Giters

facebook / create-react-app

Set up a modern web app by running one command.

Home Page:https://create-react-app.dev

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

RFC - babel-macros

threepointone opened this issue · comments

commented

Code modifiers are becoming popular in the javascript ecosystem, via babel-plugins (eg - many css-in-js libs, relay, etc). This issue is to open discussion on a generic system to include them in CRA apps, but still honouring the zero-config philosophy.

The proposal here is to have a constrained form of babel-plugins, disguised to appear like regular libraries. A short writeup on what it would look like here - https://github.com/threepointone/babel-macros/blob/master/README.md

  • modifiers/'macros' would be isolated in files with .macros.js suffixes
  • there would no real trace of them in the final bundle stripping imports of all *.macros.js
  • considering they're localised and used on demand, I expect them to have little effect on build performance, so can be enabled in both dev/prod
  • because there's no dependence on webpack, it should still work with other js tooling like jest, etc.

I'm looking for feedback on the same before I start writing code for it, as well as possible first implementations (maybe a simple css-in-js lib?)

If this doesn't belong here, feel free to close the issue. Thanks!

commented

a real world example - a theoretical, simpler port of glam would look like this -

  • npm install glam.macros
// app .js
import css from 'glam.macros'

let red = css` display: flex; color: red; `

React.render(<div className={red}>
  red text!
</div>, window.app)

would compile down to

let red = require('glam.macros/runtime')(() => 
  ['.css-7asd4a { display:flex; color: red; }']
)

React.render(<div className={red}>
  red text!
</div>, window.app)

during runtime, it would render html that looks like

<div class='css-7asd4a'>
  red text!
</div>

and add css to a stylesheet

.css-7asd4a { 
  display:flex; 
  color: red; 
}

additionally, we could also extract the css to a real .css file and automatically import them (leveraging CRA's webpack css loader automatically), and all the other goodies from glam/emotion.

another example would be relay.

instead of -

import Relay from 'relay' 
// ...

Relay.QL`...`

you'd split the two as

import Relay from 'relay'
import QL from 'relay.macros'
// ...

QL`...`

but it would still generate code similar/same to relay's own plugin

commented

Another usecase - integrating @kentcdodds' preval lib https://github.com/kentcdodds/babel-plugin-preval to inline node.js code into client side code for easily inlining file contents, etc without using a custom webpack loader(!)

commented

Cool and super-useful!

A curious question: would this mechanism be enabled on npm module macros only?
I would kinda like to be able to write a macro in the project itself without having to publish it, although I know this potentially breaks the CRA philosophy.

commented

I think the point would be that you just import the macros from anywhere and it would apply itself on the file you're importing. I can definitely see how this could be accomplished using a babel plugin with 0 configuration on the part of the end user. If there's interest in this, I'll build it (because it would be SO FUN!)

commented

So something with a name like babel-plugin-macros that looks for imports/requires of /.*\/macros/, and when it finds them, imports a babel plugin from the macros lib (eg require("glam.macros/plugin"), then applies it to the Program that contained the import/require?

Sounds awesome

commented

Yep, that's what I was thinking @suchipi 😄

commented

Not to the whole program, but to the specific call sites.

commented

The plugin would be responsible for what it does I think. Most of the time it would just find the call sites. Could make a helper to make finding call sites easier, but would want the flexibility of the whole program. I'm busy tonight, but can probably start on this tomorrow...

commented

That constraint is important, hence the proposal. Else it's pretty unrestrained and you'll have all the problems of the babel plugin ecosystem (and why this was called macros in the first place, from lisp)

This is the basic sketch https://github.com/threepointone/babel-macros/blob/master/README.md

commented

Ah, so it would only work for function calls, tagged template strings, and JSX, and could only modify the node where it was "called". That's an important distinction.

Which "problems of the babel ecosystem" are you trying to avoid?

commented

2 big ones are plugin ordering, and implicit changes (you won't be able to tell what changes will happen on a source file just by looking at it)

commented

If you want wholesale babel plugins and/or loaders, the option to eject remains, and is likely a better option.

commented

Oh, now I finally understand your babel-macro thing. Makes total sense. I hope nobody works on this before I get a chance to tomorrow. My mind is racing with ideas 😀

commented

But if course I'm just kidding 😉 if someone else wants to build this, I can't stop you 😀

commented

Sounds a lot like OCaml extension points, which was the result of people saying "letting them change the whole parser makes things too confusing".

Do you think @decorator support would be a good idea, too?

commented

Yeah I think decorators could work too.

commented

Sounds super cool

commented

(We won't support decorators for this before their proposal advances. But in theory yes.)

commented

As for idea itself, heh, I like it. If @kentcdodds (or somebody else) can make a proof of concept and demonstrate it is enough to get Relay and some CSS-in-JS libraries working, I‘m game. The only thing I don’t quite like is the .macros naming, maybe we can come up with an alternative suffix.

commented

Alright, I found 30 minutes and I've got something! Contributions welcome :) I even published it: babel-macros. See the tests.

Ok, now I've gotta run, but this is awesome. Looking forward to some PRs! Even more test cases would be cool (just make your test an object and add skip: true)

commented

The only thing I don’t quite like is the .macros naming, maybe we can come up with an alternative suffix.

The ideal solution would be to have separate syntax, e.g. importMacros, but that's likely never going to happen
Another option is to use a webpack inspired 'babel-macros?css', but that straight up looks alien
Alternate words off the top of my head - mods, plugs, socks(?), defines
(I'm no good at naming stuff haha)

commented

I want to make sure that existing tooling will work as much as I can so I want to avoid anything too weird looking... 🤔

I don't mind macros myself, but happy to consider other names 👍

commented

I want to make sure that existing tooling will work as much as I can so I want to avoid anything too weird looking...

The thing is, it also need to be explicit enough that you wouldn’t mistake it from a regular import. Because otherwise it’ll be a debugging nightmare for the person who doesn’t know what’s going on.

commented

Flow has import type, and sweet.js uses import ... for syntax. Is adding a similar syntax extension to babylon going too far for something like this?

commented

Should we move discussion over to babel-macros?

I'd be fine with it if babylon supported it via a plugin, then people could just include a preset we make (and CRA could include it in its own config theoretically). I like:

import macros glam from 'glam.macros'

But I'm open to pretty much anything else.

FYI, babel-macros now supports:

  1. import
  2. require
  3. asTag
  4. asFunction
  5. asJSX

See the tests and snapshots...

Oh, heck, saved you a click 😄

screen shot 2017-07-07 at 11 10 47 am

commented

Terrific progress, eager to hack on and use this for real stuff

commented

I've moved discussion about how to identify a macros here. Would love to talk about this so we can solidify the API and start building real things with this 😄

commented

@gaearon, do you mind outlining what you'd like to see happen with this before you consider including it in CRA? It's a little bit of a chicken and egg problem (people wont care to build macros until a big tool like CRA enables them to use it, but CRA wants to prove it out in the community first).

In my mind, I think what we need to do is encourage people to write macros for a few popular libraries already doing similar things and see how that goes. Anything else?

commented

I'd want to see the API solidify a little bit so that we don't have to make a bunch of breaking changes to react-scripts as we discover things we want to change.

commented

As it stands, we now have macros for preval, emotion, and my gut feel tells me the relay version would be super easy too.

We should now make a sample app that show these in action with CRA (using react-app-rewired or the like)

Any ideas for more macros? :) off the top of my head - explicit react-hot-loader opt-ins, redux boilerplate generators, rakt-style routes... endless opportunities.

commented

@tricoder42 was working on one for i18n:
kentcdodds/babel-plugin-macros#15

I was playing around with the idea of one that inlines function bodies at callsites, but it's more academically interesting than it'd be practical

commented

Something related to codesplitting maybe?

commented

One thing I haven't considered is that this will break type systems like flow/typescript. I'm not sure of a good workaround yet, open to ideas.

commented

For stuff like CSS-in-JS, the shape of the resulting object is pretty consistent; the macro can be safely thought of as a runtime function call that returns the object it compiles to (even though it won't be). So you can make external type definitions via flow-typed for flow and DefinitelyTyped for typescript.

That approach will probably cover the majority of macros.

commented

As an update on progress:

I could use some help spreading the word and getting more folks using babel-macros so we can get it into CRA (even if it's an undocumented "beta" feature to start).

Thanks friends!

commented

@kentcdodds how about publishing a fork? maybe named react-scripts-babel-macros.

commented

I guess I could do that... 🤔

But I'll name it unmaintained-react-scripts-babel-macros 😉

commented

Ok, so, check this out:

unmaintained-react-scripts-babel-macros diff:

screen shot 2017-09-08 at 10 36 23 am

(this would be implemented as part of babel-preset-react-app if it becomes official)

App.js

screen shot 2017-09-08 at 10 36 58 am

npm test

screen shot 2017-09-08 at 10 38 36 am

npm start

screen shot 2017-09-08 at 10 34 54 am

So that's pretty rock solid awesome... 😄

commented

Really great work @threepointone @kentcdodds 👏
@gaearon is there any way to help with speeding the process of adding macros to CRA ?

commented

babel-macros compatible plugins for some popular css-in-js libraries that already have their own babel plugin would probably drive interest (glamor, etc)

commented

That, and also working Relay and i18n solutions.

commented

@gaearon Could you expand on this thought some more, curious what all you have in mind in regards to expected functionality for macros for these.

That, and also working Relay and i18n solutions.

commented

We're pretty close to merging a babel-macro-mode for styled-components. (styled-components/styled-components#1256; import styled from 'styled-components/macro')

Would be awesome to be able to use that in create-react-app ootb to get nicer class names and proper minification just like that!

commented

Emotion already supports macros.

@gaearon Is there any clear outline of when babel-macros support could be added to CRA?

commented

The only thing that's worrying me is that we plan to jump to Babel 7 soon. (#3522)

Would that break your macros? Can you check?

commented

FWIW, I made changes to the project so it works better with Babel 7:

  1. It's been renamed to babel-plugin-macros so you don't have to do weird things in the .babelrc.js config to make it work.
  2. It's now tested with Babel 7

Unless a particular macro is doing something weird with the AST, I'm pretty sure they should still work. I believe that a macro would suffer from the same breaking changes that a normal plugin will suffer from (changes in path methods, etc. of which I believe there are very few if any).

commented

I'll go ahead and make a pull request to add babel-plugin-macros. We can leave it undocumented for now and I'll add a comment that it's experimental.

commented

Would be awesome if this will fit in the next release ... #3672

commented

Fixed by #3675.

You can start testing it with the first 2.x alphas. Instructions are here: #3815

commented

HUGE Thank you to @threepointone for the idea 👏👏👏👏

commented

Out of curiosity, is it possible to use macro to modify webpack config file? Say if I want to add a new loader into the config without ejecting? Thanks.

cc @gaearon @kentcdodds

commented

@coodoo Nope, it's not possible. Macros only transform AST nodes. There're like Babel plugins.

commented

I don’t think CRA runs their webpack config through their own babel pipeline but hahaha nice try 😆

commented

HUGEST Thank you to @kentcdodds for actually building the thing (based on a hasty gist lol) and being such a good steward! You’re the best.

commented

Oh snap, so is there anyway other than ejecting that could allow one to add new loaders to the webpack config? 😂

commented

you could write a macro that reads the file, parses/transforms it, and inlines it at the macro site call. eg -

import csvToJs from 'csvToJs.macro'

let data = csvToJs('./myfile.csv') // will inline the contents of myfile.csv as a js object here

a disadvantage of this approach is that if you require it in a different file, it'll get inlined again, increasing the size of your bundle for no good reason.

another approach is to write the contents of the file to another js file (ie - in the macro, you'd call fs.writeFileSync to a 'myfile.csv.js' file, and convert -

let data = csvToJs('./myfile.csv')

to

let data = require('./myfile.csv.js')

but this would mean you'd generate spurious files in your source folder (or whichever temp folder you output to).

either option may work for you, depending on your constraints.

commented

but this would mean you'd generate spurious files in your source folder (or whichever temp folder you output to).

I suggest writing the file to node_modules/.cache/some-namespace/path/to/file.js then you don't need to worry about generating spurious files :)

I think that there is room for someone to make a module that makes doing this with macros very easy:

const {createMacro} = require('babel-plugin-macros')
const createTempFileForMacro = require('the-module-someone-writes-to-make-this-easy')

module.exports = createMacro(csvToJsMacro)

function csvToJsMacro({references}) {
  // do stuff
  createTempFileForMacro(someRelevantArguments)
  // do more stuff
}

I think many macros would benefit from something like this (though I don't believe it's something that belongs in the core of babel-plugin-macros).

commented

One problem with Babel macro reading a file is that unless the file which uses the macro changes, it won't be called again due to the Babel loader cache, which can be very annoying.

commented

@satya164 this problem does not exist in Create React App. There is continuous cache invalidation for files using macros.
Clearing the cache manually should be avoided.