Allow --rm to remove a virtualenv by its dependencies

[gilgamezh]

What

--rm now accepts either a UUID or nothing. When given as a bare flag it removes the virtualenv matching the indicated dependencies, so this:

fades --rm $(fades --where -d django | awk -F "/" '{print $NF}')

becomes simply:

fades --rm -d django

How

• --rm uses nargs='?' with a REMOVE_BY_DEPS sentinel as const, distinguishing "not passed" / "passed with UUID" / "passed bare".
• The removal block was moved below the dependency/interpreter/options setup. With the sentinel it resolves the venv via get_venv(indicated_deps, interpreter, options=options) — the same matching used to run a venv — otherwise it uses the existing UUID lookup. Both share the destroy + warning logic.
• Backward compatible: fades --rm <uuid> behaves exactly as before.

Tests

There were no tests for the --rm flow in go(). Added RemoveVenvTestCase covering removal by UUID (found / not found / invalid env_path) and by dependencies (found / not found).

• Full suite: 296 passed
• flake8: clean

Docs

Updated man/fades.1 and README.rst (usage synopsis + command examples).

🤖 Generated with Claude Code

[facundobatista]

Awesome, thanks! There's a couple of very small details, commented inline...

[gilgamezh]

One edge case with --rm taking an optional value: when the dependencies come from a script, the script name gets consumed as the UUID.

With nargs='?', argparse grabs the next non-option token as the value, so (checked on 3.13):

command	remove	child_program
fades --rm -d foo	sentinel ✅	None
fades --rm script.py	'script.py' ❌	None
fades --rm script.py arg	'script.py' ❌	'arg'
fades --rm -- script.py	sentinel ✅	'script.py'
fades script.py --rm	None ❌	'script.py' (--rm goes to child_options)

So the natural "remove the venv for this script" (fades --rm script.py) ends up warning No virtualenv found with uuid: script.py. and exits 0, which is easy to miss. The only form that works today is --rm -- script.py.

Proposal: venv UUIDs are always str(uuid4()) (envbuilder.py), so a value that doesn't parse as a UUID can't be one. After parse_args(), reinterpret it as the child program:

if isinstance(args.remove, str):
    try:
        uuid.UUID(args.remove)
    except ValueError:
        # not a UUID: it's the child program, remove by its dependencies
        if args.child_program is not None:
            args.child_options.insert(0, args.child_program)
        args.child_program = args.remove
        args.remove = REMOVE_BY_DEPS

That makes fades --rm script.py do what it reads like, keeps fades --rm <uuid> unchanged, and needs no new flag. Plus a couple of tests for the reinterpretation.

If that's too magic, the conservative alternative is to error out on a non-UUID value, suggesting fades --rm -- script.py, instead of the current warning + exit 0.

Happy to push either one; I lean towards the first.

---

Generated by Claude Code

[gilgamezh]

Pushed 76b3571 to fix the --rm script.py edge case from the comment above, without isinstance():

• Bare --rm now uses const='' instead of an object() sentinel, so args.remove is always None or a str.
• Right after parsing, if args.remove is not None and it isn't a valid UUID, it's the child program: it's moved to child_program (any previous child_program goes to the front of child_options), and the venv is removed by its dependencies.
• In the removal block, a non-empty value means remove by UUID, and an empty one means remove by dependencies.

command	result
fades --rm -d foo	by deps
fades --rm <uuid>	by UUID
fades --rm script.py	by the script's deps (was: "No virtualenv found with uuid: script.py")
fades --rm script.py arg --opt	by the script's deps, arg --opt kept as child options (was: "child program not found")
fades --rm -- script.py	by the script's deps

Tests: the existing UUID tests now use a real UUID, and I added tests for --rm script.py, --rm script.py arg1 --opt and --rm -- script.py. The two script ones failed before the fix. Full suite: 299 passed, flake8 clean.

---

Generated by Claude Code