[FEATURE ember-module-unification] Initial implementation #15368

iezer · 2017-06-16T23:11:28Z

To support module unification, we must modify the container and registry to pass the source as second parameter to the resolver. This PR introduces a new feature flag called ember-module-unification which does just that.

I'll follow up with some PRs in the ember-resolver and glimmer-resolver repos, and an end-to-end example of an Ember app using module unification and private components.

When the feature flag is on, this change breaks two tests related to expandLocalLookup. I'm looking at a way to support the new functionality and make those tests pass. While I'm doing that, I'd appreciate some feedback on the overall design here.

EDIT

Related PRs:

@mixonic @rwjblue

mixonic

Left comments for a few minor items. The testing and feature flag stuff can be improved but the functionality here looks good at a glance. Please do cross-link to the PRs and issues elsewhere.

mixonic · 2017-06-21T16:38:02Z

packages/container/lib/container.js

@@ -1,5 +1,5 @@
 /* globals Proxy */
-import { assert } from 'ember-debug';
+import { assert, isFeatureEnabled } from 'ember-debug';


@rwjblue am I correct in suggesting this should be using ember/features now? Like so:

import { EMBER_MODULE_UNIFICATION } from 'ember/features';

source

Hm I don't think we have any docs on this in CONTRIBUTING.md.

mixonic · 2017-06-21T16:56:22Z

packages/container/lib/container.js

+        // with ember-module-unification, if expandLocalLookup returns something,
+        // pass it to the resolve without the source
+        fullName = expandedFullName;
+        delete options.source;


I wonder if here and at the other delete call further down we should not be mutating the passed-in object? Instead allocate a new one.

mixonic · 2017-06-21T16:59:00Z

packages/container/lib/registry.js

+      return name;
+    }
+
+    return options && options.source ? `${options.source}:${name}` : name;


Can you wrap the test here in parens just for legibility?

return (options && options.source) ? `${options.source}:${name}` : name;

mixonic · 2017-06-21T18:17:27Z

packages/container/tests/container_test.js

+
+    let container = registry.container();
+
+    strictEqual(container.factoryFor('component:my-input', { source: src }).class, PrivateComponent, 'The correct factory was provided');


Tests are passed assert as an argument and should use that over the global assertion methods. For example:

QUnit.test('The container can pass a source to factoryFor', function(assert) { /* ... */ assert.strictEqual( container.factoryFor('component:my-input', {source: src}).class, PrivateComponent, 'The correct factory was provided' ); /* ... */

In fact re-using the string 'component:my-input' instead of re-typing it would clean things up as well, as would using source instead of src as the source string variable name (then this can be {source}).

mixonic · 2017-06-21T18:19:30Z

packages/container/tests/container_test.js

+    let container = registry.container();
+
+    strictEqual(container.factoryFor('component:my-input', { source: src }).class, PrivateComponent, 'The correct factory was provided');
+    strictEqual(container.factoryManagerCache[`template:routes/application:component:my-input`].class,


Instead of asserting the contents of the private cache, can the cache be make _ private and here you can assert resolve or the method on the resolver isn't called a second time? or that it is called with a different source? Test the behavior, not the state of the cache?

mixonic · 2017-06-25T04:25:56Z

packages/container/lib/container.js

+        // pass it to the resolve without the source
+        normalizedName = expandedFullName;
+        options = copy(options, true);
+        delete options.source;


Copying all properties then deleting one is not ideal. It likely causes the object to go into hash table mode which is a slower data structure.

It would be ideal to create an object with only the data we want to pass forward. Kind of silly that we are doing all this work since I think there are no other properties on options :-p

But please just make an object. Even options = {} would be better if there are no other options to copy.

This is scenario that will never get hit in practice because it would only be executed if expandedLocalLookup is implemented and returns something. Are you suggesting that I do options = {};? I'm concerned that some upstream caller might add an addition property to options and expect it to be passed to the resolver.

@iezer Right, options as an argument to lookup is only a contract with the lookup method. We don't document the argument as resolverOptions where it is implied the object is passed transparently.

A patch that build the object up by iterating Object.keys but skipping source would be ok, but really IMO just building the new object (in this case passing an empty object) is fine. Building a whole new object by iterating would be slow, calling the delete after copying is slow, and this is fairly performance sensitive code. Doing the fast and API-conservative thing is ok.

Ok I'll make that change

mixonic

Talked with @dgeb, and I think this psuedo code would be correct for us to ensure we get the higher precedence for local/private lookups over global when resolving a pair of things:

`
a:/my-app/widgets/x-mark
b:/my-app/widgets/top-level/x-mark
`

let source = 'b:/my-app/widgets/top-level';
let a = di.lookup('a:x-mark', {source})

let pairedB;
if (a) {
  pairedB = di.lookup('b', {source: a})
}

let b;
if (pairedB) {
  b = pairedB;
} else {
  b = di.lookup('b:x-mark', {source})
  if (b) {
    let pairedA = di.lookup('a', {source: b})
    a = pairedA;
  }
}

let paired = [a,b]
return paired;

iezer · 2017-06-30T19:25:11Z

@mixonic I'm not sure I follow that example. When you call di.lookup you still have no way of knowing if the lookup found the factory with the source or without.

* Pass `source` to the resolver's `resolve` method as a second argument * Update the logic for component pairs to ensure local/private resolution of a component/template are only matched with another local/private resolution.

mixonic · 2017-07-01T17:40:07Z

I've added the commit prefix and an entry to FEATURES.md along with some other comments in the code. I think this is gtg as a feature flag.

dgeb

Thanks for all your hard work @iezer. I think you've walled everything off appropriately behind the feature flag, which should allow us to iterate on the implementation.

mixonic requested changes Jun 21, 2017

View reviewed changes

iezer force-pushed the isaac/referrer-to-resolver branch from 7288544 to 1f40d58 Compare June 23, 2017 16:24

This was referenced Jun 23, 2017

local lookup changes glimmerjs/glimmer-resolver#16

Closed

Pass Referrer to Glimmer Resolver ember-cli/ember-resolver#196

Closed

mixonic requested changes Jun 25, 2017

View reviewed changes

mixonic requested changes Jun 26, 2017

View reviewed changes

iezer force-pushed the isaac/referrer-to-resolver branch from 1f40d58 to bd5ad81 Compare June 30, 2017 19:16

iezer force-pushed the isaac/referrer-to-resolver branch 4 times, most recently from e7cdb3f to 14b5e36 Compare July 1, 2017 16:19

[FEATURE ember-module-unification] Initial work

5e0a1fd

* Pass `source` to the resolver's `resolve` method as a second argument * Update the logic for component pairs to ensure local/private resolution of a component/template are only matched with another local/private resolution.

mixonic force-pushed the isaac/referrer-to-resolver branch from 14b5e36 to 5e0a1fd Compare July 1, 2017 17:35

mixonic approved these changes Jul 1, 2017

View reviewed changes

dgeb approved these changes Jul 1, 2017

View reviewed changes

mixonic changed the title ~~Module Unification: Pass source to resolver~~ [FEATURE ember-module-unification] Initial implementation Jul 1, 2017

mixonic merged commit d807315 into emberjs:master Jul 1, 2017

This was referenced Jul 2, 2017

Local lookup ember-cli/ember-resolver#191

Closed

module unification feature flag #15350

Closed

iezer mentioned this pull request Dec 12, 2017

[FEATURE ember-module-unification] Support name-spaced lookups #15967

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[FEATURE ember-module-unification] Initial implementation #15368

[FEATURE ember-module-unification] Initial implementation #15368

iezer commented Jun 16, 2017 •

edited

Loading

mixonic left a comment

mixonic Jun 21, 2017

mixonic Jun 21, 2017

mixonic Jun 21, 2017

mixonic Jun 21, 2017

mixonic Jun 21, 2017

mixonic Jun 25, 2017

iezer Jun 30, 2017

mixonic Jun 30, 2017

iezer Jun 30, 2017

mixonic left a comment •

edited

Loading

iezer commented Jun 30, 2017

mixonic commented Jul 1, 2017

dgeb left a comment


		let container = registry.container();

		strictEqual(container.factoryFor('component:my-input', { source: src }).class, PrivateComponent, 'The correct factory was provided');

[FEATURE ember-module-unification] Initial implementation #15368

[FEATURE ember-module-unification] Initial implementation #15368

Conversation

iezer commented Jun 16, 2017 • edited Loading

mixonic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mixonic left a comment • edited Loading

Choose a reason for hiding this comment

iezer commented Jun 30, 2017

mixonic commented Jul 1, 2017

dgeb left a comment

Choose a reason for hiding this comment

iezer commented Jun 16, 2017 •

edited

Loading

mixonic left a comment •

edited

Loading