Node interpreter setup documentation

[FossPrime]

• Operating system and version: Linux 5

• How did you install nvm?
  install script in readme

• What steps did you perform?
  nvm install 12

• What happened?
  Everything worked wonderfully.

But, I have some node scripts that depend on /usr/local/bin/node existing. I can execute them just like any shell script, they have #!/usr/local/bin/node as the interpreter, and I run them with ./my-script.js myparams from the terminal.

• What did you expect to happen?
  I expected some documentation about where to symlink /usr/local/bin/node to. Instead I get node: bad interpreter: No such file or directory errors when executing those scripts from the shell.

it seems like this is not an expected use case... I can't find any symlinks to the default node anywhere. it would be nice to have a default node symlink in /alias/node. Same for npx and npm. That way we could symlink it to a standard location, or be asked if we want to do that at installation the get go.

• Is there anything in any of your profile files (.bashrc, .bash_profile, .zshrc, etc) that modifies the PATH?
  no.

[ljharb]

If so, those scripts are broken. They should be using a shebang of #!/usr/bin/env node instead, so that the node path isn't hardcoded in those scripts.

[FossPrime]

If so, those scripts are broken. They should be using a shebang of #!/usr/bin/env node instead, so that the node path isn't hardcoded in those scripts.

Yes but that depends on the environment vars, which makes it hard to run in Cron, LaunchDeamons and systemd, where the environment is non existent or rudimentary. I remember having only /bin and /sbin in the path with the default cron shell

[ljharb]

The tradeoff there is that you need to hardcode your node path, which means you have to change it for every system and/or every node version.

[FossPrime]

Redhat distros are very consistent about it's location... It's just always outdated. I don't deal with Debian enough to tell where apt-get likes to out it, but I imagine it's near /usr/local/bin/node

The main issue here is there is no easy way to point /usr/local/bin/node to whatever the NVM default is

[ljharb]

The problem is that there isn’t an nvm default - there’s only per-shell PATH modifications. So if PATH isn’t available in a context, nvm isn’t really applicable or useful.

[FossPrime]

I got my inid.d script to work with by loading the bash profile up top... . /etc/profile. Not efficient, but works. I still think that we should add a nvm/current symbolically linked directory on POSIX. That would be more efficient and reliable.

[ljharb]

There is no “current” outside of the current shell session.

[FossPrime]

Went with this instead, due to serious slowdowns when network not available. I think I bricked Ubuntu on a previous machine with NVM because I ran it too early in profile.d.

export NVM_DIR="/opt/nvm"
if [ ! -f /usr/local/bin/node ]; then
        [ -s "$NVM_DIR/nvm.sh" ] && \. "$NVM_DIR/nvm.sh"  # This loads nvm
        [ -s "$NVM_DIR/bash_completion" ] && \. "$NVM_DIR/bash_completion"  # This loads nvm bash_completion
 
        cd /usr/local/bin
        ln -s `which node`
        ln -s `which npm`
        ln -s `which npx`
        cd - &> /dev/null
fi

[ljharb]

You definitely do not want to set NODE_PATH; global modules should never be requireable.

[FossPrime]

You're right... I was using it to run node -r esm for scripts with #!/usr/bin/env esm

I updated the esm script to

#!/bin/bash

if [ -x "$(command -v npm)" ]; then
  env NODE_PATH=`npm root -g` node -r esm $*
fi

could be useful for typescript bash replacement scripts too.

These methods turned out to be ... WAY faster and more resilient to network issues. Specially when using in profile.d scripts and sudo su'ing around a lot.