lib: make the global console [[Prototype]] an empty object #23509

joyeecheung · 2018-10-12T17:25:42Z

lib: make the global console [[Prototype]] an empty object

From the WHATWG console spec:

For historical web-compatibility reasons, the namespace object for
console must have as its [[Prototype]] an empty object, created as
if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%.

Since in Node.js, the Console constructor has been exposed through
require('console'), we need to keep the Console constructor but
we cannot actually use new Console to construct the global console.

This patch changes the prototype chain of the global console object,
so the console.Console.prototype is not in the global console prototype
chain anymore.

const proto = Object.getPrototypeOf(global.console);
// Before this patch
proto.constructor === global.console.Console
// After this patch
proto.constructor === Object

But, we still maintain that

global.console instanceof global.console.Console

through a custom Symbol.hasInstance function of Console that tests
for a special symbol kIsConsole for backwards compatibility.

This fixes a case in the console Web Platform Test that we commented
out.

Refs: https://console.spec.whatwg.org/#console-namespace
Refs: whatwg/console#3

make -j4 test (UNIX), or vcbuild test (Windows) passes
tests and/or benchmarks are included
commit message follows commit guidelines

nodejs-github-bot · 2018-10-12T17:25:48Z

@joyeecheung build started: https://ci.nodejs.org/blue/organizations/jenkins/node-test-pull-request-lite-pipeline/detail/node-test-pull-request-lite-pipeline/1214/pipeline

joyeecheung · 2018-10-12T17:26:29Z

Defensively marked as semver-major. cc @nodejs/tsc

CI: https://ci.nodejs.org/job/node-test-pull-request/17753/

mcollina

LGTM

TimothyGu

I’d like to measure userland breakage though, in case anyone is depending on console instanceof console.Console being true. If so maybe we can have a custom Symbol.hasInstance function to mitigate that.

joyeecheung · 2018-10-12T23:26:35Z

Master CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1581/
PR CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1582/

cc @TimothyGu

joyeecheung · 2018-10-14T05:14:59Z

(Previous CITGM failed because disk on test-osuosl-aix61-ppc64_be-2 was full)
Master CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1585/
PR CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1586/

thefourtheye · 2018-10-15T14:03:33Z

lib/console.js

+  ConsoleImpl.call(this, ...args);
+}
+
+// Reflect.ownKeys() is used here for for retrieving Symbols


Nit: Double for

thefourtheye · 2018-10-15T14:04:21Z

lib/console.js

-function noop() {}
+function Console(...args) {
+  if (!(this instanceof Console)) {
+    return new Console(...arguments);


I think ...args would be better. I prefer to avoid arguments as much as possible.

This is a copy-paste but yeah probably also makes sense to change it in this patch

jdalton · 2018-10-15T17:11:11Z

I've seen interesting Console use in jest, I don't have the specifics handy though. I suspect there may be breaks for a few so the major bump makes sense.

lib/console.js

jdalton · 2018-10-15T17:41:11Z

lib/console.js

+// we cannot actually use `new Console` to construct the global console.
+// Therefore, the console.Console.prototype is not
+// in the global console prototype chain anymore.
+const consolePrototype = Object.create(Object.prototype);


Isn't Object.create(Object.prototype) just {} ?

I don't have an opinion on this, it just somehow seems more readable to be if we just do exactly what the spec says (but probably {} is more readable for people who skip the comments).

thefourtheye · 2018-10-15T19:04:46Z

@jdalton Then it would make sense to include jest in CITGM if it is not there already. cc @nodejs/citgm

richardlau · 2018-10-15T19:46:21Z

@jdalton Then it would make sense to include jest in CITGM if it is not there already. cc @nodejs/citgm

Currently blocked because Jest uses yarn for tests: nodejs/citgm#560

jdalton · 2018-10-15T20:16:12Z

lib/console.js

-function noop() {}
+function Console(...args) {
+  if (!(this instanceof Console)) {
+    return new Console(...arguments);


Could use Reflect.construct(Console, args)

jdalton · 2018-10-15T20:16:41Z

lib/console.js


-function noop() {}
+function Console(...args) {
+  if (!(this instanceof Console)) {


instead of !(this instanceof Console) could use !new.target

Not that anybody does that, but these would behave differently if somebody did ES5-style inheritance from this class, right? The current version is also a bit more idiomatic, imo (same for the comment above…)

jdalton · 2018-10-15T20:27:30Z

I'm a little fuzzy on the need for ConsoleImpl. It looks like the gist of this PR is to make the global console no longer an instance of Console which should be doable, by copying the own and inherited props of a new Console(stdout, stderr) instance (minus constructor) to an empty object {}, without rewiring Console to use an internal ConsoleImpl.

joyeecheung · 2018-10-16T00:54:58Z

by copying the own and inherited props of a new Console(stdout, stderr) instance (minus constructor) to an empty object {} without rewiring Console to use an internal ConsoleImpl.

I am a bit conflicted on the two approach (adding props to an object v.s. removing props from a Console). One upside about using the name ConsoleImpl is so that we don't have to put a bunch of comments to remind whoever wants to modify the file later that the global console is not actually an instance of Console - the code there are just implementations that will be wired to the global console. Somehow it seems to me that if we go with removing things, then we could accidentally forget to remove something added later, but I don't have strong opinions on this, it's just a hunch.

jdalton · 2018-10-16T01:03:42Z

I think this is overcomplicating the PR. Here's a snippet from my own implementation in a related project.

const globalConsole = assignProperties({}, new Console(stdout, stderr))

There is no need to rewire Console or abstract through an intermediate ConsoleImpl. The line above accomplishes the goal of making globalConsole assigned to a plain object without a heavy rewrite. The assignProperties util is doing a loop over Reflect.ownKeys() and Reflect.defineProperty(o, k, Reflect.getOwnPropertyDescriptor(s, k))

devsnek · 2018-10-16T01:06:01Z

personally i would do this by having a ConsolePrototype object which is used with Object.create like in our ReadableStreamAsyncIterator implementation. https://github.com/nodejs/node/blob/master/lib/internal/streams/async_iterator.js#L130

nvm i realize that it has to be a plain object. current implementation seems fine.

jdalton

Please look into simplifying this PR. I believe the gist of the PR can be done in a handful of lines without major refactoring to Console. See #23509 (comment).

joyeecheung · 2018-10-16T01:18:34Z

@jdalton

What do you think about #23509 (comment) ? So far 33 contributors have committed in this file, so I believe it's worth the cost to rename the class to hint future contributors that this is an implementation without having them to scroll all the way down to see the comments.

jdalton · 2018-10-16T01:26:44Z

@joyeecheung

What do you think about #23509 (comment) ?

Class renaming, or other deeper refactoring, is probably better suited for a PR specifically for that.
This PR, for a WHATWG console spec compliance fix, is less great for that.

Trott · 2018-10-16T04:48:23Z

Class renaming, or other deeper refactoring, is probably better suited for a PR specifically for that.
This PR, for a WHATWG console spec compliance fix, is less great for that.

@jdalton I don't think the question was if we should do refactoring in this PR. I think the question was what you thought about the argument made in that comment that the approach currently in this PR is more maintainable in that it will be less error-prone for future contributors to modify. Here's the comment:

One upside about using the name ConsoleImpl is so that we don't have to put a bunch of comments to remind whoever wants to modify the file later that the global console is not actually an instance of Console - the code there are just implementations that will be wired to the global console. Somehow it seems to me that if we go with removing things, then we could accidentally forget to remove something added later, but I don't have strong opinions on this, it's just a hunch.

jdalton · 2018-10-16T05:03:44Z

@Trott

I think the question was what you thought about the argument made in that comment that the approach currently in this PR is more maintainable in that it will be less error-prone for future contributors to modify.

Oh, I don't think the approach in this PR promotes those things.
I'd be against it as is, but wouldn't be against further iterating on it in a more focused PR.

SimenB · 2018-10-22T22:05:13Z

Jest team member here 👋 Saw the link from the CITGM PR, and in Jest we do class OurConsole extends Console and super.log('stuff'). I haven't looked at the code in this PR, but do you think that would break? If it's just changing instanceof I don't think that matters to us.

Code:
https://github.com/facebook/jest/blob/cdf7a222420ae2ad21c76f15dd155e2ebacad79d/packages/jest-util/src/Console.js
https://github.com/facebook/jest/blob/cdf7a222420ae2ad21c76f15dd155e2ebacad79d/packages/jest-util/src/buffered_console.js

An instance of that subclass is injected into the user code (running as a vm.Script), overwriting their global.console. We do this both so we can buffer up console output so it doesn't interfere with our own output, so we can give a line and column of where the log came from, and so we can pass it to reporters.

joyeecheung · 2018-10-23T00:29:11Z

@SimenB Hi! Thanks for the explanation, based on that I don't think this would break Jest, since after this patch the lookups on the global console and other Console instances still arrive to the same stuff - just a matter of through the prototype chain or not.

trygve-lie · 2018-10-23T06:35:51Z

I have a module where this will be breaking: https://github.com/trygve-lie/abslog/blob/master/lib/log.js#L8

The module is an abstract log API making it possible for modules to have log statements without depending on a full log library. A consumer of a module using this abstraction are then able to pass in a logger of their choice as long it comply with the API of the abstraction. This is btw a pattern used in fastify.js, though this module differ from the one in fastify.js that this one support passing in console as an exception despite that it does not comply with the API.

If so maybe we can have a custom Symbol.hasInstance function to mitigate that.

Having something like that would be very handy.

joyeecheung · 2018-10-23T08:05:52Z

@trygve-lie Thanks for the explanation! Indeed, we should implement a Symbol.hasInstance to mitigate that - I'll find sometime later this week to update this PR.

From the WHATWG console spec: > For historical web-compatibility reasons, the namespace object for > console must have as its [[Prototype]] an empty object, created as > if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%. Since in Node.js, the Console constructor has been exposed through require('console'), we need to keep the Console constructor but we cannot actually use `new Console` to construct the global console. This patch changes the prototype chain of the global console object, so the console.Console.prototype is not in the global console prototype chain anymore. ``` const proto = Object.getPrototypeOf(global.console); // Before this patch proto.constructor === global.console.Console // After this patch proto.constructor === Object ``` But, we still maintain that ``` global.console instanceof global.console.Console ``` through a custom Symbol.hasInstance function of Console that tests for a special symbol kIsConsole for backwards compatibility. This fixes a case in the console Web Platform Test that we commented out. Refs: https://console.spec.whatwg.org/#console-namespace Refs: whatwg/console#3

joyeecheung · 2018-10-25T12:56:20Z

@jdalton I looked into the alternative implementation a bit, and noticed that simply assigning properties of a Console instance to a new object would result in a behavior change - the methods called on the global console will have the temporary Console instance as their contexts since the Console constructor binds all the method to the instance being constructed (so that users can do const { log } = console). So for instance, monkey patching properties of the global console could be broken:

console._stdout = undefined;
// Should've thrown because we have broken _stdout, but doesn't because log
// actually gets called on another object.
console.log('test');

In the context of the Web, this is expected because console is a namespace and stuff like console._stdout are not a thing, but in the context of Node.js, this is probably a bit too breaking - I am pretty sure people are monkey patching these properties in the wild, or assume that they carry the states as altered by the methods, so these are probably properties that we even need to think twice before starting a proper deprecation cycle. That's probably a bit too far away from the intention of this PR, though.

I've patched that through a condition in the loop to rebind the methods to the global console.

joyeecheung · 2018-10-25T13:00:46Z

Another note: I went with new.target in the constructor to implement Symbol.hasInstance, and added a few tests for the instanceof behavior.

CI: https://ci.nodejs.org/job/node-test-pull-request/18145/
Master CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1598/
PR CITGM: https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1599/

jdalton · 2018-10-25T14:47:23Z

@joyeecheung Thank you for the update on your findings and the PR!

danbev · 2018-10-26T03:43:58Z

Landed in 6223236.

From the WHATWG console spec: > For historical web-compatibility reasons, the namespace object for > console must have as its [[Prototype]] an empty object, created as > if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%. Since in Node.js, the Console constructor has been exposed through require('console'), we need to keep the Console constructor but we cannot actually use `new Console` to construct the global console. This patch changes the prototype chain of the global console object, so the console.Console.prototype is not in the global console prototype chain anymore. ``` const proto = Object.getPrototypeOf(global.console); // Before this patch proto.constructor === global.console.Console // After this patch proto.constructor === Object ``` But, we still maintain that ``` global.console instanceof global.console.Console ``` through a custom Symbol.hasInstance function of Console that tests for a special symbol kIsConsole for backwards compatibility. This fixes a case in the console Web Platform Test that we commented out. PR-URL: #23509 Refs: whatwg/console#3 Refs: https://console.spec.whatwg.org/#console-namespace Reviewed-By: Gus Caplan <[email protected]> Reviewed-By: Matteo Collina <[email protected]> Reviewed-By: Tiancheng "Timothy" Gu <[email protected]> Reviewed-By: Denys Otrishko <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Rich Trott <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Sakthipriyan Vairamani <[email protected]> Reviewed-By: John-David Dalton <[email protected]>

joyeecheung · 2018-10-26T10:32:37Z

Post-mortem: the two new CITGM failures did not appear to be relevant:

https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1599/nodes=fedora-last-latest-x64/testReport/junit/(root)/citgm/bluebird_v3_5_2/
https://ci.nodejs.org/view/Node.js-citgm/job/citgm-smoker/1599/nodes=win-vs2017/testReport/junit/(root)/citgm/body_parser_v1_18_3/

From the WHATWG console spec: > For historical web-compatibility reasons, the namespace object for > console must have as its [[Prototype]] an empty object, created as > if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%. Since in Node.js, the Console constructor has been exposed through require('console'), we need to keep the Console constructor but we cannot actually use `new Console` to construct the global console. This patch changes the prototype chain of the global console object, so the console.Console.prototype is not in the global console prototype chain anymore. ``` const proto = Object.getPrototypeOf(global.console); // Before this patch proto.constructor === global.console.Console // After this patch proto.constructor === Object ``` But, we still maintain that ``` global.console instanceof global.console.Console ``` through a custom Symbol.hasInstance function of Console that tests for a special symbol kIsConsole for backwards compatibility. This fixes a case in the console Web Platform Test that we commented out. PR-URL: nodejs#23509 Refs: whatwg/console#3 Refs: https://console.spec.whatwg.org/#console-namespace Reviewed-By: Gus Caplan <[email protected]> Reviewed-By: Matteo Collina <[email protected]> Reviewed-By: Tiancheng "Timothy" Gu <[email protected]> Reviewed-By: Denys Otrishko <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Rich Trott <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Sakthipriyan Vairamani <[email protected]> Reviewed-By: John-David Dalton <[email protected]>

Specifically for v11.x. Refs: nodejs#23509

From the WHATWG console spec: > For historical web-compatibility reasons, the namespace object for > console must have as its [[Prototype]] an empty object, created as > if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%. Since in Node.js, the Console constructor has been exposed through require('console'), we need to keep the Console constructor but we cannot actually use `new Console` to construct the global console. This patch changes the prototype chain of the global console object, so the console.Console.prototype is not in the global console prototype chain anymore. ``` const proto = Object.getPrototypeOf(global.console); // Before this patch proto.constructor === global.console.Console // After this patch proto.constructor === Object ``` But, we still maintain that ``` global.console instanceof global.console.Console ``` through a custom Symbol.hasInstance function of Console that tests for a special symbol kIsConsole for backwards compatibility. This fixes a case in the console Web Platform Test that we commented out. PR-URL: #23509 Refs: whatwg/console#3 Refs: https://console.spec.whatwg.org/#console-namespace Reviewed-By: Gus Caplan <[email protected]> Reviewed-By: Matteo Collina <[email protected]> Reviewed-By: Tiancheng "Timothy" Gu <[email protected]> Reviewed-By: Denys Otrishko <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Rich Trott <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Sakthipriyan Vairamani <[email protected]> Reviewed-By: John-David Dalton <[email protected]>

Specifically for v11.x. PR-URL: #25420 Refs: #23509 Reviewed-By: Ruben Bridgewater <[email protected]> Reviewed-By: Anna Henningsen <[email protected]>

From the WHATWG console spec: > For historical web-compatibility reasons, the namespace object for > console must have as its [[Prototype]] an empty object, created as > if by ObjectCreate(%ObjectPrototype%), instead of %ObjectPrototype%. Since in Node.js, the Console constructor has been exposed through require('console'), we need to keep the Console constructor but we cannot actually use `new Console` to construct the global console. This patch changes the prototype chain of the global console object, so the console.Console.prototype is not in the global console prototype chain anymore. ``` const proto = Object.getPrototypeOf(global.console); // Before this patch proto.constructor === global.console.Console // After this patch proto.constructor === Object ``` But, we still maintain that ``` global.console instanceof global.console.Console ``` through a custom Symbol.hasInstance function of Console that tests for a special symbol kIsConsole for backwards compatibility. This fixes a case in the console Web Platform Test that we commented out. PR-URL: nodejs#23509 Refs: whatwg/console#3 Refs: https://console.spec.whatwg.org/#console-namespace Reviewed-By: Gus Caplan <[email protected]> Reviewed-By: Matteo Collina <[email protected]> Reviewed-By: Tiancheng "Timothy" Gu <[email protected]> Reviewed-By: Denys Otrishko <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Rich Trott <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Sakthipriyan Vairamani <[email protected]> Reviewed-By: John-David Dalton <[email protected]>

Specifically for v11.x. PR-URL: nodejs#25420 Refs: nodejs#23509 Reviewed-By: Ruben Bridgewater <[email protected]> Reviewed-By: Anna Henningsen <[email protected]>

nodejs-github-bot added the console Issues and PRs related to the console subsystem. label Oct 12, 2018

joyeecheung added the semver-major PRs that contain breaking changes and should be released in the next major version. label Oct 12, 2018

devsnek approved these changes Oct 12, 2018

View reviewed changes

mcollina approved these changes Oct 12, 2018

View reviewed changes

TimothyGu approved these changes Oct 12, 2018

View reviewed changes

lundibundi approved these changes Oct 12, 2018

View reviewed changes

jasnell approved these changes Oct 12, 2018

View reviewed changes

Trott approved these changes Oct 12, 2018

View reviewed changes

cjihrig approved these changes Oct 14, 2018

View reviewed changes

thefourtheye approved these changes Oct 15, 2018

View reviewed changes

jdalton reviewed Oct 15, 2018

View reviewed changes

lib/console.js Show resolved Hide resolved

jdalton reviewed Oct 15, 2018

View reviewed changes

jdalton suggested changes Oct 16, 2018

View reviewed changes

trygve-lie mentioned this pull request Oct 23, 2018

Track and cater for upcomming changes in console.Console trygve-lie/abslog#14

Closed

joyeecheung force-pushed the fix-console branch from 29c4bd0 to 5556e47 Compare October 25, 2018 12:51

jdalton approved these changes Oct 25, 2018

View reviewed changes

Trott added the author ready PRs that have at least one approval, no pending requests for changes, and a CI started. label Oct 25, 2018

danbev closed this Oct 26, 2018

joyeecheung mentioned this pull request Jan 1, 2019

console: lazy load process.stderr and process.stdout #24534

Closed

3 tasks

joyeecheung added a commit to joyeecheung/node that referenced this pull request Jan 9, 2019

console: create the global console from Console constructor

c67021a

Specifically for v11.x. Refs: nodejs#23509

joyeecheung mentioned this pull request Jan 9, 2019

[v11.x] backport console related changes #25420

Closed

2 tasks

BridgeAR pushed a commit that referenced this pull request Jan 9, 2019

console: create the global console from Console constructor

b2b0645

Specifically for v11.x. PR-URL: #25420 Refs: #23509 Reviewed-By: Ruben Bridgewater <[email protected]> Reviewed-By: Anna Henningsen <[email protected]>

BridgeAR added the backported-to-v11.x label Jan 9, 2019

BridgeAR mentioned this pull request Jan 16, 2019

v11.7.0 proposal #25537

Merged

MylesBorins mentioned this pull request Jan 24, 2019

v11.8.0 proposal #25687

Merged

lib: make the global console [[Prototype]] an empty object #23509

lib: make the global console [[Prototype]] an empty object #23509

Conversation

joyeecheung commented Oct 12, 2018 • edited Loading

nodejs-github-bot commented Oct 12, 2018

joyeecheung commented Oct 12, 2018

mcollina left a comment

Choose a reason for hiding this comment

TimothyGu left a comment

Choose a reason for hiding this comment

joyeecheung commented Oct 12, 2018

joyeecheung commented Oct 14, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jdalton commented Oct 15, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

thefourtheye commented Oct 15, 2018

richardlau commented Oct 15, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jdalton commented Oct 15, 2018 • edited Loading

joyeecheung commented Oct 16, 2018

jdalton commented Oct 16, 2018 • edited Loading

devsnek commented Oct 16, 2018 • edited Loading

jdalton left a comment

Choose a reason for hiding this comment

joyeecheung commented Oct 16, 2018

jdalton commented Oct 16, 2018

Trott commented Oct 16, 2018

jdalton commented Oct 16, 2018

SimenB commented Oct 22, 2018 • edited Loading

joyeecheung commented Oct 23, 2018

trygve-lie commented Oct 23, 2018

joyeecheung commented Oct 23, 2018

joyeecheung commented Oct 25, 2018 • edited Loading

joyeecheung commented Oct 25, 2018 • edited Loading

jdalton commented Oct 25, 2018

danbev commented Oct 26, 2018

joyeecheung commented Oct 26, 2018 • edited Loading

joyeecheung commented Oct 12, 2018 •

edited

Loading

jdalton commented Oct 15, 2018 •

edited

Loading

jdalton commented Oct 16, 2018 •

edited

Loading

devsnek commented Oct 16, 2018 •

edited

Loading

SimenB commented Oct 22, 2018 •

edited

Loading

joyeecheung commented Oct 25, 2018 •

edited

Loading

joyeecheung commented Oct 25, 2018 •

edited

Loading

joyeecheung commented Oct 26, 2018 •

edited

Loading