Background
I am designing a CLI for a container build tool I am making. It uses Gentoo’s Portage behind the scenes
Question
I want to give the user the ability to specify a custom package repository. The repository must have a name, URI and sync type.
custom_repo: {
uri: 'https://...',
name: 'custom',
sync_type: 'git',
}
How do I have the user represent this in the CLI? keep in mind, this is not the main input and is optional.
One way is to make this only provide-able via a config file using JSON or another structured data representation. But I want to see if theres a good way to do it in the CLI
What I am thinking of:
command --custom-repo uri='https://...',name=custom,sync_type=git --custom-repo ... [main input]
Is this the best way of doing this?
Just pass in the name of a json file as a CLI input (or default the name and act on it if present or use it if indicated [e.g. /U == use json.config]).
I will definitely make that an option, but I would still want it to be invokable via CLI only if the user chooses. It makes scripting easier sometimes.
How about a command-line flag to name an input file, but also process input as JSON, so someone can pipe it to your command or hand-write it if they’re crazy?
What language? This would be simple with Python’s argparse or Go’s pflag.
If the json payload is small with finite keys you can support separate args for those keys. If you really need arbitrary json what you have described is fairly reasonable as a shorthand, similar to AWS CLI shorthand.
Honestly passing optional/advanced args as json via CLI isn’t usually too bad since you can quote it with single quotes.
command --git-url https://... --alias myalias --svn-url http://... --alias mysvnalias
You may process it as a stack.
When reading within the program from stdin I recommend a state machine.
You could read json from standard input. Ex:
echo << EOF | command --read-stdin Some JSON EOF
For something like that i’d take a parameter like this (repeated as necessary):
--custom-repo=<name>=<synctype>+<url>
for example:
--custom-repo=custom=git+https://github.com/matcha/custom
command --custom-repo-uri https://foo.com --custom-repo-name repo_name --custom-repo-sync-type git
There could be multiple custom repos, so it would be difficult to know which uri goes with which repo name, and so on.
Can’t this all be deduced from the URI?
The .git suffix indicates git, the project name is the stem (project).
Nix does something like this with the protocol specifier: e.g.
git+https://...
I’m not sure what
name
means here exactly, but it might make sense to treat that separately, like git remotes:tool add [name] git+https://foo
That is assuming it’s hosted on github.
That is an example URI.
Ok, then I don’t understand at all. What happens if I host my git project on
https://myawesomeproject.dev/
? How can the application infer anything by this URL?Then replace “github.com” with “myawesomeproject.dev”. There’s more to the URI than just the hostname.
But you can’t assume that it follows the github format of
https://<domain>/<user>/<project>.git
. In my example, I meant that you would just use that url to clone it:git clone https://myawesomeproject.dev
One real-world example of this is ziglings.org (though it’s technically just a redirect).
That’s not a GitHub format; it’s a git format.
No, it isn’t. Git doesn’t care what the url is, as long as it uses a supported transport protocol.