infinitered / gluegun Goto Github PK

These files are pretty much copy & paste of each other:

• https://github.com/skellock/gluegun/blob/master/packages/gluegun/src/loaders/command-loader.js
• https://github.com/skellock/gluegun/blob/master/packages/gluegun/src/loaders/extension-loader.js

Do the right thing.

For printing the header in the CLI, if we're in brand mode, we will do the first of this list:

• execute the default function in brand/header.js
• render brand/header.txt as straight up text
• render nothing

We only show the gluegun header if we're not in branded mode.

Will need to setup CI to protect this code base from myself.

I've only used semaphore, but I'm open to anything really.

globals

brew install nodejs
npm i -g n
n 7.2
npm i -g yarn

monorepo setup

git clone [email protected]/skellock/gluegun
cd gluegun
npm i

gluegun setup

cd packages/gluegun
yarn
yarn test
yarn run lint
yarn run watch

adding a feature

git branch fun
# furious typing
git commit -m "Adds fun"
# submit a PR with an awesome gif
# win

These two would be great additions to our filesystem extension:

context.filesystem.eol = os.EOL is https://nodejs.org/api/os.html#os_os_eol

context.filesystem.separator = path.sep is https://nodejs.org/api/path.html#path_path_sep

When invoking gluegun, it currently shows all commands. This could quickly spiral. Let's switch to this:

gluegun

# show a list of namespaces

gluegun something

# show a list of commands in the something namespace

Currently we capture errors, tuck them away, and never show them.

https://github.com/infinitered/gluegun/blob/master/packages/gluegun/src/domain/runtime.js#L141

Instead of forcing CLI's to print these with:

const context = await runtime.run()

if (context.error) {
  context.print.error(error.message)
  context.print.debug(context.error)
}

Let's add a new builder function called .rethrowErrorsWhenRunning(true) that people can opt-into when developing.

@GantMan pointed out that commander has a dsl that might work well for us.

We could use that in the front matter for the *.js files by including it:

@command name <thing> [otherThing]

Also, lets promote & clobber onto the context.parameters object the things the user names (e.g. thing and otherThing).

Allow templates to be pseudo-forked.

Provide some kind of directory we can copy the *.njk files to. When we then run a generate() call, if the template has been overridden, we'll call that one instead.

This configuration will be handled in the app's configuration file.

I just donned on me that we might conflicting with other jsdoc things or whatever.

Let's prefix ours with @gg-name, etc

And put a way to execute async and capture output in it.

Inside the app .toml file, let's add:

[defaults]
silent = true

When true (default is false), it will suppress all print commands triggered through the print extension.

There's no tests in the CLI. It kinda just grew and grew until it started doing important things. Now it needs tests.

Let's setup coveralls. I know my tests coverage is slipping, but having an automated score will motivate me.

Let's add another few paths to the mix.

~/.gluegun/plugins/*

These are the gateway to the system. And they've become a bit messy now. Let's clean these up so people wanting to hack at the core aren't confused.

It still says 'staple gun'.

This was unhinged when I switched from inquirer to enquirer.

Let's put a version back that works.

Currently we have

  .load('one_plugin_dir') // loads a directory as plugin
  .loadAll('many_plugins_dir') // loads the immediate sub-directories as plugins

But what we could use this as well:

  .loadAll('node_modules', 'ignite-*') // same, but only ones that match a jetpack pattern

Plugins have unique names. I call these namespaces. But there's nothing magical or special about them. I'm thinking for simplicity sake, change the name from "namespace" to "name".

Namespace seems to imply there's stuff to learn. But there isn't. It's just a name.

Then the description of the command line goes:

<cliname> <pluginname> <commandname>

Right now the identification of the meta tags are looking for // @. That's misleading because of JavaScript's other style of commenting.

Let's support both.

We need a way to load the extensions dynamically from plugins. The sample way we do it with commands, but this time from an extensions directory.

I'd like to be language independent, so let's get out of the js wheelhouse for configuration.

The printWtf script should be opened up so we can see which plugins & commands & extensions aren't loading and why.

Right now when you generate files, the root is the plugin's root. Let's change this to be the templates sub directory.

And put fs-jetpack in it.

@GantMan suggested this.

And put inquirer in it.

Currently a command can't live on it's own. It needs to live under a namespace.

With spork, i made that name space. And since namespaces aren't commands, you can't just type it. So I created default as the command.

So now it's:

gluegun spork default

Which is lame.

I want:

gluegun spork.

So we need to build the concept of a default command. default.js would be fine, but I think index.js is a little nicer because of the convention on the web and React Native.

Prevent these namespaces from being defined in 3rd party plugins:

• spork
• project
• gluegun
• default

If we see those namespaces, we should just go into a badnamespace error state on the plugin.

When a plugin, extension, or command is in a bad state, there's no way of knowing. Let's provide a --debug flag people can run which will emit the errors somehow.

This seems to be broken after a batch of refactoring. The tests didn't seem to catch this.

https://github.com/infinitered/gluegun/blob/master/docs/context-api.md

not inquirer anymore

Let's make this askToOverwrite prompt be more like askYesOrNo or something like that.

We can pass in { message: 'something', yes: 'Yup', no: 'Newp' } or something like that.

@GantMan helped me clarify what I was after. 🌮

@GantMan convinced me to use apisauce over axios because of the response type opinions. I agree i guess.

This will be the extension that lives under context.http.

As violent delights have violent ends, so do default commands have unpruned params
-- Dolores

running ignite add ui gives parameters.first the value of "default"

This looks like some leakage.

👙 Bikini Bonus:
Would be helpful to have parameters.last - which is how I plan on fixing this for futureproof ways. But I'm not a fan of that method.

We would like support for hidden plugins & commands.

This is useful if CLIs want to have their own plugins in the gluegun plugin format, but prevent the end user from having to know of them.

You would still be able to call these from runtime.run({}).

I recommend adding an object as a 2nd parameter to:

.load('dir', { hidden: true })
.loadDefault('dir', { hidden: true })
.loadAll('dir', { hidden: true })

Problem is, loadAll already has an optional 2nd parameter for matching sub-directories, so let's upgrade that to:

.loadAll('dir', { hidden: true, matching: 'ignite-*' })`

This would be a breaking change, but now is the time!

Allow a command line option to pass to the CLI which specifies a default namespace.

gluegun --gluegun-namespace ignite

This namespace then gets priority over all others. In fact, it will be checked even before the other namespace names.

This allows people to brand their own cli like: alias ignite="gluegun --gluegun-namespace ignite" or even in their package.json bin scripts.

I'd like to see a string extension. These would be the things that are automatically placed on the filters list in nunjucks. Let's grow this slowly and start with the cases that we already put into the generate extension.

They're useful to many people, so let's hang them on context.