Reject empty type and field names in namedutils

[avalyset]

The validation loop in namedtuple and namedlist checks names with:

if not all(c.isalnum() or c == "_" for c in name):

all() returns True for an empty iterable, so an empty name passes this check. It then passes _iskeyword("") as well, and reaches name[0].isdigit() — which raises IndexError instead of the intended ValueError:

=== boltons.namedutils.namedtuple ===
  empty field name ["x", ""]     -> IndexError: string index out of range
  empty typename ""              -> IndexError: string index out of range
  control: invalid char ["x-y"]  -> ValueError: Type names and field names can only contain
                                    alphanumeric characters and underscores: "x-y"
  control: valid ["x", "y"]      -> accepted
=== boltons.namedutils.namedlist ===
  empty field name ["x", ""]     -> IndexError: string index out of range
  empty typename ""              -> IndexError: string index out of range

Both factories are affected, and both an empty field name and an empty typename trigger it.

Why this looks like an oversight rather than a decision

The rename branch seven lines above the validation loop already guards for this exact case:

if (not all(c.isalnum() or c == "_" for c in name)
    or _iskeyword(name)
    or not name                     # <-- here
    or name[0].isdigit()
    ...

So the guard exists in the same function; it is just missing from the loop that raises. CPythons collections.namedtuple, which this module is a variant of, rejects the same input properly:

=== CPython collections.namedtuple ===
  empty field name ["x", ""]     -> ValueError: Type names and field names must be valid identifiers: ""
  empty typename ""              -> ValueError: Type names and field names must be valid identifiers: ""

The change

Adds or not name to the validation loop in both functions, matching the form the rename branch already uses.

I also extended the message. The existing wording — "can only contain alphanumeric characters and underscores" — is vacuously true of the empty string: it states a rule the empty string does not actually break, so it would not explain the failure. The new wording is "must be non-empty and can only contain alphanumeric characters and underscores". No test asserted on the old text. Happy to drop that part and keep the one-word condition change if you would rather leave the message alone.

The rename=True branches are untouched.

Tests

Added to tests/test_namedutils.py, parametrized over both factories: empty field name, empty typename, an invalid character (still ValueError), and a valid pair (still accepted).

Verified the tests pin the behaviour rather than just passing — against an unmodified namedutils.py the four empty-name cases fail with the IndexError above:

FAILED tests/test_namedutils.py::test_empty_field_name_raises_value_error[namedtuple]
FAILED tests/test_namedutils.py::test_empty_field_name_raises_value_error[namedlist]
FAILED tests/test_namedutils.py::test_empty_type_name_raises_value_error[namedtuple]
FAILED tests/test_namedutils.py::test_empty_type_name_raises_value_error[namedlist]
4 failed, 7 passed

With the fix, and using the command from tox.ini:

pytest --doctest-modules boltons/namedutils.py tests/test_namedutils.py
13 passed

pytest tests
480 passed

Conflicts

#382 also touches boltons/namedutils.py, but its hunks are at lines 45, 120-122 and 279-281, while this change is at 163 and 322 — no overlap.