Cobra's bash completion V2 has the following benefits:
- aligned with the completion logic for the other shells
- provides completion descriptions as for the other shells
- uses a 300-line shell script (versus 4K lines for V1)
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
Cobra provides powershell auto-completion out of the box. This commit
teaches helm how to use it.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
The "helm completion zsh" and "helm completion fish" commands accept the
"--no-descriptions" flag, therefore we should not disable the addition
of "[flags]" to the usage line when printing help.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
With the move to Cobra 1.1.1, all commands that have sub-commands
have file completion automatically disabled, so helm need not do it
itself.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
With Cobra 1.0, it is now possible to control when file completion
should or should not be done. For example:
helm list <TAB>
should not trigger file completion since 'helm list' does not accept
any arguments.
This commit disables file completion when appropriate and adds tests to
verify that file completion is properly disabled.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
Making each shell a subcommand of the 'completion' has multiple
advantages:
- simplifies the code,
- allows to have different flags for each shell,
for example, a future `--no-descriptions` flag for fish only,
- allows to tailor the help text per shell.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
When doing zsh completion, the -n flag would not be handled properly.
Doing
helm -n<TAB>
would not add the space after the -n.
This was caused by the fact that the -n flag was being swallowed by
the echo command.
To fix this, we use printf instead.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
If a user renames the helm binary, the shell needs to be told that this
new name corresponds to the helm completion function. For example if
the user decides to rename the helm v3 binary to 'helm3', then the
following command extra command would need be run by the user:
complete -o default -F __start_helm helm3.
This commit automates this by adding this extra command to the generated
shell completion script by looking at what binary was used to call the
helm completion command. For example, if the user renames the binary
to helm3 and then calls
helm3 completion bash
the completion script will include a hook for the binary 'helm3' to
the completion script.
A binary rename is one of the two options recommended when users need to
run both helm2 and helm3.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
This allows that any addition to the bash completion logic be
automatically added to the zsh logic.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
This is a bug I ran into when working on Helm completion.
I was surprised that it didn't happen when I was using
kubectl, so I investigated and found a PR that fixed this
bug in kubectl:
https://github.com/kubernetes/kubernetes/pull/48553
I duplicated the code in this commit which:
Removes __helm_declare, which is safe to do since
`declare -F` is already replaced to `whence -w` by
__helm_convert_bash_to_zsh().
The problem was that calling "declare" from inside a function
scopes the declaration to that function only. So "declare"
should not be called through __helm_declare() but instead
directly.
To reproduce:
1- setup helm completion in zsh
2- helm --kubeconfig=$HOME/.kube/config statu<TAB>
you will get the error:
__helm_handle_flag:27: bad math expression: operand expected at end of string
Co-authored-by: Kazuki Suda <kazuki.suda@gmail.com>
Signed-off-by: Marc Khouzam <marc.khouzam@ville.montreal.qc.ca>
Cobra provides some out-of-the-box debugging for bash completion.
To use it, one must set the variable BASH_COMP_DEBUG_FILE to
some file where the debug output will be written. Many of the
debug printouts indicate the current method name; they do so
by using bash's ${FUNCNAME[0]} variable. This variable is
different in zsh. To obtain the current method name in zsh
we must use ${funcstack[0]}.
This commit adds the proper sed modification to convert from
bash to zsh.
Signed-off-by: Marc Khouzam <marc.khouzam@montreal.ca>
Updated the 'helm completion' command to take in the shell
as a parameter. Currently acceptable options are 'bash' and
'zsh'. Also fixed the completions to work with zsh in a manner
similar to what kubectl does.
Also updated the docs to reflect this change.
Closes#2201
Marc Khouzam
Alexandre Garnier
Scaat Feng
knrt10
Matt Butcher
Matthew Fisher
Matt Farina
Marc Khouzam
Anton
Marc Khouzam
Peter Stalman
Bartel Sielski
Adam Reese
Sushil Kumar
Michelle Noorali
Nikhil Manchanda
rodcloutier