src: use new V8 API in vm module

[fhinkel]

This PR removes the CopyProperties() hack in the vm module, i.e., Contextify.
Instead, it uses different V8 API methods, that allow interception of
DefineProperty() and do not flatten accessor descriptors to
property descriptors.

Move many known issues to test cases. Factor out the last test in
test-vm-context.js for
#10223
into its own file, test-vm-strict-assign.js.

Part of this CL is taken from a stalled PR by
https://github.com/AnnaMag
#13265

Refs: #6283
Refs: #15114
Refs: #13265

Fixes: #2734
Fixes: #10223
Fixes: #11803
Fixes: #11902

This PR requires a backport of
37a3a15 otherwise some tests will fail. Cherry pick PR.

Checklist

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

Affected core subsystem(s)

src

Checklist

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

Affected core subsystem(s)

src

[jasnell]

largely rubber stamp lgtm

[bnoordhuis]

Can you use the overload that takes a Local<Context>?

[fhinkel]

Done.

[bnoordhuis]

I'd write this as an if statement. It's rather hard to read.

[fhinkel]

Agree. Done.

[bnoordhuis]

I suppose the conditional isn't needed?

[fhinkel]

Why? If we intercept, it signals that the property was found. I think that's not correct if the property is not present.

[bnoordhuis]

Undefined is the default return value of API callbacks. I.e., the return value is going to be the same, with or without the IsUndefined() check.

[fhinkel]

ReturnValue on the CallbackArgs is set to undefined, but the call to the interceptor returns true or false depending on whether the return value was set. This then changes the control flow, see objects.cc.

But taking a closer look, I think the logic can be substantially simplified: take the value from the sandbox if present, otherwise proceed regularly. What do you think?

[bnoordhuis]

This fits on one line (well, two, if you count the LHS.)

[fhinkel]

Done.

[bnoordhuis]

Stray debug code, I presume?

[fhinkel]

oops

[bnoordhuis]

If this function gets called a lot (which it looks like it is), it's arguably better to use snprintf() instead to avoid the extra allocation, or even to do manual stringification. C++17 adds std::to_chars() but we can't use that yet, unfortunately.

It might even be worthwhile to implement this as Uint32::New(...)->ToString(...) because that would reuse V8's number-to-string cache.

[fhinkel]

Yes, if there's any larger array, this is called for every index. Using Uint32::New()->ToString().

[bnoordhuis]

You fixed the style for this further up but not here.

[fhinkel]

Done.

[bnoordhuis]

Debug code again.

There's a lot of duplication. Couldn't you implement them by calling UIntToName(isolate, index) and passing the result to GlobalPropertyDefinerCallback(), etc.?

[fhinkel]

It's all duplication. Let me think about how to get rid of the code duplication while maintaining readability.

[bnoordhuis]

Can you drop the comment?

[fhinkel]

Done.

[bnoordhuis]

You can drop the copyright boilerplate, we don't use it in new files.

[fhinkel]

Done

[fhinkel]

@cpojer, since this is fixing bugs/altering behavior, do you want to check if that causes problems for Jest?

[TimothyGu]

I'll run this through the jsdom test suite as well.

[fhinkel]

Nits addressed, refactoring the code duplication part. Will squash afterwards.

[bnoordhuis]

See my comment about the default return value but otherwise LGTM. Nice work, Franziska and @AnnaMag.

[bnoordhuis]

Some extraneous blank lines here.

[fhinkel]

Done.

[bnoordhuis]

Can you remove the blank line here?

[cpojer]

@fhinkel mind running this against Jest's version on master?

[fhinkel]

@cpojer Jest's tests helped me find a fatal error on my side 😄
I get the same test failures with Node master with and without my patch on Jest master. I guess that means you're not relying on any of our broken behavior that this patch fixes.

jest/integration_tests/coverage-threshold/__tests__/a-banana.js
1:1 error Delete `⏎······` prettier/prettier

[cpojer]

Can you just try yarn jest on the repo, that'll run the JS tests? It seems like there is some weird eslint issue on your state of the Jest repo (master is green, so not sure what's up).

[fhinkel]

@cpojer I'm not getting very far with Node master, opend an issue here jestjs/jest#4731.

[SimenB]

As mentioned in the Jest issue, using node 9 RC fails worker-farm's tests as well. Jest uses worker-farm to parallelise test runs.

https://github.com/rvagg/node-worker-farm

Details

$ npm version{ 'worker-farm': '1.5.0', npm: '5.3.0', ares: '1.13.0', cldr: '31.0.1', http_parser: '2.7.0', icu: '59.1', modules: '58', nghttp2: '1.25.0', node: '9.0.0-rc.0', openssl: '1.0.2l', tz: '2017b', unicode: '9.0', uv: '1.15.0', v8: '6.1.534.42', zlib: '1.2.11' }

> worker-farm@1.5.0 test /Users/simbekkh/repos/node-worker-farm
> node ./tests/
TAP version 13
# simple, exports=function test
ok 1 pid makes sense
ok 2 pid makes sense
ok 3 rnd result makes sense
ok 4 workerFarm ended
# simple, exports.fn test
ok 5 pid makes sense
ok 6 pid makes sense
ok 7 rnd result makes sense
ok 8 workerFarm ended
# single worker
ok 9 only a single process (by pid)
ok 10 workerFarm ended
# two workers
ok 11 only two child processes (by pid)
ok 12 workerFarm ended
# many workers
ok 13 pids are all the same (by pid)
ok 14 workerFarm ended
# auto start workers
ok 15 child has been up before the request
ok 16 child has been up before the request
ok 17 child has been up before the request
ok 18 workerFarm ended
# single call per worker
ok 19 one process for each call (by pid)
ok 20 workerFarm ended
# two calls per worker
ok 21 one process for each call (by pid)
ok 22 workerFarm ended
# many concurrent calls
ok 23 processed tasks concurrently (173ms)
ok 24 workerFarm ended
# single concurrent call
ok 25 processed tasks sequentially (304ms)
ok 26 workerFarm ended
# multiple concurrent calls
ok 27 processed tasks concurrently (277ms)
ok 28 workerFarm ended
# durability
events.js:195
throw er; // Unhandled 'error' event
^
Error [ERR_IPC_CHANNEL_CLOSED]: Channel closed
at ChildProcess.target.send (internal/child_process.js:606:16)
at Object.send (/Users/simbekkh/repos/node-worker-farm/lib/fork.js:24:17)
at Farm.stopChild (/Users/simbekkh/repos/node-worker-farm/lib/farm.js:131:11)
at Farm.<anonymous> (/Users/simbekkh/repos/node-worker-farm/lib/farm.js:96:10)
at ontimeout (timers.js:478:11)
at tryOnTimeout (timers.js:302:5)
at Timer.listOnTimeout (timers.js:262:5)
npm ERR! Test failed. See above for more details.

[SimenB]

git bisect blames f2b01cb

#4670

Bisect script if anyone wants to double-check

Details

#!/bin/sh
./configure
make -j4 ||exit 125 # an exit code of 125 asks "git bisect" to "skip" the current commit# run the application and check that it produces good output
./node ../node-worker-farm/tests 2>&1| grep 'ERR_IPC_CHANNEL_CLOSED'if [ $?-eq 0 ];thenexit 1
fiexit 0

I can confirm that reverting f2b01cb makes both worker-farm's and jest's test suites pass

[fhinkel]

@SimenB Do you mind putting your code snippets in details, as they are not related to this issue? Thanks!

[fhinkel]

CI: https://ci.nodejs.org/job/node-test-pull-request/10919/

[fhinkel]

New CI: https://ci.nodejs.org/job/node-test-pull-request/10920/

[fhinkel]

Landed in f1d6b04

[targos]

Nice. Thank you and @AnnaMag for doing this!