[lntutor]

Summary

• add native rtk bun and rtk bunx commands that preserve the underlying Bun executable
• rewrite bun test, option-prefixed bun commands, bun run, Bun script shortcuts, and bunx invocations
• remove Bun run/test banners and only the immediate echoed script command while preserving program output
• classify Bun usage under JavaScript telemetry and document the new commands

Fixes #1316

Verification

• cargo fmt --all -- --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test --all-features (2,505 passed, 8 ignored)
• cargo build --release --all-features
• live execution: rtk bun -e console.log command completed successfully
• exact rewrite verified for an environment-prefixed bun test command

[CLAassistant]

All committers have signed the CLA.

[KuSh]

Thanks for this, and sorry it sat — closing as superseded by #988, which covers bun plus deno and has been through five rounds of review with fixtures captured from real bun 1.2.20 / deno 2.9.6.

Being concrete about what your PR does that #988 does not, because it is a real gap and I would like you to pick it up:

#988 only rewrites an allowlist of subcommands — ^bun\s+(install|add|remove|test|build|run|pm ls|pm|x)\b. Yours matches ^bun(?:\s|$), so these route through rtk for you and run raw under #988:

bun --env-file=.env.test test tests/unit    # option-prefixed
bun dev                                     # script shortcut
bun drizzle-kit generate                    # package binary
bun -e 'console.log(1)'

That is a coverage gap, not a correctness one — those commands still work under #988, they just get no filtering.

Why #988 can't simply take the wider pattern as-is. run_filtered buffers the whole stream until the child exits (FilterMode::CaptureOnly). ^bun(?:\s|$) routes bun dev — a dev server — into it, so it prints nothing at all until Ctrl-C and then loses the buffer. #988 spent a full review round on exactly this class and now sends watch-mode runs through unfiltered on every entry point. Widening the pattern needs the same guard for long-running scripts, which is the harder half of the problem.

Two smaller things worth carrying over when you do: filter_bun_output drops every blank line from arbitrary program output, which reflows anything a script prints deliberately; and the banner check line.contains(|c: char| c.is_ascii_digit()) will match a script whose own output starts with bun run v — worth anchoring on the version shape.

The follow-up I'd welcome after #988 lands: widen the rewrite to option-prefixed and script-shortcut forms, reusing is_watch_mode (and something equivalent for dev servers) so nothing long-running ends up captured. That is genuinely missing today and it is the valuable part of this PR.

Also relevant: #1316, which you referenced, stays open until that follow-up rather than being closed by #988.