor.
  • Ctrl-L: Clear the screen.
  • Ctrl-Left / Ctrl-Right: Move to previous/next whitespace.
  • Home: Jump to the start of the line.
    • When the "emacs" feature (on by default) is enabled, Ctrl-A has the same effect.
  • End: Jump to the end of the line.
    • When the "emacs" feature (on by default) is enabled, Ctrl-E has the same effect.
  • Ctrl-C, Ctrl-D: Send an Eof event.
  • Ctrl-C: Send an Interrupt event.
  • Extensible design based on crossterm's event-stream feature.

Examples

cargo run --example terminal_async
cargo run --example spinner

How to use this crate

[TerminalAsync::try_new()], which is the main entry point for most use cases

  1. To read user input, call [TerminalAsync::get_readline_event()].
  2. You can call [TerminalAsync::clone_shared_writer()] to get a SharedWriter instance that you can use to write to stdout concurrently, using std::write! or std::writeln!.
  3. If you use std::writeln! then there's no need to [TerminalAsync::flush()] because the \n will flush the buffer. When there's no \n in the buffer, or you are using std::write! then you might need to call [TerminalAsync::flush()].
  4. You can use the TerminalAsync::println and TerminalAsync::println_prefixed methods to easily write concurrent output to the stdout (SharedWriter).
  5. You can also get access to the underlying Readline via the Readline::readline field. Details on this struct are listed below. For most use cases you won't need to do this.

Readline overview (please see the docs for this struct for details)

  • Structure for reading lines of input from a terminal while lines are output to the terminal concurrently. It uses dependency injection, allowing you to supply resources that can be used to:

    1. Read input from the user, typically crossterm::event::EventStream.
    2. Generate output to the raw terminal, typically std::io::Stdout.
  • Terminal input is retrieved by calling [Readline::readline()], which returns each complete line of input once the user presses Enter.

  • Each Readline instance is associated with one or more SharedWriter instances. Lines written to an associated SharedWriter are output to the raw terminal.

  • Call [Readline::new()] to create a Readline instance and associated SharedWriter.

  • Call [Readline::readline()] (most likely in a loop) to receive a line of input from the terminal. The user entering the line can edit their input using the key bindings listed under "Input Editing" below.

  • After receiving a line from the user, if you wish to add it to the history (so that the user can retrieve it while editing a later line), call [Readline::add_history_entry()].

  • Lines written to the associated SharedWriter while readline() is in progress will be output to the screen above the input line.

  • When done, call [crate::pause_and_resume_support::flush_internal()] to ensure that all lines written to the SharedWriter are output.

[Spinner::try_start()]

This displays an indeterminate spinner while waiting for a long-running task to complete. The intention with displaying this spinner is to give the user an indication that the program is still running and hasn't hung up or become unresponsive. When other tasks produce output concurrently, this spinner's output will not be clobbered. Neither will the spinner output clobber the output from other tasks. It suspends the output from all the SharedWriter instances that are associated with one Readline instance. Both the terminal_async.rs and spinner.rs examples shows this (cargo run --example terminal_async and cargo run --example spinner).

[tracing_setup::init()]

This is a convenience method to setup Tokio tracing_subscriber with stdout as the output destination. This method also ensures that the SharedWriter is used for concurrent writes to stdout. You can also use the TracingConfig struct to customize the behavior of the tracing setup, by choosing whether to display output to stdout, stderr, or a SharedWriter. By default, both display and file logging are enabled. You can also customize the log level, and the file path and prefix for the log file.

Video series on developerlife.com YT channel on building this crate with Naz

Why another async readline crate?

This crate & repo is forked from rustyline-async. However it has mostly been rewritten and re-architected. Here are some changes made to the code:

  • Rearchitect the entire crate from the ground up to operate in a totally different manner than the original. All the underlying mental models are different, and simpler. The main event loop is redone. And a task is used to monitor the line channel for communication between multiple [SharedWriter]s and the [Readline], to properly support pause and resume, and other control functions.
  • Drop support for all async runtimes other than tokio. Rewrite all the code for this.
  • Drop crates like pin-project, thingbuf in favor of tokio. Rewrite all the code for this.
  • Drop simplelog and log dependencies. Add support for tokio-tracing. Rewrite all the code for this, and add tracing_setup.rs.
  • Remove all examples and create new ones to mimic a real world CLI application.
  • Add spinner_impl, readline_impl, and public_api modules.
  • Add tests.

More info on blocking and thread cancellation in Rust

License: Apache-2.0

Dependencies

~26–40MB
~514K SLoC