Add --import <module> flag for pre-loading ESM modules

[Qix-]

Is your feature request related to a problem? Please describe.
-r can be used only on CommonJS modules, as documented here. Pre-loading an .mjs or an ESM module package is currently not possible.

Describe the solution you'd like
Either expansion of -r to support ESM modules (might break some people, not sure the scope of those changes) or the inclusion of a --module, -m <module> flag. Neither the long nor short arguments appear to exist in either Node or v8 according to the above docs (correct me if I'm wrong).

Describe alternatives you've considered
None, I perceive this to be a hole in the CLI option set.

[targos]

@nodejs/modules

[bmeck]

Would love and endorse this (likely under --import instead of --module). It has been tried a few times but async hook tests almost all blow up when you make the bootstrap async instead of the current sync bootstrap. I do not have the time for the fairly hefty testing refactor to make this work but the actual implementation effort is small.

[Qix-]

Agreed, --import is much better - good catch :)

[github-actions]

There has been no activity on this feature request for 5 months and it is unlikely to be implemented. It will be closed 6 months after the last non-automated comment.

For more information on how the project manages feature requests, please consult the feature request management document.

[ljharb]

bump

[Qix-]

@benjamingr cc #43382 (comment)

[benjamingr]

@nodejs/loaders if I was to find someone to work on this does this feature have reasonable consensus?

[aduh95]

The issue has been open for almost a year with no objections in response, and even some support for it; I'd say it's reasonable to assume it has consensus, it only need someone to do the work. Note that now that Node.js supports chaining loaders, using --loader already gives you more or less that functionality, maybe we don't need a new flag after all, just to adjust the --loader to account for this use-case.

[benjamingr]

@MoLow is this something you'd be interested in working on?

[MoLow]

@MoLow is this something you'd be interested in working on?

Id like to yes :)

[GeoffreyBooth]

I second what @aduh95 wrote. I don't think --loader is a good substitute for --import; there are reasons to load a package absolutely first, like polyfills or environment variable stuff, that loaders could do but are a poor fit for. As we're trying for parity with CommonJS, having a flag equivalent to --require makes sense.

[Flarna]

Not sure if this was discussed already somewhere but I haven't found it:

• --import will be processed before or after --loaders?
• Which priority/sequence is intended for --require vs --import?
• I guess --import will be allowed multiple times like --require. Correct?

[targos]

I don't think it was discussed already. Maybe it's worth a loaders-agenda label?

I guess --import will be allowed multiple times like --require. Correct?

That's what I'd expect.

[GeoffreyBooth]

Not sure if this was discussed already somewhere but I haven't found it:

• --import will be processed before or after --loaders?

After. Loaders should always be first, before --require too.

• Which priority/sequence is intended for --require vs --import?

If both are used together, you mean? I don’t know. I would think that all packages defined by either are loaded in the order defined, like --import one –require two –import three –require four would be interspersed; but I can see how that might be an issue depending on which loader is handling the entry point. I think ideally try to match the user’s input, the same as if all had been --require, and if that's not possible we'll have to investigate the best alternative (and document it well).

• I guess --import will be allowed multiple times like --require. Correct?

Yes.

[aduh95]

Why do we need two flags exactly? It seems to me that it would be much simpler to re-use the --loader flag here. Adding another flag raises more questions than it answers..

[GeoffreyBooth]

Why do we need two flags exactly? It seems to me that it would be much simpler to re-use the --loader flag here. Adding another flag raises more questions than it answers..

I'm not sure I follow. The loader flag expects a file with particular exports, like an exported resolve function. I think the use case for this feature is to load arbitrary packages that aren't loaders, like what --require is used for.

[aduh95]

The loader flag expects a file with particular exports, like an exported resolve function.

It would gladly accept a file with no exports though. The only use case for having a separate flag would be if one wanted to run a flag that exports whose name coincide with a loader, but not as a loader. That seems very unlikely, and quite easy to workaround anyway. --require is used to inject CJS loaders as well, so why not keep it to one flag in ESM world as well?

[GeoffreyBooth]

Because I don't think code in --loader is in the same scope as application code? Especially not after we move it off thread.

[ljharb]

That --require is also used for CJS loaders is because CJS has no official loader mechanism; a common use case for it is side-effecting polyfills, which if authored in ESM, would need --import.

[link89]

How about to start with implementing --experimental-import to make it work first. Currently there is no workaround to preload ESM.

[MoLow]

How about to start with implementing --experimental-import to make it work first. Currently there is no workaround to preload ESM.

👍🏻 I am working on this and hope to ship a initial PR this/next week

[bpstrngr]

I'm trying to upgrade from the old --loader to the new --import flag for module customization hooks now, but i'm encountering an issue with process.argv going missing when the --import flag is used (both in the module registration worker, and the main process), unlike without the flag being used:


( for full picture, this is where the hooks get called in each case: )

I have seen similar occur with the --watch flag not being present in execArgv when used, but that is less critical and i can imagine a reason for it.
Is there a reason for process.argv disappearing with --import modules?
Tested with Node 20.9 and 21.1, happens in both versions.

[GeoffreyBooth]

@bpstrngr please open a new issue.