Fix: Use builtin cd for nvm_cd function

[glensc]

Use a more reliable built-in command invocation.

Fixes #3835

The previous fix 942e9ab (#1284) was not sufficient.

[ljharb]

either way, this would need a regression test.

[glensc]

Seems the builtin is not available everywhere:

+ builtin cd /workspace/.cache/src/node-v0.10.7/files
./install from source: 33: ./install from source: builtin: not found

• https://github.com/nvm-sh/nvm/actions/runs/25285934932/job/74130273083?pr=3836#step:8:14727

Alternative fixes:

1. use chdir, less commonly overwritten
2. use more detection, unlikely \cd and builtin both unavailable

[ljharb]

if there's no universal posix-compliant way to bypass functions, then there's not much that can be done - can you modify your cd function so it takes the same options as cd?

[glensc]

I also thought about unset cd, but can't do that, affects the effective shell for the user....

[glensc]

seems command cd works on some systems, even though it's supposedly executing an external command, it seems to execute a builtin

macOS 10.15.8

$ bash -c 'command cd /; pwd'
/

$ sh -c 'command cd /; pwd'
/

$ zsh -c 'command cd /; pwd'
/Users/glen

Debian 12

$ bash -c 'command cd /; pwd'
/

$ sh -c 'command cd /; pwd'
/

$ zsh -c 'command cd /; pwd'
bash: zsh: command not found

[glensc]

I propose this version as next:

nvm_cd() {
  if command -v builtin >/dev/null 2>&1; then
    builtin cd "$@"
  else
    \cd "$@"
  fi
}

WDYT?

[ljharb]

It'd be nice to define one of two functions conditionally rather than always have to check for builtin (penalizing the 99.99999% case where it's not needed), but in general that seems like it should work.

[glensc]

I did think about conditional function definition, but as the rest of the code doesn't do anything like this, I proposed to skip that part.

Very likely, defining a function conditionally causes some incompatibility issue, perhaps there's a shell that doesn't allow it, i.e., it considers such code as function re-definition.

[glensc]

Unclear why two jobs failed in CI. No error is printed, just bad exit code.

  set +e
  . $NVM_DIR/nvm.sh && nvm --version
  shell: /usr/bin/bash -e {0}
  env:
    ref: v0.40.0
Error: Process completed with exit code 3.

• https://github.com/nvm-sh/nvm/actions/runs/25291441762/job/74143741003?pr=3836#step:8:1
• https://github.com/nvm-sh/nvm/actions/runs/25291441762/job/74143741047?pr=3836#step:8:3

[ljharb]

Those are supposed to fail; you can ignore them.

[ljharb]

this will need some tests

[glensc]

Those are supposed to fail; you can ignore them.

They make CI red. That's intentional?

[glensc]

Seems it's this issue:

• Potential 0.40.0 regression on Github Actions runners #3405

Found from nvm-install-test.yml:48:

• nvm/.github/workflows/nvm-install-test.yml

  Line 48 in 3e4b0f0

  continue-on-error: ${{ matrix.ref == 'v0.40.0' }} # https://github.com/nvm-sh/nvm/issues/3405

And long-standing GitHub bug of not having a proper UI for that:

• Please support something like "allow-failure" for a given job actions/runner#2347

[glensc]

urchin tests / tests (sourcing, dash) (pull_request)
urchin tests / tests (sourcing, dash) (pull_request) Failing after 48s
urchin tests / tests (sourcing, sh) (pull_request)
urchin tests / tests (sourcing, sh) (pull_request) Failing after 51s

What to do with these?

✗ Sourcing nvm.sh should ignore a shadowed cd function when auto-detecting NVM_DIR
fake cd error was emitted while sourcing