Deprecating _third_party_main

[BridgeAR]

I stumbled upon some very old code in the loader with the following code comment:

// To allow people to extend Node in different ways, this hook allows
// one to drop a file lib/_third_party_main.js into the build
// directory which will be executed instead of Node's normal loading.

So instead of the regular loader it is going to load the new file. However, that is only the case if that file is also added to node.gyp. This "feature" has never been documented and seems to be used very rarely when searching for it with google. I am not really sure if the best way to deal with this is to document this or to deprecate it but keeping it as is does not seem to be ideal. Is this something we really want to support officially? This functionality seems rather odd to me.

Ping @nodejs/tsc

[addaleax]

I was talking about this with @rubys at JSI – They were looking into splitting the bootstrap function into two parts as a better way for addressing the embedder use case here (the parts are a) basically everything that happens before _third_party_main, i.e. setup code and b) everything after it, i.e. starting userland scripts).

If embedders are given the option of only executing bootstrap code up to the point where we’d currently call _third_party_main, and make the rest optional, they should be able to achieve the same things in a cleaner fashion.

[rubys]

@addaleax I agree. Giving people a better alternative then deprecating _third_party_main is the way to go.

[rubys]

See #23265

[Trott]

Since it's undocumented and is a compile-time/build option, I don't think the doc-deprecation and runtime-deprecations are applicable. I think we can go right to a semver-major removal, if that's what we wish to do. I guess that should wait for the alternative approach proposed by @rubys and @addaleax? Is that something that is in the "oh yeah, we're totally implementing it" phase? Or more of the "we kicked this idea around and kinda like it, but that's as far as we got" phase?

[bnoordhuis]

This issue saw its anniversary last Friday. Should it remain open?

[addaleax]

I would keep it open for a bit, I’ve started doing some more work towards a proper embedder API and maybe something will come out of that in terms of a better solution than this for embedders.