Revert "child_process: change windowsHide default to true"

[Trott]

This reverts commit 420d8af.

I'm opening this not because I'm convinced the commit needs to be reverted, but to hopefully elicit a prompt conversation and decision about whether or not it should be reverted.

Refs: #21316 (comment) and several subsequent comments.

I'm particularly interested in what embedders, especially Electron folks, think. @MarshallOfSound has already weighed in that they don't believe a revert is warranted, but also expressed it as a personal opinion (so, I assume, not speaking on behalf of Electron). Would love to hear from @codebytere or @zeke or @groundwater or someone, either to get a consensus of personal opinions or else to get a semi-official "Electron prefers revert/no-revert".

@nodejs/embedders

IIUC, this is only affecting developers and not end users. Doing this out of an abundance of caution and in case I"m wrong about that.

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

[nodejs-github-bot]

@Trott build started: https://ci.nodejs.org/blue/organizations/jenkins/node-test-pull-request-lite-pipeline/detail/node-test-pull-request-lite-pipeline/1458/pipeline

[Trott]

CI: https://ci.nodejs.org/job/node-test-pull-request/18304/

[bzoz]

FWIW this was added because it was requested by electron in libuv/libuv#1878

[ryzokuken]

@bzoz totally! You'd have to agree though that this did kinda bamboozle Electron. I mean, it was intended to hide phantom console windows from popping up, GUI windows were never in the equation (why would we even want that?).

That said, it's a problem that could be easily solved on Electron's end so this might not be necessary.

[refack]

I think the actual problem that was experienced by Electron was fixed in:
libuv/libuv#1558
and
libuv/libuv#1659
IIUC it was that even when windowsHide was set to true a console window would flash and disappear.

[ryzokuken]

@refack well, that'll be great news in favor of reverting.

I'll let y'all know what the consensus is on this. That'll have to wait until Monday though.

[refack]

This has some WIN32 minutia behind it. And I'm writing the following with some approximations, and only with my own understanding.
The Windows native binary format (PE) has in it's header a field named subsystem, which is commonly either console or windows. Only windows executables can interact with the user, so console binaries are hosted inside conhost.exe which is Windows' analog to a terminal emulator:

Node.js for Windows is compiled as a console app so for it to interact with the user it also runs through conhost:

So essentially you can only directly spawn console applications applications from node without creating a new Window. Also if you want to spawn a "detached" process (one that's life-cycle is independent of its parent) it has to have it's own window. That window can be "hidden", but it's there.

---

For reference, the most similar "native" APIs are

• The START command of CMD.exe - It has an explicit /MIX to open new window minimized and /B to not create a new window:
  
• PowerShell's Start-Process - Has an explicit -NoNewWindow to not spawn a new window:
  

(other runtimes such as ActiveState Perl or cpython , have a similar API with the default to hidden=false)

[refack]

@yodurr or @bitcrazed could you help me correct any mistakes in the above ^^^^

[codebytere]

Electron can work around this, so we'd be +0 on a revert, i would say!

[Trott]

Explicitly pinging the author and approvers of the commit potentially being reverted: @cjihrig @TimothyGu @bzoz @apapirovski @trivikr @addaleax @ryzokuken @BridgeAR

[cjihrig]

In a perfect world, I think defaulting to hiding the window is the correct behavior. This doesn't impact me personally, so I'm fine with deferring to the people that are impacted. I will say though that I was under the impression that only console windows would be impacted by the change.

[Trott]

FWIW this was added because it was requested by electron in libuv/libuv#1878

Talked a bit with @codebytere tonight and the upshot seems to be that a change in libuv was requested, but that this change in Node.js was not quite what they wanted. They can live with this, but do have a slight preference to it being reverted.

[Trott]

Approvals on this from two Electron team members and two Windows-centric collaborators. I'm going to land on master in 24 hours if there's no objections.

Since this is reverting a semver-major, landing on v11.x-staging may require some TSC buy-in so: ping @nodejs/tsc for concerns/objections. I discussed this briefly with @ofrobots yesterday and he was in favor of reverting on 11.x rather than waiting for more people to rely on the new behavior. I concur, so that's two of us....

[sindresorhus]

It seems like everyone agrees that hiding is the best default for users, so I'm surprised you're making Node.js worse for end-users just because of an embedder. Electron can work around this. If this is merged, every usage of childProcess.spawn() will have to include this option. I strongly oppose this change.

[Trott]

Last CI run was all green, but it was also 10 days ago, and a lot of code has landed since then. So just to be super extra special careful, here's another CI: https://ci.nodejs.org/job/node-test-pull-request/18563/

[Fishrock123]

Wait was this actually not desirable?

[Trott]

Resume Build CI: https://ci.nodejs.org/job/node-test-pull-request/18590/

[Trott]

Landed in ad6ead3