The run and exec subcommands

[indirect]

We're planning to add commands named exec and run to rv, mostly modeled on how uv handles those commands, but with Bundler, npm, and Cargo in mind.

For more information about the history and functionality, read our recent blog post Why are exec and run so confusing?, and then come back with your thoughts.

How would you expect the rv exec and rv run commands to work? What options would you expect them to have?

[jaredcwhite]

@indirect I'd love to read the blog post, but I'm getting a 404!

[indirect]

@jaredcwhite Whoops, missed removing draft = true from that post. Fixed!

[kaspth]

Having read the blog post through, I'm noticing the use of global and local to help explain then need here. In that case, I'm wondering if the commands are rvg for rv global run and rvl for rv local run? We could still have rv exec and rv run for npm/uv compatibility.

Alternatively, I have been trying to drill into my mind that rvx means external, but then I can't find a "local" synonym for rvr.

Here's some proposed documentation that might be useful too:

• rv exec executes global commands external to your project, installing the gem if needed.
• rv run runs a local command/file/script internal to your project, installing and using gems from your Gemfile.

[indirect]

When I think about them, rather than "local" and "global", I imagine them as "run a file that's mine" and "exec a package from someone else".

For example, the official docs for the Cloudflare Workers CLI wrangler always tell you to run npx wrangler, even though it is the CLI for developing and managing your Cloudflare Workers project. In the opposite direction, I might want to uv run project-helper in many projects, but that file is either on my PATH or just an actual path directly to a local file on disk at ./project-helper.

[jaredcwhite]

I like the thrust of this, absolutely. Run for project-specific tasks, exec for global tasks. Makes perfect sense to me.

[deivid-rodriguez]

I discussed with @indirect and @kaspth briefly and yeah, this distinction makes sense.

I argue that most cases can be disambiguated easily depending on context, namely, whether rv is run in a project context or not, and when run from a project context, whether the target command is a project dependency or not.

So I feel the great majority of use cases don't need separate commands and can do the most expected thing without the user explicitly asking for it. In this sense, I really like how npm exec has this dual behavior and I believe it would work fine for Ruby projects.

That said, there are a lot of edge cases, so there will most likely be a need to be able to disambiguate what happens under the hood explicitly. Whether that's through separate commands, different flags for the same command, configuration, explicit edition of project files/manifests... I'm not sure.

[kaspth]

It sounds like we need a prototype command to map some of this out, and see whether or not these edge cases are easy enough to disambiguate in practice.