watchexec/CONTRIBUTING.md

Contribution guidebook

This is a fairly free-form project, with low contribution traffic.

Maintainers:

• Félix Saparelli (@passcod) (active)
• Matt Green (@mattgreen) (original author, mostly checked out)

There are a few anti goals:

• Calling watchexec is to be a simple exercise that remains intuitive. As a specific point, it should not involve any piping or require xargs.

• Watchexec will not be tied to any particular ecosystem or language. Projects that themselves use watchexec (the library) can be focused on a particular domain (for example Cargo Watch for Rust), but watchexec itself will remain generic, usable for any purpose.

Debugging

To enable verbose logging in tests, run with:

$ env RUST_LOG=watchexec=trace,info RUST_TEST_THREADS=1 RUST_NOCAPTURE=1 cargo test --test testfile -- testname

To use Tokio Console:

1. Add --cfg tokio_unstable to your RUSTFLAGS.
2. Run the CLI with the dev-console feature.

PR etiquette

• Maintainers are busy or may not have the bandwidth, be patient.
• Do not change the version number in the PR.
• Do not change Cargo.toml or other project metadata, unless specifically asked for, or if that's the point of the PR (like adding a crates.io category).

Apart from that, welcome and thank you for your time!

Releasing

A release goes like this:

1. A maintainer launches the "Open a release PR" workflow.

2. A PR bumping the chosen crate's version is opened. Maintainers may then add stuff to it if needed, like changelog entries for library crates. Release notes for CLI releases go directly on the PR.

3. When the PR is merged, the release is tagged. CLI releases also get built and distributed.

4. A maintainer then manually publishes the crate (automated publishing is blocked on crates.io implementing scoped tokens).

Release order

Use this command to see the tree of workspace dependencies:

$ cargo tree -p watchexec-cli | rg -F '(/' --color=never | sed 's/ v[0-9].*//'

Overview

The architecture of watchexec is roughly:

• sources gather events
• events are debounced and filtered
• event(s) make it through the debounce/filters and trigger an "action"
• on_action handler is called, returning an Outcome
• outcome is processed into managing the command that watchexec is running
  • outcome can also be to exit
• when a command is started, the on_pre_spawn and on_post_spawn handlers are called
• commands are also a source of events, so e.g. "command has finished" is handled by on_action

And this is the startup sequence:

• init config sets basic immutable facts about the runtime
• runtime starts:
  • source workers start, and are passed their runtime config
  • action worker starts, and is passed its runtime config
• (unless --postpone is given) a synthetic event is injected to kickstart things

Guides

These are generic guides for implementing specific bits of functionality.

Adding an event source

• add a worker for "sourcing" events. Looking at the signal source worker is probably easiest to get started here.

• because we may not always want to enable this event source, and just to be flexible, add runtime config for the source.

• for convenience, probably add a method on the runtime config which configures the most common usecase.

• because watchexec is reconfigurable, in the worker you'll need to react to config changes. Look at how the fs worker does it for reference.

• you may need to add to the event tag enum.

• if you do, you should add support to the "tagged filterer", but this can be done in follow-up work.

Process a new event in the CLI

• add an option to the args if necessary

• add to the runtime config when the option is present

• process relevant events in the action handler

---

vim: tw=100