Nvm: Cannot call `nvm exec` in bash script with `set -e`

[errm]

If I call nvm exec node foobar.js in a bash script with set -e everything exits early, this is not great for being able to use nvm in an automated build for example . . .

[ljharb]

Using inline nonzero return codes is a common practice, and set -e is an intrusive option that isn't compatible with nvm. Rather than using the catchall of -e, I'd recommend explicit error checking and exits where errors are possible.

An alternative is to source nvm.sh prior to setting the -e option, and use nvm use rather than nvm exec.

[joscha]

For anyone using nvm in a CI environment, where non-zero exit codes are usually used to indicate a problem, I have this little snippet:

export NODE_VERSION="5.9.0"
export NPM_VERSION="3.7.3"
export NVM_DIR="`pwd`/.nvm" # truly local nvm install not needing sudo
mkdir -p $NVM_DIR

curl -o- https://raw.githubusercontent.com/creationix/nvm/v0.31.0/install.sh | bash
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"  # This loads nvm

nvm install $NODE_VERSION
nvm use --silent $NODE_VERSION

set -e # make the script fail if anything goes wrong from here

export npm_install=$NPM_VERSION
curl -L https://npmjs.org/install.sh | sh

npm install your-module
node `npm bin`/your-module

[ljharb]

@joscha is this still a problem if you move the set -e above the sourcing of nvm.sh?

While I do think that set -e is a silly option, and failures should explicitly be checked for, I don't want to make life hard for nvm users who use set -e. If there's still issues with this, and it's possible to solve them in a practical way, I'd like to (which is why the issue is still open)

[joscha]

Yes, it doesnt't work, because most nvm xxx tasks return with 127 instead of 0. Especially for nvm exec $NODE_VERSION npm test this is dangerous.

[ljharb]

Hmm, 127 should only be the exit code when an incorrect command was entered, and the help output goes to stderr. Do you have a specific line where it exits?

[joscha]

Unfortunately the build logs where I had problems with that (hence why I found this ticket) are gone, so I am not sure which of the commands failed exactly, but I remember that an order like this:

set -e
export NVM_DIR="`pwd`/.nvm"
mkdir -p $NVM_DIR

curl -o- https://raw.githubusercontent.com/creationix/nvm/v0.31.0/install.sh | bash
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"

nvm install 5.9.0
nvm use --silent 5.9.0

export npm_install=3.7.3
curl -L https://npmjs.org/install.sh | sh

nvm exec --silent 5.9.0 npm install your-module
nvm exec --silent 5.9.0 node index.js

did not work.

[ljharb]

Executing this script works as expected with and without a default, with and without an .nvmrc:

#!/usr/bin/env bash
set -e

. ~/.nvm/nvm.sh
echo $(nvm current)

I'm going to close this for now. Please reopen if the issue occurs again and if anyone can be more specific about which command fails.

[joscha]

I looked into this again and I think in some of our builds the $NODE_VERSION was empty and the command was this:

nvm exec --silent $NODE_VERSION npm -v

so I believe that it actually only fails when the version is omitted. The OP of this ticket also uses it without a version, so that would make sense.

[ljharb]

@joscha when the version is omitted and no .nvmrc is present, it does fail (and should) - is that the situation for you?

[joscha]

Yes, it fails with exit code 3

[ljharb]

Great - that's intentional then :-)

[kncs]

I don't know if it could help someone but I just solve a similar problem using set -e and nvm into a bash script. The problem was a missing line break into the .nvmrc file.

How to reproduce it :
1) Create a bash file

#!/bin/bash
set -e
source ~/.nvm/nvm.sh
nvm --version 
nvm use 

2) Remove line break into the .nvmrc

3) Launch script and observe.

[ljharb]

@kncs hm, you're saying that if the nvmrc file omits a trailing line break (something every file in unix should always have), it breaks? That would be a concrete bug I could fix.

[kncs]

My bad, of course I wouldn't ask you to fix it, my comment was only to help people that could have similar problems

[ljharb]

@kncs i'll try to reproduce (i want to fix it!) but your confirmation would help :-)

[kncs]

Yes it seems that it breaks something. I join a zip containing a broken nvmrc and the bash script to help you reproduce it.
test.zip

[ljharb]

@kncs thanks - i was able to find the line causing the error. Definitely every file ever should have a trailing newline, but, I'll figure out a way to make this not break.

[hedefalk]

I'm having this problem using drone ci https://github.com/drone/dronehttps://github.com/drone/drone

Drone automatically creates a script with set -e on top of my "user commands" so can't really do anything about it other than || trueon all nvm commands.

[ljharb]

@hedefalk if you can narrow down which commands are failing, and ideally using set -x which place they are failing at, please open a new issue for each and I'll tackle them one at a time.

[andyearnshaw]

It appears nvm install <version> fails when set -e is used also. Here's an example script I'm using:

#!/bin/sh
set -e   
[ -z "${DEBUG}" ] || set -x

green="$(tput setaf 2)"
reset="$(tput sgr0)"

export NVM_DIR="$HOME/.nvm"

NODE_VERSION="7.8"
BASEDIR="$(dirname "$0")/.."
cd "$(dirname "$0")/.."

SHORT_PIPELINE_VERSION=${PIPELINE_VERSION#*/}

if [ ! -d $NVM_DIR ]; then
    echo $green "\n==> Installing Node Version Manager" $reset
    (
        git clone -n https://github.com/creationix/nvm.git "$NVM_DIR"
        cd "$NVM_DIR"
        git checkout -q `git describe --abbrev=0 --tags --match "v[0-9]*" origin`
    )
fi

. $NVM_DIR/nvm.sh

echo $green "\n==> Installing node" $reset

nvm install $NODE_VERSION

The offending non-zero returning function is nvm_version_greater, called from nvm_version_path. Should I open a new ticket for this?

[ljharb]

@andyearnshaw yes, please; a new issue would be great.

[matheusfaustino]

Executing this script works as expected with and without a default, with and without an .nvmrc:

#!/usr/bin/env bash
set -e

. ~/.nvm/nvm.sh
echo $(nvm current)

I'm going to close this for now. Please reopen if the issue occurs again and if anyone can be more specific about which command fails.

I just wanna point out that this solved my problem with Bitbucket Pipeline using atlassian/ssh-run. I added this part on the script because only when it was running by the Pipeline it couldn't find node installation. Thanks ^^