Prefix unusable with structured argv

[sanga]

Since #1067 got merged, prefix became unusable with structured argv. Which is documented behavior. However I'd like to continue to use hk's ruff definition but I can't now as that took structured argv into use and we need prefix. It occurred to me that hk should be able to support this by letting prefix be a list of strings rather than a string? (Or actually an enum of string and list of strings). Do you think this is reasonable? Would you accept an llm generated pr for this?

[jdx]

Yes, I consider this a regression. The restriction added in #1067 was intentional for string prefixes, but migrating Builtins.ruff to structured argv made previously valid builtin overrides fail based on an internal representation change.

#1175 restores the lost composition without weakening argv safety. Structured commands can now use an explicit list prefix:

["ruff"] = (Builtins.ruff) {
  prefix = List("mise", "x", "--")
}

String prefixes remain shell syntax for string commands; list prefixes are rendered and prepended as literal argv entries. The PR also covers builtin composition, group inheritance, literal filenames, invalid cross-mode combinations, and batching size accounting.

AI-assisted — Tool: Codex; model: unavailable/unavailable; version: unavailable.

[jdx]

sorry about this, I didn't think of pkl inheritance breaking changes, this is the best idea I had for resolving it

[sanga]

No worries - cheers for hk! I haven't tested your fix yet but that looks about like what I had in mind and what opus cooked up locally, so: all good.