validate-family.py member-duplicate runs only with the members checked and compares project: exactly, so --no-members passes a family.yaml that names one row twice, and rows IRRS and irrs with distinct ids pass, though members/IRRS and members/irrs are one directory on macOS and Windows #331

Lane: openreposhape-3 (openRepoShape-3)

What happens. member-duplicate is raised in _member_project_name, which only _check_member calls, and _check_members returns before that loop under --no-members. _check_member_ids_unique (#221) runs before the return, but it compares project in projects exactly as well. Measured on be2790e with a family holder from family.py init, a hand-written family.yaml of two rows, and --no-members:

two rows `project: IRRS` (repositories InkRouter/IRRS twice)   ->  family ok: Fam (fam), 2 member(s)    exit 0
rows `IRRS` and `irrs` with different ids                       ->  family ok: Fam (fam), 2 member(s)    exit 0

With members checked the comparison is if project in seen, exact (read from the code; not run with two mounted members).

Where. templates/family-root/scripts/validate-family.py:_member_project_name (:215-228), _check_members (the --no-members return), _check_member_ids_unique (:525).

Why it matters. --no-members is the run a private family's CI falls back to when its members cannot be checked out, so a duplicate row passes the one gate that runs. family.py add now refuses a case variant, so only a hand edit makes such rows. The file is one of the copies a holder pins, so a change is a shape update. Direction: move member-duplicate ahead of the --no-members return, like member-duplicate-id, and compare project: case-folded.

Source. PR #221, ## Left out, second and third bullets.