@dbalatero @goldjacobe Sorry to be a pest and ping you directly, but any chance this could be reviewed?

This helps simply the graphite CLI install to a single step.

Thanks for these PRs, I'll integrate soon (we have some other inflight changes to the release process I want to finalize first)

this diff doesn't seem to work as expected.

@goldjacobe What issue are you seeing? No completions installed?

I had run into some issues during the last rebased, but thought I fixed everything.

I'm rebasing to test again.

@goldjacobe I just rebased and tested and it's installing the shell completions for me (I'm using zsh). Are you not seeing completions installed?

For zsh, the shell completions are installed to $(brew --prefix)/share/zsh/site-functions/_graphite. Depending on your setup, you may have to load the completions from Homebrew's completion directory: https://docs.brew.sh/Shell-Completion (not needed if you use brew installed shells)

Following up here since this stalled — I dug into why this diff may not have worked as expected, and opened #30 which supersedes it (full diagnosis and testing evidence there). Credit to @jalaziz for the original approach — the install-time generate_completions_from_executable call was the right idea. Two root causes, both quirks of yargs rather than this diff's structure:

1. yargs can only emit bash and zsh completion scripts, selected via the SHELL env var that generate_completions_from_executable sets per shell. Requesting :fish made yargs fall back to emitting a bash script into fish's completions directory, so the fish output was broken from the start.

2. The zsh script yargs emits only registers the completer — compdef runs at the end of the file, but nothing invokes the function (Generate zsh autoload completion function file yargs/yargs#2402). As an autoloaded fpath file, the first <Tab> of every shell session therefore completes nothing, and it works from the second attempt onward. I suspect this explains the impasse above: "this diff doesn't seem to work" and "it's installing the completions for me" were both accurate observations.

#30 limits generation to [:bash, :zsh], appends a _gt_yargs_completions "$@" invocation to the installed _gt so the first Tab works, and patches the formula template plus both rendered formulas so the release pipeline won't clobber the change on the next version bump. (It also chmods the binary before invoking it — the downloaded release artifact carries no exec bit during install and fails with EACCES otherwise.)