fix: Added missing chrome privacy api settings by Manvel · Pull Request #205 · mozilla/webextension-polyfill

Manvel · 2019-10-14T15:46:24Z

This PR is for fixing #204

Rob--W · 2019-10-24T13:26:29Z

-      },
-    };
+    // Wrap all "network", "services", "websites" APIs in chrome.privacy.*
+    if (extensionAPIs.privacy) {


This dereferences the privacy namespace even when not used.

Could you make this a lazy getter? Maybe a getter for each of network, services and websites.

Good point, will update.

Updated, I have created separated getters for each of the privacyTypes as you have asked, but I feel like that implementation below is simpler:

// Wrap all "network", "services", "websites" APIs in chrome.privacy.* const privacy = { get supported() { return extensionAPIs.privacy; }, }; if (privacy.supported) { apiMetadata.privacy = {}; for (const privacyType of ["network", "services", "websites"]) { apiMetadata.privacy[privacyType] = {}; for (const privacyName of Object.keys(extensionAPIs.privacy[privacyType])) { apiMetadata.privacy[privacyType][privacyName] = settingMetadata; } } }

What you think?

Rob--W · 2019-10-29T16:26:58Z

+      apiMetadata.privacy = {};
+      for (const privacyType of ["network", "services", "websites"]) {
+        apiMetadata.privacy[privacyType] = {};
+        for (const privacyName in extensionAPIs.privacy[privacyType]) {


This would also include enumerable properties from Object.prototype, which shouldn't be exported.

Good point, will update.
Guess using Object.keys() might be better.

Manvel · 2019-10-29T19:25:30Z

@Rob--W I have addressed the comments, but seems like there is some problems with ChromeDriver SessionNotCreatedError: session not created: This version of ChromeDriver only supports Chrome version 76.

Most probably I won't be available tomorrow, but will ensure to address anything needed the day after.

Rob--W

@Rob--W I have addressed the comments, but seems like there is some problems with ChromeDriver SessionNotCreatedError: session not created: This version of ChromeDriver only supports Chrome version 76.

Most probably I won't be available tomorrow, but will ensure to address anything needed the day after.

We should probably bump the chromedriver version again... As done in #199. If you open a pull request in the same format, I'd approve and merge it.

Rob--W · 2019-10-29T19:31:36Z

+        return {};
+      }});
+    }
+    if (privacy.supported) {


This is not a lazy getter.

As soon as you dereference the extensionAPIs.privacy object, the browser has to initialize the extension bindings for chrome.privacy object.

A lazy getter is one where you avoid dereferencing the extensionAPIs.privacy object until absolutely necessary.

For example, something like this:

apiMetadata.privacy = { get network() { // <-- This is a getter. You should probably use Object.defineProperty instead. let target = extensionAPIs.privacy && extensionAPIs.privacy.network; if (!target) { let network = {}; for (let k of Object.keys(target)) { network[k] = settingMetadata; } this.network = network; // <-- This caches the result, hence the "lazy getter" is run only once. return network; } }, };

Now I feel so stupid, thanks will update.

@Rob--W Updated, hope now it looks better.

Manvel · 2019-10-29T21:10:39Z

@Rob--W I have addressed the comments, but seems like there is some problems with ChromeDriver SessionNotCreatedError: session not created: This version of ChromeDriver only supports Chrome version 76.
Most probably I won't be available tomorrow, but will ensure to address anything needed the day after.

We should probably bump the chromedriver version again... As done in #199. If you open a pull request in the same format, I'd approve and merge it.

Done -> #208

Rob--W · 2019-10-29T23:47:40Z

+
+    apiMetadata.privacy = {};
+    for (const type of ["network", "services", "websites"]) {
+      Object.defineProperty(apiMetadata.privacy, type, {


Add configurable: true, to the properties of the parameter to Object.defineProperty, to ensure that the property can be replaced later.

See https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Object/defineProperty

Rob--W · 2019-10-29T23:49:23Z

+            return privacyType;
+          }
+        },
+        set: (value) => value,


This is only used internally; no setter necessary.

When I don't use setters, getters only error is thrown when caching the value as the strict mode is used.

Not sure why this is not being caught by tests.

Steps to reproduce:

npm run build.

Load dist/browser-polyfill.js into a chrome extension.

execute browser.privacy.services.passwordSavingEnabled.get({});

Observer behavior:

Uncaught TypeError: Cannot set property services of #<Object> which has only a getter

Expected:
No error is thrown.

This is because you're triggering the setter (at line 505).

There are two ways to replace a getter property:

delete the original property and assign the value.

Use Object.defineProperty` to replace the property descriptor with one that only has a value.

In a unit test, you should count the number of accesses to chrome.privacy.network when you dereference the chrome.privacy.network object.

After looking at the implementation, it is apparent that the metadata is dereferenced only once and cached afterwards. I guess that you don't even have to replace the getter, so the code can be simplified.

implementation where metadata is dereferenced once:

webextension-polyfill/src/browser-polyfill.js

Lines 256 to 261 in 5f7a845

(hasOwnProperty(wrappers, prop) ||

hasOwnProperty(metadata, prop))) {

// This is an object that we need to do some wrapping for the children

// of. Create a sub-object wrapper for it with the appropriate child

// metadata.

value = wrapObject(value, wrappers[prop], metadata[prop]);

In a unit test, you should count the number of accesses to chrome.privacy.network when you dereference the chrome.privacy.network object.

I don't understand this comment, you want me to update the tests in a way?

After looking at the implementation, it is apparent that the metadata is dereferenced only once and cached afterwards. I guess that you don't even have to replace the getter, so the code can be simplified

Sounds good, then I guess I can just remove the line where I overwrite the getter and that's it?

After looking at the implementation, it is apparent that the metadata is dereferenced only once and cached afterwards. I guess that you don't even have to replace the getter, so the code can be simplified

Done, removed the cache. Hope I understood your comment correctly.

In a unit test, you should count the number of accesses to chrome.privacy.network when you dereference the chrome.privacy.network object.

I need a hint here, not sure how I can implement that.

Here is an existing test that verifies that a property is accessed only once:

webextension-polyfill/test/test-proxied-properties.js

Lines 140 to 183 in ad207a8

describe("without side effects", () => {

it("should proxy non-wrapped methods", () => {

let lazyInitCount = 0;

const fakeChrome = {

get runtime() {

// Chrome lazily initializes API objects by replacing the getter with

// the value. The initialization is only allowed to occur once,

// after that `undefined` is returned and a warning is printed.

// https://chromium.googlesource.com/chromium/src/+/4d6b3a067994ce6dcf0ed9a9efd566c083736952/extensions/renderer/module_system.cc#414

//

// The polyfill should invoke the getter only once (on the global chrome object).

++lazyInitCount;

const onMessage = {

addListener(listener) {

equal(this, onMessage, "onMessage.addListener should be called on the original chrome.onMessage object");

},

};

const value = {onMessage};

Object.defineProperty(this, "runtime", {value});

return value;

},

get tabs() {

ok(false, "chrome.tabs should not lazily be initialized without explicit API call");

},

};

return setupTestDOMWindow(fakeChrome).then(window => {

// This used to be equal(lazyInitCount, 0, ...), but was changed to

// accomodate a change in the implementation of the polyfill.

// To verify that APIs are not unnecessarily initialized, the fakeChrome

// object has a "tabs" getter that fails the test upon access.

equal(lazyInitCount, 1, "chrome.runtime should be initialized because chrome.runtime.id is accessed during polyfill initialization");

window.browser.runtime.onMessage.addListener(() => {});

equal(lazyInitCount, 1, "chrome.runtime should be initialized upon accessing browser.runtime");

window.browser.runtime.onMessage.addListener(() => {});

equal(lazyInitCount, 1, "chrome.runtime should be re-used upon accessing browser.runtime");

window.chrome.runtime.onMessage.addListener(() => {});

equal(lazyInitCount, 1, "chrome.runtime should be re-used upon accessing chrome.runtime");

});

});

});

You would have to add a new test that verifies that fakeChrome.privacy.network is accessed only once, even if the test gets window.browser.privacy.network.somePropertyName twice.
It would probably also be good to verify that if fakeChrome.privacy.services is undefined, that window.privacy.network.services is undefined as well.

@Rob--W I have added a test, looking into existing implementation you provided, but I'm bit unsure about it.

It would probably also be good to verify that if fakeChrome.privacy.services is undefined, that window.privacy.network.services is undefined as well

Not sure if I understood this one, are you referring to something like?

equal(window.browser.privacy.services, undefined);

… has only a getter

Rob--W · 2019-10-30T19:08:48Z

        window.chrome.runtime.onMessage.addListener(() => {});
        equal(lazyInitCount, 1, "chrome.runtime should be re-used upon accessing chrome.runtime");
+
+        window.browser.privacy.network.networkPredictionEnabled();


Don't call the function, just check that the returned object looks like a Setting object (object with set, get and clear methods).

Rob--W · 2019-10-30T19:09:56Z

+
+            const networkPredictionEnabled = () => true;
+            const value = {networkPredictionEnabled};
+            Object.defineProperty(fakeChrome.privacy, "network", {value});


Don't replace the property. Now we're interested in how often privacy.network is dereferenced.

Noted, done.

It fails, but I can confirm it's not a regression as it was calling twice for the old implementation as well.

Any idea why it's so? Might be a problem with my test?

It fails, but I can confirm it's not a regression as it was calling twice for the old implementation as well.

This was wrong assumption, apparently I was running test without running npm run build first, Sorry for this, will have a look into it later today, the old tests were passing correctly, the problem is in the new implementation.

After taking a closer look into implementation, now I see what is the problem.

The problem is that I can't fix it without touching wrapAPIs() code, unless I hardcode everything.

I'm thinking about something like:

const settingMetadata = { clear: {minArgs: 1, maxArgs: 1}, get: {minArgs: 1, maxArgs: 1}, set: {minArgs: 1, maxArgs: 1}, }; apiMetadata.privacy = {"*": {"*": settingMetadata}};

and updating wrapObject():

if (hasOwnProperty(metadata, "*")) { value = wrapObject(value, wrappers[prop], metadata["*"]); }

I'm quite uncomfortable updating wrapObject(), but I'll give it a shot and revert it if you say so.

@Rob--W Done.

Rob--W · 2019-10-30T19:10:25Z

  describe("without side effects", () => {
    it("should proxy non-wrapped methods", () => {
      let lazyInitCount = 0;
+      let lazyInitPrivacyCount = 0;


Could you create a new describe(...) block to test the privacy API?

Rob--W · 2019-11-02T11:50:38Z

-        referrersEnabled: settingMetadata,
-      },
-    };
+    apiMetadata.privacy = {"*": {"*": settingMetadata}};


I'm not immediately against this implementation, since it is still somewhat readable and concise.
I do wonder whether we should explicitly spell out network, services, and websites.

@rpl Thoughts?

@Rob--W Any thought on this? Happy to explicitly spell out network, services, and websites, if that holds the progress.

CC: @rpl

👍 from me on spelling out the first level and use the "*" to cover all the properties inside those.

Thanks @rpl and @Rob--W, done.

I was thinking about 2 different implementation:

const privacyType = {"*": settingMetadata}; apiMetadata.privacy = { network: privacyType, services: privacyType, websites: privacyType, };

apiMetadata.privacy = {}; for (const type of ["network", "services", "websites"]) { apiMetadata.privacy[type] = {"*": settingMetadata}; }

If you have preference for one or another please let me know which to use. This seem to be most consistent comparing to settingMetadata.

Keep the current code. It is more readable than the proposed alternatives.

Rob--W · 2019-11-07T14:08:09Z

+        },
+      };
+
+      return setupTestDOMWindow(fakeChrome).then(window => {


equal(lazyInitCount, 0, ...); to check that chrome.privacy is not accessed at first.

Manvel · 2019-11-14T12:52:42Z

@Rob--W Is there anything blocking this issue to be merged into master?

Thanks in advance.

Rob--W · 2019-11-21T14:21:28Z

Thanks for the patch @Manvel !

Manvel changed the title ~~#204 - Added missing chrome privacy api settings~~ fix: Added missing chrome privacy api settings Oct 14, 2019

Manvel force-pushed the chrome-privacy-api-update branch 2 times, most recently from 66869a3 to b082082 Compare October 14, 2019 19:14

Rob--W reviewed Oct 29, 2019

View reviewed changes

Manvel added 8 commits October 30, 2019 12:43

fix: Added missing chrome privacy api settings

6a65805

Wrap all network, services, websites APIs in chrome.privacy.*

1d72428

Ensure that privacy API is available

585cd2d

Using lazy getters

8e32e6d

Include only own properties

43962ef

Use lazy getter

cf49ab7

Added setter to avoide Cannot set property network of #<Object> which…

e2c1116

… has only a getter

Added configurable: true

8277c03

Manvel force-pushed the chrome-privacy-api-update branch from e735f85 to 8277c03 Compare October 30, 2019 10:50

Manvel added 2 commits October 30, 2019 21:18

Remove cache

95de267

Updated the tests

5b3311f

Rob--W reviewed Oct 30, 2019

View reviewed changes

Manvel added 3 commits October 31, 2019 00:12

Moved to own description Added test for getters and setters

9c6766b

Added clear and

371ca99

Wrap all properties in * namespace

2e55c6f

Rob--W reviewed Nov 2, 2019

View reviewed changes

Rob--W reviewed Nov 7, 2019

View reviewed changes

explicitly spell out network, services, and websites

b2affbd

Rob--W approved these changes Nov 7, 2019

View reviewed changes

Rob--W merged commit 87bdfa8 into mozilla:master Nov 21, 2019

Rob--W mentioned this pull request Nov 21, 2019

Half of the chrome privacy API is not available #204

Closed

	(hasOwnProperty(wrappers, prop) \|\|
	hasOwnProperty(metadata, prop))) {
	// This is an object that we need to do some wrapping for the children
	// of. Create a sub-object wrapper for it with the appropriate child
	// metadata.
	value = wrapObject(value, wrappers[prop], metadata[prop]);

	describe("without side effects", () => {
	it("should proxy non-wrapped methods", () => {
	let lazyInitCount = 0;
	const fakeChrome = {
	get runtime() {
	// Chrome lazily initializes API objects by replacing the getter with
	// the value. The initialization is only allowed to occur once,
	// after that `undefined` is returned and a warning is printed.
	// https://chromium.googlesource.com/chromium/src/+/4d6b3a067994ce6dcf0ed9a9efd566c083736952/extensions/renderer/module_system.cc#414
	//
	// The polyfill should invoke the getter only once (on the global chrome object).
	++lazyInitCount;

	const onMessage = {
	addListener(listener) {
	equal(this, onMessage, "onMessage.addListener should be called on the original chrome.onMessage object");
	},
	};
	const value = {onMessage};
	Object.defineProperty(this, "runtime", {value});
	return value;
	},
	get tabs() {
	ok(false, "chrome.tabs should not lazily be initialized without explicit API call");
	},
	};
	return setupTestDOMWindow(fakeChrome).then(window => {
	// This used to be equal(lazyInitCount, 0, ...), but was changed to
	// accomodate a change in the implementation of the polyfill.
	// To verify that APIs are not unnecessarily initialized, the fakeChrome
	// object has a "tabs" getter that fails the test upon access.
	equal(lazyInitCount, 1, "chrome.runtime should be initialized because chrome.runtime.id is accessed during polyfill initialization");

	window.browser.runtime.onMessage.addListener(() => {});
	equal(lazyInitCount, 1, "chrome.runtime should be initialized upon accessing browser.runtime");

	window.browser.runtime.onMessage.addListener(() => {});
	equal(lazyInitCount, 1, "chrome.runtime should be re-used upon accessing browser.runtime");

	window.chrome.runtime.onMessage.addListener(() => {});
	equal(lazyInitCount, 1, "chrome.runtime should be re-used upon accessing chrome.runtime");
	});
	});
	});

Uh oh!

Conversation

Manvel commented Oct 14, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manvel Oct 29, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manvel commented Oct 29, 2019

Uh oh!

Rob--W left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Manvel commented Oct 29, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Manvel Oct 29, 2019 •

edited

Loading