[andrewstuart2]

> There are a lot of JS tools, but none of them do what we want. We’re trying to coordinate them. We want to provide a good default experience out of the box, so we’re building a CLI to: scaffold, skeleton files, set up build, set up testing environment, possibly even deployment

Was there something wrong with the yeoman & grunt/gulp combo? The yeoman tool is great for the scaffolding and skeleton story, and even for setting up your build, test, deployment environments using whatever combination of grunt & gulp you want to build into your generator.

I'm starting to get weary of this constant need to reinvent the incredible tools that we have already instead of iterating and improving them.

[adambratt]

The problem is that a lot of these js tools that are being built are being built on top of other hobbyist js libraries that are built on top of others that creates this weird chain of dependency that appears to arise from nothing other than the author's coding style preference.

Take a look at just one of gulp's dependencies (and arguably one of the most important) vinyl-fs which it uses for file watching. This is the actual chain of abstractions for the file watching functionality in gulp.

vinyl-fs < glob-watcher < gaze < graceful-fs

    * gulp.watch uses vfs.watch provided by vinyl-fs

    * vfs.watch is actually just glob-watcher (lol)

    * glob-watcher is a simple 20 line wrapper around gaze watch

    * gaze uses graceful-fs to watch full paths

4 levels of library abstraction for file watching. Something that the built-in fs.watch handles natively.

Here's the actual description of the gaze library "A globbing fs.watch wrapper built from the best parts of other fine watch libs."

Yet apparently that wasn't enough as the gulp authors had to toss in 2 more layers on top of it. And at the end of the day gulp still sucks for watching files on linux.

I understand that there are other pieces of vfs that are being used by gulp...but it still doesn't make any sense to me how there can be SOOO many accepted libraries dealing with just file watching. And instead of making a pull request for new features on existing libraries, developers are just building other libraries around them to do what they want.

There's so much fragmentation in the JS world and no one can seem to ever agree to work together on anything. In comparison, in the Python world there's pretty much one agreed upon library for something like file watching "watchdog". There are alternatives but it stands out above all the rest.

[contrahax]

fs.watch actually doesn't work just fine, not sure what gave you that impression. We went with the best file watcher that was around at the time because we didn't feel the need to write our own. The layers of abstraction are there so other people can make use of them as modules. We are (and have been) evaluating other solutions for file watching but gaze is still the best option for us to use under the hood.

[thealphanerd]

I've been stuck in file watching hell before... CHOKIDAR!!!!!

I actually had a really interesting conversation with Bert Belder about some of the lower level problem with file watching. It came up that perhaps there could be a working group around fixing this issue.

[kansface]

The problem with file watching in JS land isn't the libraries, its node. I'll point you at the documentation [1]. fs.watch is inconsistent, unreliable, unperformant on certain platforms, and a layer too low for most applications. For that matter, npm is worlds better than python which is why you don't tend to see crazy dependency graphs in python.

[1] https://nodejs.org/api/fs.html#fs_caveats

[Wintamute]

I think the deep dependency graphs enabled by npm and a community rigorously embracing semver are fantastically exciting. I'm of the opinion that it is one of the platform's biggest strengths.

[mixonic]

Yeoman, Grunt, and Gulp are (I believe) all various implementations of a task runner. They chain tasks in different ways, but at the end of the day think of your asset build as a set of discrete steps.

Broccoli (https://github.com/broccolijs/broccoli), the build tool written by Jo Liss and supported by the Ember community, is truly an asset pipeline tool. It is concerned with transformations to filesystems and file contents, not tasks.

This makes Broccoli a faster and easier to use tool for implementing complex asset pipelines. The main domain object is "trees". A tree represents a directory hierarchy of files that will be regenerated on each build. Broccoli uses directories of symlinks in tmp/ to cache un-changed files from step to step, allowing it to restart a build at any depth and only process those files that change.

Some links you might enjoy on Broccoli:

* http://aexmachina.info/intro-to-broccoli/

* http://moduscreate.com/better-builds-begin-with-broccoli/

* http://hashrocket.com/blog/posts/broccoli-the-build-tool-not... (how Broccoli is used in Ember-CLI)

I can't speak to the weariness, but I can say that Broccoli is completely unlike Grunt, Gulp, or Yeoman and without it (or something like it) Ember-CLI and the next generation of build tools would be impossible. Broccoli has been in development since fall of 2013, and been used aggressively by Ember-CLI users since summer 2014. Give it a look!

[pfooti]

In my experience using grunt, gulp and broccoli as task runner / build systems for complex applications, they all end up doing about the same thing.

In a way, it makes a lot of sense that broccoli is part of the Ember-CLI system - I found broccoli to be the most opinionated of the runners. As long as your structure mapped easily onto the opinions of broccoli, things were easy and speedy. If you want to do something not supported, or customize how things work, not so much. As an example, I submitted a PR to configure where and how broccoli stores its temp directory structure a year ago (./tmp wasn't right for my project), which is still open (and admittedly wasn't the most elegant solution).

Gulp and Grunt, on the other hand, make far fewer assumptions, provide very little in the way of out-of-the-box "just works", and are far more easy to tweak. In my own experience, my issue was getting certain cat/minify/sort processes to work in broccoli (to deal with angular module declaration ordering) was annoyingly complex enough that I just moved to gulp where it was easier (and someone else had already written a gulp plugin).

[mixonic]

> In my experience using grunt, gulp and broccoli as task runner / build systems for complex applications, they all end up doing about the same thing.

This may be true for a codebase of trivial size. With a sizable codebase and many build steps (defeaturefy, babel, es6 modules, es3recast, jshint, jscs) you quickly see grunt and gulp fall flat.

Gulp and Grunt absolutely have a simpler API for blindly chaining tasks. And they have the ability (as task runners) to define multiple tasks and compose them.

Broccoli is tricky to learn. In Ember-CLI we go out of our way to make it invisible to devs who shouldn't need to care what tool is running the internals. For the internals however, it is definitely the correct tool.

[pfooti]

For what it's worth, my gulp build handles both ES6 and Coffee files (although coffee is being phased out it is still present) on both the front (angular) and back (node) ends. It then takes the built files, merges the two streams (as well as any passed-through content from bower etc), and conditionally on development / beta / production sends them through sourcemap, angular-annotate, cat, minify and md5-rev (for cache busting) layers. It also handles injecting correct filenames into index.html and my karma.conf, sorting angular angular files so they're catted in dependency order, livereload, restarting the backend server, catching build errors, and quite a bit more. Plus sass, of course. It does this with a live asset server for development or building to static files for production. As a bonus, I also can use gulp to run various other important tasks like my ersatz database migration (rolled my own with node and some raw SQL fles) process.

I'd say: I ended up building all the features I wanted in broccoli as part of a gulp build system. It did not "fall flat", rather, at the time (about eight months ago), I found broccoli had fallen flat- it didn't handle what I wanted, and even when it did it was impossible to do incremental rebuilds (say) of files - change one thing and the whole system rebuilt itself leading to a livereload on * rather than a single changed file. On top of everything else, it was dog-slow on medium-to-large codebases. I understand that this has changed since, but the frustration I had originally getting it to do what I want coupled with the length of time it took to run rebuilds (nobody wants 10k msec builds when the same process elsewhere takes 300msec) made me transition to gulp. I'm sure broccoli is just fine nowadays, and my read of the emails (I still am watching the repo) seems to indicate that Stefan ironed out a lot of the performance issues around watching and rebuilding (symlinks on OSX, IIRC, were a dog).

As I said the first time around: you can do just about anything with gulp or grunt, you just have to DIY. If you happen to enjoy spending some time really getting to know your toolchain, they all end up doing about the same thing, and with grunt and gulp it's marginally easier to configure certain aspects of the process.

[spiralganglion]

"defeaturefy" is a Googlewhackblatt (only result is this thread). Care to explain what this term means? My curiosity is piqued.

[jannes]

I guess he/she meant https://github.com/thomasboyt/defeatureify but forgot an I in the name.

[woah]

I used to be really into gulp, until I discovered that I could replace my entire gulp file with a 1 line shell script in package.json

[contrahax]

If you can do it in one line of shell then do it - you don't need a build system. I don't understand why everyone sees it as a competition - just use the right tool for the problem you need to solve.

[andrewstuart2]

Grunt and gulp are task runners, yeoman is a scaffolding tool that does exactly that, and all the generators I've used and written will scaffold out Grunt or Gulp tasks which take care of the build/test/deploy story. Usually with live-reload and filesystem watches for autoupdate.

[abritishguy]

Yeoman is for scaffolding

[sehr]

We’re working with the Ember CLI team who are extracting reusable bits. Working with Joe from broccoli and reusing those bits. Current changing the Angular build from gulp to broccoli. Working with the NPM team on package management and resolution. The package managers that exist today aren’t good, but NPM is the closest of all of them.

Sounds like they're doing exactly that!

[dpritchett]

I wish they'd gotten Jo's name right. Luckily her work speaks for itself.