`nvm which` + `.nvmrc` vs. `nvm which current`

[0cjs]

This is with nvm v0.40.3 on Debian 12.

The Problem

The documentation says that

...nvm which will use the version specified in the .nvmrc file if no version is supplied on the command line.

The help output says:

nvm which [current | ]
Display path to installed node version. Uses .nvmrc if available and version is omitted.

These are all correct, but the design is annoying because npm which fails if there is no .nvmrc found:

$ nvm which
No .nvmrc file found
[...full help output here deleted...]
$ echo exitcode=$?
exitcode=127

In this case you must use nvm which current, which always shows the version that would be selected if .nvmrc were not found. (A related problem: nvm current does the same.)

Solutions

Probably nvm which should be changed to use current if no .nvmrc is found. Without this, it's a bunch of extra work to reliably use nvm which when a .nvmrc file may or may not be present.

It's unclear to me what current is supposed to mean, and this appears to be undocumented. Is it the version of node in the path, if present, or the version that will be used by nvm exec etc? (These are not the same if a .nvmrc file is present.) I just noticed that nvm exec etc. accept an undocumented current argument for the version, where it appears to be the latter:

$ nvm current
v24.11.1
$ nvm which
Found '/home/cjs/tmp/nvm/.nvmrc' with version <v22>
/home/cjs/.nvm/versions/node/v22.20.0/bin/node
$ nvm which current
/home/cjs/.nvm/versions/node/v24.11.1/bin/node
$ nvm exec true
Found '/home/cjs/tmp/nvm/.nvmrc' with version <v22>
Running node v22.20.0 (npm v11.6.2)
$ nvm exec current true
Running node v24.11.1 (npm v11.6.2)

So probably the documentation should be updated to indicate what current is and exactly how it works, including documenting current in the help output for all subcommands that accept it.

[ljharb]

current is whatever's currently nvm used.

I'm happy to review PRs improving the docs.

As for changing the default behavior, I could be convinced to add a flag to opt in to that behavior, but I don't think it's a good idea for "no nvmrc found" to succeed and do something different.

[0cjs]

current is whatever's currently nvm used.

I don't think this is correct.

$ ls -l .nvmrc
ls: cannot access '.nvmrc': No such file or directory
2 $ nvm use current
Now using node v24.14.0 (npm v11.9.0)
$ nvm which
No .nvmrc file found
[rest of help here, and exit with code 127]

[ljharb]

Try nvm which current.

[0cjs]

Try nvm which current.

That doesn't work, because as you say:

current is whatever's currently nvm used"

But I want to know what nvm exec and nvm run will use.

$ nvm --version
0.40.3
$ nvm which current .
/home/cjs/.nvm/versions/node/v24.14.0/bin/node
$ nvm exec -- node --version
No .nvmrc file found
Running node v24.14.0 (npm v11.9.0)
v24.14.
$ echo v24.13.1 > .nvmrc
$ nvm which current
/home/cjs/.nvm/versions/node/v24.14.0/bin/node
$ nvm exec -- node --version
Found '/home/cjs/tmp/z/.nvmrc' with version <v24.13.1>
Running node v24.13.1 (npm v11.8.0)
v24.13.

So what I appear to want is just nvm which. Except we're back to where we
were before:

$ nvm which
Found '/home/cjs/tmp/z/.nvmrc' with version <v24.13.1>
/home/cjs/.nvm/versions/node/v24.13.1/bin/node
$ nvm exec -- node --version
Found '/home/cjs/tmp/z/.nvmrc' with version <v24.13.1>
Running node v24.13.1 (npm v11.8.0)
v24.13.1
$ rm .nvmrc
$ nvm which
No .nvmrc file found

Node Version Manager (v0.40.3)

[extensive help text deleted]
[exited with exit code 127]
127 $ nvm exec -- node --version
No .nvmrc file found
Running node v24.14.0 (npm v11.9.0)
v24.14.0

Can you just post the code that will, whether a .nvmrc is found or not,
give me the path to the node binary that will be used by nvm exec?

[ljharb]

That's not functionality we've ever had. nvm which isn't for "what will be used", it's for "what's being used now", just like POSIX which itself.

Can you elaborate on your use case?

[0cjs]

Except that nvm which does not give you "what's being used now" in various circumstances, as I have shown in many examples above where nvm which returns an error rather than showing the node that is run by nvm exec -- node --version in that exact same environment.

How is it good for anyone that, in a specific circumstance when nvm exec will successfully run a particular node binary, nvm which will sometimes tell you what that binary is, but sometimes will instead return an error instead of telling you what that binary is? (It's clear that nvm must know that path even when it won't tell you, or nvm exec wouldn't work. It's just refusing to tell you that path it knows.) What is the use case where you want this behaviour?

[ljharb]

True - nvm which requires a node version be supplied, through nvmrc or the command line. "What nvm exec will use" however should be the same thing - without a version supplied, it uses nvmrc; with one, it uses the command line.

What's confusing to me is that nvm exec -- node --version with no nvmrc file runs anything - that should be erroring out.

[0cjs]

What's confusing to me is that nvm exec -- node --version with no nvmrc file runs anything - that should be erroring out.

Oh, interesting! Ok, we have two quite different perceptions of what nvm behaviour should be, which makes me think that the documentation isn't quite clear (though it's been a long time since I've read it).

My understanding was that nvm use set what should be used by nvm exec in the absence of a .nvmrc file. But it sounds to me as if you intended the design to be, nvm exec works only with .nvm files and doesn't work if it can't find one.

And actually the latter probably makes more sense, as nvm use also (and perhaps primarily) sets the version of node added to the path, which of course can never be influenced by a .nvmrc because it's directly running a node binary that has no knowledge of or connection with nvm.

It seems to me that having a complete split between .nvmrc-using commands and "path"-related commands, such that each individual command uses/sets only one or the other, and the .nvmrc commands fail if they can't find a .nvmrc, would help avoid confusion about what one is supposed to be using, and probably catch some errors, such as:

1. In a project intended to use node from the path, I modify a script but unthinkingly use nvm exec instead of just using node directly; when I run the script that error is caught because nvm exec complains that there is no .nvmrc.

2. In a project intended to use only the version in the .nvmrc, I unthinkingly add a node ... command to a script, but this is caught (in my case anyway) because by default I don't have a node in my path: I execute nvm use only when I don't intend to use .nvmrc files.

[ljharb]

nvm exec has always been intended to be entirely unrelated to "the currently used node", yes. It does not only work with .nvmrc files, because it takes a version argument directly - .nvmrc files only matter when the version argument is omitted (just like with nvm use and nvm install).

[0cjs]

Ok. So the help file isn't clear about that; it says "Uses .nvmrc if available and version is omitted" where the "if available" bit seems strongly into indicate that it's not required if the version is omitted. And the behaviour is wrong, too:

$ nvm exec -- node --version
No .nvmrc file found
Running node v24.14.0 (npm v11.9.0)
v24.14.0

Various other commands suffer from similar or related issues:

• nvm run: As nvm exec.

• nvm which:

  • The help output needs the same clarification as above.
  • Rather than giving you no error message and printing the help, implyingy you'd supplied an invalid command line argument, nvm which with no .nvmrc should print an error. (The same error that nvm exec should generate in the same situation.)

• nvm uninstall: As per nvm which.

• nvm install: The one command it seems that correctly prints a No .nvmrc file found error message, it still then prints another hundred lines of help message. If you don't scroll back and carefully examine the first line, it looks as if you've made an error in the command line arguments. Just the message should be printed. And has the same issues as the rest in its help entry.

The error message would also be better to be not just No .nvmrc file found but instead No .nvmrc file found and no version number given so that it's more immediately clear this split between the "use only .nvmrc version" and "use only command-line argument version" works.

And, of course, any doc updates for the website etc. that are relevant to help clarify all this.

[ljharb]

it's pretty common to print help output on any failure in command line tools, and if you don't have an nvmrc file and you fail to provide a version, you have made an error in the command line arguments.

I'd definitely be happy to improve the error message, and happy to review whatever documentation updates you have in mind.

As far as behavior changes, nvm exec -- node --version with no nvmrc file should fail - it's a bug that it doesn't. Same with nvm run.

[0cjs]

I cannot recall any other program that prints help when the command-line syntax is correct. I'd be very interested to see any examples you have of that.

Your argument that your command line argument syntax depends on the environment is interesting, but I don't buy it. Normally people don't think that "if I look at this statement in a language, I cannot tell from the statement alone if the syntax is correct; I also need to know what files happen to exist or not exist in the filesystem." What you're saying is essentially that if someone asks, "is nvm exec -- node version correct syntax for an nvm command line?" the answer is, "You can't tell without knowing what files are in $CWD and eveyr directory above it." That's not what people normally think of as "syntax." (Certainly cat does not work like this: cat .nvmrc is correct syntax regardless of whether .nvmrc exists, and the error you get when it doesn't exist is not a syntax error in the arguments to cat, and does not give you a help message.)

And the implication for people doing scripting is, the only way you can call nvm exec correctly is first to entirely replicate nvm's way of finding a .nvmrc file and try to find it before you call nvm so that you can tell if the nvm ... command you want to run has correct syntax or not.

At any rate, since we agree that nvm is broken as far as not failing nvm exec -- node --version when there's no .nvmrc file, can you fix that?

[ljharb]

Certainly that can be fixed.

The implication for people doing scripting is always to 100% know the entirety of the filesystem, including the presence or absence (or possibility of either) of an .nvmrc file. You can't ever successfully write a script without that knowledge.

As for help text, try typing git - many programs (but certainly not all) work this way. It's fair that perhaps we should have a shorter help output for the error cases.

[0cjs]

The implication for people doing scripting is always to 100% know the entirety of the filesystem, including the presence or absence (or possibility of either) of an .nvmrc file. You can't ever successfully write a script without that knowledge.

This is simply not correct. I write scripts all the time without this knowledge, especially when the user is supplying parameters to the script. For example, I do not use

[[ -r $file ]] || {
    echo 1>&2 "Error: cannot read $1"
    exit 1
}
cat "$file"

But instead just use cat "$file"; cat will fail with an appropriate error message if the user supplies a filename that doesn't exist. (Actually a better message, because cat will give more precise messages based on the error, distinguishing between 'No such file or directory,' 'Permission denied,' etc.) And then it's my choice to use set -e to get an automatic exit there, use cat "$file" || ... if I wish to do something else after displaying the error, whatever.

The same is true for scripts using nvm. I do not control the directory in which my script is run: the user decides that. To ask scripters to do the same search for a .nvmrc that nvm does is a waste of their time and, only if you're lucky, generates duplicate code. If you're unlucky the scripter doesn't implement it in the exact same way and will either find .nvmrc files that nvm itself doesn't find or not find .nvmrc files that nvm does find.

As for help text, try typing git - many programs (but certainly not all) work this way.

git never gives help output if the command line is valid syntax. (git with no parameters is never valid syntax.) Try typing git status when you're not in a repo; as with my cat example above, it prints fatal: not a git repository (or any of the parent directories): .git, rather than relying on you to figure out if your CWD is within a Git repo or not.

[0cjs]

Perhaps I can put this in another way. If I just agree to disagree with you that it's required to know about the contents of the filesystem to determine if nvm which, and go with what you say I need to do, what code do I need to put in my script either to run nvm which without an error or know that I can't do that and give the user an error message?