Our team has built a small CLI used for maintenance. The package.json specifies a path for with the bin property, and everything works great; "bin": { "eddy": "./dist/src/cli/entry.js"}
Autocompletion is achived by using yargs@17.0.1. However we recently converted the project to use es6 modules, because of a migration to Sveltekit, i.e. the package.json now contains type: module. Because of this, the CLI now only works if we run with:
what works
node --experimental-specifier-resolution=node ./dist/src/cli/entry.js help
However, if we run this without the flag, we get an error "module not found":
Error [ERR_MODULE_NOT_FOUND]: Cannot find module...
So the question is
Can we somehow "always" add the experimental-specifier-resolution=node to the CLI - so we can continue to use the shorthand eddy, and utilize auto completion?
There are two probable solutions here.
Your entry.js file should start with a shebang like #!/usr/bin/env node. You cannot specify the flag directly here, however, if you could provide the absolute path to node directly in the shebang, you can specify the flag.
Assuming you have node installed in /usr/bin/node, you can write the shebang in entry.js like:
#!/usr/bin/node --experimental-specifier-resolution=node
(Use which node to find the absolute path)
However, this is not a very portable solution. You cannot always assume everyone has node installed in the same path. Also some may use nvm to manage versions and can have multiple version in different path. This is the reason why we use /usr/bin/env to find the required node installation in the first place. This leads to the second solution.
You can create a shell script that would intern call the cli entry point with the required flags. This shell script can be specified in the package.json bin section.
The shell script (entry.sh) should look like:
#!/usr/bin/env bash
/usr/bin/env node --experimental-specifier-resolution=node ./entry.js "$@"
Then, in your package.json, replace bin with:
"bin": { "eddy": "./dist/src/cli/entry.sh"}
So when you run eddy, it will run the entry.js using node with the required flag. The "$@" in the command will be replaced by any arguments that you pass to eddy.
So eddy help will translate to /usr/bin/env node --experimental-specifier-resolution=node ./entry.js help
Just add a script to your package.json:
Assuming index.js is your entry point and package.json is in the same directory
{
"scripts": {
"start": "node --experimental-specifier-resolution=node index.js"
}
}
Then you can just run on your console:
npm start