mirrors
/
semaphore
mirror of https://github.com/NickColley/semaphore.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
semaphore/CONTRIBUTING.md

6.9 KiB
Raw Blame History

Contributing to Pinafore

Installing

To install with dev dependencies, run:

yarn

Dev server

To run a dev server with hot reloading:

yarn run dev

Now it's running at localhost:4002.

Linux users: for file changes to work, you'll probably want to run export CHOKIDAR_USEPOLLING=1 because of this issue.

Linting

Pinafore uses JavaScript Standard Style.

Lint:

yarn run lint

Automatically fix most linting issues:

yarn run lint-fix

Integration tests

Integration tests use TestCafé and a live local Mastodon instance running on localhost:3000.

Running integration tests

The integration tests require running Mastodon itself, meaning the Mastodon development guide is relevant here. In particular, you'll need a recent version of Ruby, Redis, and Postgres running. For a full list of deps, see bin/setup-mastodon-in-travis.sh.

Run integration tests, using headless Chrome by default:

npm test

Run tests for a particular browser:

BROWSER=chrome yarn run test-browser
BROWSER=chrome:headless yarn run test-browser
BROWSER=firefox yarn run test-browser
BROWSER=firefox:headless yarn run test-browser
BROWSER=safari yarn run test-browser
BROWSER=edge yarn run test-browser

If the script isn't able to set up the Postgres database, try running:

sudo su - postgres

Then:

psql -d template1 -c "CREATE USER pinafore WITH PASSWORD 'pinafore' CREATEDB;"

Testing in development mode

In separate terminals:

1. Run a Mastodon dev server:

yarn run run-mastodon

2. Run a Pinafore dev server:

yarn run dev

3. Run a debuggable TestCafé instance:

npx testcafe --hostname localhost --skip-js-errors --debug-mode chrome tests/spec

Test conventions

The tests have a naming convention:

  • 0xx-test-name.js: tests that don't modify the Mastodon database (read-only)
  • 1xx-test-name.js: tests that do modify the Mastodon database (read-write)

In principle the 0- tests don't have to worry about clobbering each other, whereas the 1- ones do.

Mastodon used for testing

There are two parts to the Mastodon data used for testing:

  1. A Postgres dump and a tgz containing the media files, located in fixtures
  2. A script that populates the Mastodon backend with test data (restore-mastodon-data.js).

The reason we don't use a Postgres dump for everything is that Mastodon will ignore changes made after a certain period of time, and we don't want our tests to randomly start breaking one day. Running the script ensures that statuses, favorites, boosts, etc. are all "fresh".

Updating the test data

You probably don't want to do this, as the 0xx tests are pretty rigidly defined against the test data. Write a 1xx test instead and insert what you need on-the-fly.

If you really need to, though, you can either:

  1. Add new test data to mastodon-data.js

or

  1. Comment out await restoreMastodonData() in run-mastodon.js
  2. Make your changes manually to the live Mastodon
  3. Run the steps in the next section to back it up to fixtures/

Updating the Mastodon version

  1. Run rm -fr mastodon to clear out all Mastodon data
  2. Comment out await restoreMastodonData() in run-mastodon.js to avoid actually populating the database with statuses/favorites/etc.
  3. Update the GIT_TAG in run-mastodon.js to whatever you want
  4. Run yarn run run-mastodon
  5. Run yarn run backup-mastodon-data to overwrite the data in fixtures/
  6. Uncomment await restoreMastodonData() in run-mastodon.js
  7. Commit all changed files
  8. Run rm -fr mastodon/ and yarn run run-mastodon to confirm everything's working

Check mastodon.log if you have any issues.

Note that we also run db:migrate just to play it safe, but updating the fixtures/ should make that a no-op.

Unit tests

There are also some unit tests that run in Node using Mocha. You can find them in tests/unit and run them using yarn run test-unit.

Debugging Webpack

The Webpack Bundle Analyzer report.html and stats.json are available publicly via e.g.:

This is also available locally after yarn run build at .sapper/client/report.html.

Codebase overview

Pinafore uses SvelteJS and SapperJS. Most of it is a fairly typical Svelte/Sapper project, but there are some quirks, which are described below. This list of quirks is non-exhaustive.

Prebuild process

The template.html is itself templated. The "template template" has some inline scripts, CSS, and SVGs injected into it during the build process. SCSS is used for global CSS and themed CSS, but inside of the components themselves, it's just vanilla CSS because I couldn't figure out how to get Svelte to run a SCSS preprocessor.

Lots of small files

Highly modular, highly functional, lots of single-function files. Tends to help with tree-shaking and code-splitting, as well as avoiding circular dependencies.

Preact is loaded dynamically

This is a Svelte project, but emoji-mart is used for the emoji picker, and it's written in React. So we lazy-load the React-compatible Preact library when we load emoji-mart.

Some third-party code is bundled

For various reasons, a11y-dialog, autosize, and timeago are forked and bundled into the source code. This was either because something needed to be tweaked or fixed, or I was trimming unused code and didn't see much value in contributing it back, because it was too Pinafore-specific.

Every Sapper page is "duplicated"

To get a nice animation on the nav bar when you switch columns, every page is lazy-loaded as LazyPage.html. This "lazy page" is merely delayed a few frames to let the animation run. Therefore there is a duplication between src/routes and src/routes/_pages. The "lazy page" is in the former, and the actual page is in the latter. One imports the other.

There are multiple stores

Originally I conceived of separating out the virtual list into a separate npm package, so I gave it its own Svelte store (virtualListStore.js). This never happened, but it still has its own store. This is useful anyway, because each store has its state maintained in an LRU cache that allows us to keep the scroll position in the virtual list e.g. when the user hits the back button.

Also, the main store.js store is explicitly loaded by every component that uses it. So there's no store inheritance; every component just declares whatever store it uses. The main store.js is the primary one.

There is a global event bus

It's in eventBus.js. This is useful for some stuff that is hard to do with standard Svelte or DOM events.