If you are using the generator you can create a completion command by running
```bash
cobra add completion
```
Update the help text show how to install the bash_completion Linux show here [Kubectl docs show mac options](https://kubernetes.io/docs/tasks/tools/install-kubectl/#enabling-shell-autocompletion)
Writing the shell script to stdout allows the most flexible use.
```go
// completionCmd represents the completion command
var completionCmd=&cobra.Command{
Use: "completion",
Short: "Generates bash completion scripts",
Long: `To load completion run
. <(bitbucket completion)
To configure your bash shell to load completions for each session add to your bashrc
# ~/.bashrc or ~/.profile
. <(bitbucket completion)
`,
Run: func(cmd *cobra.Command, args []string){
rootCmd.GenBashCompletion(os.Stdout);
},
}
```
**Note:** The cobra generator may include messages printed to stdout for example if the config file is loaded, this will break the auto complete script
`out.sh` will get you completions of subcommands and flags. Copy it to `/etc/bash_completion.d/` as described [here](https://debian-administration.org/article/316/An_introduction_to_bash_completion_part_1) and reset your terminal to use autocompletion. If you make additional annotations to your code, you can get even more intelligent and flexible behavior.
## Creating your own custom functions
Some more actual code that works in kubernetes:
```bash
const (
bash_completion_func=`__kubectl_parse_get()
{
local kubectl_output out
ifkubectl_output=$(kubectl get --no-headers "$1" 2>/dev/null);then
The `BashCompletionFunction` option is really only valid/useful on the root command. Doing the above will cause `__kubectl_custom_func()`(`__<command-use>_custom_func()`) to be called when the built in processor was unable to find a solution. In the case of kubernetes a valid command might look something like `kubectl get pod [mypod]`. If you type`kubectl get pod [tab][tab]` the `__kubectl_customc_func()` will run because the cobra.Command only understood "kubectl" and "get."`__kubectl_custom_func()` will see that the cobra.Command is "kubectl_get" and will thus call another helper `__kubectl_get_resource()`. `__kubectl_get_resource` will look at the 'nouns' collected. In our example the only noun will be `pod`. So it will call `__kubectl_parse_get pod`. `__kubectl_parse_get` will actually call out to kubernetes and get any pods. It will thenset`COMPREPLY` to valid pods!
## Have the completions code complete your 'nouns'
In the above example "pod" was assumed to already be typed. But if you want `kubectl get [tab][tab]` to show a list of valid "nouns" you have to set them. Simplified code from `kubectl get` looks like:
The aliases are not shown to the user on tab completion, but they are accepted as valid nouns by
the completion algorithm if entered manually, e.g. in:
```bash
# kubectl get rc [tab][tab]
backend frontend database
```
Note that without declaring `rc` as an alias, the completion algorithm would show the list of nouns
in this example again instead of the replication controllers.
## Mark flags as required
Most of the time completions will only show subcommands. But if a flag is required to make a subcommand work, you probably want it to show up when the user types [tab][tab]. Marking a flag as 'Required' is incredibly easy.
```go
cmd.MarkFlagRequired("pod")
cmd.MarkFlagRequired("container")
```
and you'll get something like
```bash
# kubectl exec [tab][tab][tab]
-c --container= -p --pod=
```
# Specify valid filename extensions for flags that take a filename
In this example we use --filename= and expect to get a json or yaml file as the argument. To make this easier we annotate the --filename flag with valid filename extensions.
