Fix reinstall from cache cause optionalDependencies being deleted from package-lock.json

[Brooooooklyn]

• Close [BUG] Platform-specific optional dependencies not being included in package-lock.json when reinstalling with node_modules present #4828

[nlf]

unfortunately this isn't quite the right approach here, this changed behavior means that we'll always try to install an optional dependency, even if we're asked specifically not to. what we'll have to do is try to write a test that demonstrates the incorrect behavior, then work backwards and figure out why it's happening.

i'm working on a related problem with optional peer dependencies that may address this too, when that work is done i'll circle back here and see if anything changes

[Brooooooklyn]

this changed behavior means that we'll always try to install an optional dependency, even if we're asked specifically not to.

That's the right behavior, right? The package-lock.json is designed to use cross-platform. Skip installation on one platform doesn't mean it needs to be skipped on the other platform.
Even if it happened on the same machine, I mean, delete package-lock.json and install again with node_modules existing, it should try to install the optionalDependencies again.
For example, if one of the optionalDependencies failed to install because lack of system components like gcc. Then the user configures the required system components and installs again, shouldn't try again in this scenario?

[nlf]

this changed behavior means that we'll always try to install an optional dependency, even if we're asked specifically not to.

That's the right behavior, right?

if we're explicitly asked to not install a package, then no this is not the right behavior. consider npm i --omit=optional; with your changes here we would install optional dependencies despite having been instructed not to. i think you're on the right path, but we need to figure out a more specific approach so that we allow an optional node to fail installation, correctly omit it when asked and still record it in the package-lock.json

[paulrutter]

unfortunately this isn't quite the right approach here, this changed behavior means that we'll always try to install an optional dependency, even if we're asked specifically not to. what we'll have to do is try to write a test that demonstrates the incorrect behavior, then work backwards and figure out why it's happening.

i'm working on a related problem with optional peer dependencies that may address this too, when that work is done i'll circle back here and see if anything changes

@nlf did you ever circle back to this one? See the related thread.