tools: remove gyp from tools directory

[maclover7]

Rely on gyp via npm's node-gyp dependency. All changes to gyp
must now be sent to https://github.com/nodejs/node-gyp.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[nodejs-github-bot]

@maclover7 sadly an error occured when I tried to trigger a build :(

[maclover7]

This is on hold until #22342 lands, which contains [email protected]

[refack]

We can't do that. The node-gyp that is a part of npm is managed by npm not us. We should not add a dependency on that.

[refack]

Rely on gyp via npm's node-gyp dependency. All changes to gyp
must now be sent to nodejs/node-gyp.

RE that; those two forks have been "intentionally" diverging. node/tools/gyp is an internal tool, and we (sort of) have control of it's usage. Hence we can be more liberal with patching it.
node-gyp is a user-land tool, we have much more limited control of it's usage, and ported usage. Hence we need to be very careful with patching this one.

IMHO a better solution would be to take responsibility of CIing patches, and vendoring to those two locations. I'm working on that, but time...

BTW: we have many opportunities for optimizations and streamlining of our build process via GYP.

/CC @nodejs/gyp

[maclover7]

those two forks have been "intentionally" diverging. node/tools/gyp is an internal tool, and we (sort of) have control of it's usage.

I had thought of node core's gyp as being a kind of testing ground for the community version distributed via node-gyp. In theory, any patches that fix bugs in gyp should be rolled out to all end users via node-gyp, and shouldn't exclusively live in node core's gyp. Also, it has been difficult keeping more than one copy of gyp in sync, and this commit creates a single place where gyp maintenance efforts can be concentrated. Since Node.js maintains this codebase now, this alone is a good enough reason, for me at least, to land this.

[BridgeAR]

Since Node.js maintains node-gyp as well, I think this is fine. Especially since each npm update that would change something would result in failures on our side and we would be able to solve them then.
I like the single source of truth and it should also result in a smaller binary.

[BridgeAR]

@nodejs/gyp

[refack]

Also, it has been difficult keeping more than one copy of gyp in sync, and this commit creates a single place where gyp maintenance efforts can be concentrated.

IMHO it's hasn't been that difficult.

Since Node.js maintains this codebase now, this alone is a good enough reason, for me at least, to land this.

That is not officially true, the upstream project is not dead, and the node org did not official take responsibility for GYP

Anyway, adding a dependency on npm IMHO is sub-optimal. Especially since:

node/doc/guides/maintaining-npm.md

Lines 7 to 9 in 2ce0380

	Two weeks after the "latest" release has been promoted it can land on master
	assuming no major regressions are found. There are no additional constraints
	for Semver-Major releases.

i.e. if a bug or update is needed, we'll need to patch node-gyp, release a version, get it landed in npm, wait for npm release, then wait at least two weeks before it can land.
This also essentially excludes fixing in backports.

tl;dr I'm a hard -1, if needed this can go the TSC.

[gibfahn]

i.e. if a bug or update is needed, we'll need to patch node-gyp, release a version, get it landed in npm, wait for npm release, then wait at least two weeks before it can land.
This also essentially excludes fixing in backports.

So we have to accept that we may end up carrying patches on top of npm, in the same way we carry patches on OpenSSL and V8 etc.?

[refack]

So we have to accept that we may end up carrying patches on top of npm, in the same way we carry patches on OpenSSL and V8 etc.?

Why do that, when the status quo is IMHO better? We have an easy to use and maintain version of the tool. The benefits of removing this copy are minimal, and outweighed by the cost.

[refack]

Also as I wrote above, our current dependency is on npm as a whole. Breaking that abstraction and adding a dependency on a non public component of npm is super fragile (Ref: #21975).

If we were to replace GYP with a direct dependency on node-gyp that might be better, but IMHO it is still a reverse dependency. A change that has minimal benefit IMHO. We simply have two projects that are dependant on GYP (four if you count node-chakracore and libuv). I don't think we should change the dependency graph without refactoring the tool and it's usage.

[maclover7]

The benefits of removing this copy are minimal, and outweighed by the cost.

IMHO the question is about the tradeoffs of maintaining an internal version of a tool (node core's copy of gyp) while we already have to maintain/release/support essentially the same thing, but for the whole community (node-gyp's copy of gyp). The patches from node core's copy of gyp have already been applied to node-gyp, so we are now carrying two essentially identical gyp directories in nodejs/node (building addons doesn't even use node core's copy of gyp). It would seem to be better, to me at least, to centralize long-term maintenance efforts around the community version, so everyone can benefit from bugfixes, etc, and then to treat node core like any other downstream user, and float patches as necessary.

[Trott]

@maclover7 Any chance you're still around and want to see this through? If so, let's get it on TSC agenda. If not, anyone else interested in this or should it be closed? @nodejs/build @BridgeAR

[maclover7]

@Trott Yeah, I'm happy to see this through — agree that the next step would be to take this to the TSC due to the -1 from @refack

[rvagg]

Regarding Refael's concerns: on one hand npm have been very responsive to taking node-gyp updates, plus we have a precedent of patching the npm we release where it's served us (including node-gyp, e.g. 8b4af64, 0454725). On the other hand .. what about when this makes its way into LTS branches where we don't upgrade npm much but node-gyp becomes a problem for us for whatever reason and the fixes involve changes that might conflict with npm's use of node-gyp—or at least be undetermined and the risk of us breaking npm for downstream LTS users is large? That's going to put us in a bit of a corner with the easiest path to re-introduce our own fork into tools/ or deps/ again.

[BridgeAR]

@nodejs/tsc it was asked to have a look at this due to the -1.

@maclover7 this needs a rebase.

[Trott]

Maybe @nodejs/python or @nodejs/gyp would have thoughts. Based on @rvagg's comment above, I'd be inclined to go with whatever the most active Python/gyp folks on the project think is best, because they're the ones who are most likely to do the work to address any pain associated with whatever approach we take here.

[mcollina]

I share some of @refack concerns, but I’m happy to delegate this decision to the people that are going to do the work.

[refack]

I'd be inclined to go with whatever the most active Python/gyp folks on the project think is best,

I’m happy to delegate this decision to the people that are going to do the work.

Well this has stalled for a log time, and based on the resent feedback IMHO it could be closed, so I'm going to cast the die.
The door is always open for reconsideration in the future...