templates/family-root/scripts/siblings.py: the layout table pads each member's name by code points, not by the columns a terminal gives it, so a wide or combining name's state column stands out of line (the family-root twin of #1034) #1216

Goal: the layout table that make siblings prints pads each name by the columns a terminal gives it, so a wide or combining name's state column starts where every other row's does, and an ASCII name prints byte for byte as it does today.

Filed by lane openRepoShape-3, from RV1178's independent review of #1178 (#1034) at fa56168a (#1178 issuecomment-6351653129), its follow-up for templates/family-root/scripts/siblings.py:729; the delta review at c4beba19 (#1178 issuecomment-6353960809) is READY and takes the follow-up no further. It did not hold the landing. #1178 is open at c4beba19, and its diff against main ef619c74 leaves siblings.py, scripts/validate-repository-naming.py and scripts/repo_shape.py alone, so the defect is on main now. Line numbers are main's at ef619c74.

The row pads a name by code points

State. Sibling.row (templates/family-root/scripts/siblings.py:725) returns f" {kind:<9} {spelled(self.project):<18} {self.state:<22} " and then spelled(self.branch) (:729-730). A format width counts code points. row is every row of the layout: the holder, sibling and pinned rows (:1623, :1639, :1641) and the dispatch's rows (:1289). Each table's header is a literal (:1287 and :1635, {'what':<9} {'name':<18} {'state':<22} branch). spelled (scripts/repo_shape.py:3645) escapes controls, surrogates, line and paragraph separators and bidirectional controls and, its docstring says, prints every other character as it is, so a wide character or a combining mark reaches the pad whole. self.project is a family manifest's project: text.

What happens. Arithmetic from the code and the standard library's unicodedata, not a run of the tool: RV1178 measured adopt-project.py's table and listed this one as a follow-up without a measurement. Under the rule #1178 uses (no column for categories Mn, Me and Cf, two for East Asian width W or F, one otherwise):

• a name of the three characters U+65E5 U+672C U+8A9E is 3 code points, so the pad adds 15 spaces and the field is 21 columns wide, 3 more than an ASCII name's 18: its state starts 3 columns to the right of every other row's;
• e U+0301 cole is 6 code points and 5 columns, so its field is 17 columns wide and its state starts 1 column to the left.

On a stdout that cannot hold a character, escape_unencodable_stdout (called at :1743) makes the handler backslashreplace, so U+65E5 reaches the terminal as \u65e5: six columns for one code point that the pad counted once (by reading that function's docstring in scripts/repo_shape.py; not run).

Direction. Pad by display columns, as #1178 does for adopt-project.py's plan table (_columns at adopt-project.py:6437 and _padded at :6460 at c4beba19, with ZERO_COLUMN_CATEGORIES, TWO_COLUMN_WIDTHS and the two Hangul jamo sets), and measure the text as stdout will write it. An ASCII name keeps its width exactly; a name wider than the field is left as it is, with no truncation. No command and no flag.

• Where the rule lives is the writer's choice, and the PR says which. This file is a pinned copy every holder carries. A holder that kept an older repo_shape.py (update-shape.py apply --accept-local) must still run it, and an import of a name that copy lacks would stop make siblings, park and resume (the comment at :1520-1524). So the simple form is the rule written in this file with unicodedata, which it can import as os and re are imported.

• The change moves this file's bytes, so holders see an upstream-changed row for it, as bootstrap.py, and the family root's validate-family.py, bootstrap.py and siblings.py: a root whose own .git this user cannot read is a PermissionError traceback on 3.12, or not-a-git-repo where exists() answers False, not refused by name #1148 says of this file and three other pinned copies.

• The test is tests/test_siblings_family_name.py. Its layout() (:111) collapses runs of spaces (" ".join(line.split())), so no row there can see a pad; the new rows read the raw line, with the member rows its fixtures already build (ROWS, with_rows, about :290-300).

• A member whose project: is U+65E5 U+672C U+8A9E: the state word starts at the same display column as in an ASCII member's row, on a UTF-8 stdout. The row fails before and passes after.

• The same for a name holding e U+0301.

• The same two rows on an ASCII stdout (NARROW, tests/test_bootstrap_spelling.py:86): the escaped form is counted as stdout writes it.

• An ASCII name prints byte for byte as before: the file's existing rows pass unchanged.

• One mutant, the pad back to len, fails a new row; each mutant run is restored and checked with cmp.

Related: #1178, #1034 (adopt-project.py's plan table, the same defect), #1148 (names this file among four pinned copies), and #459 (the tracker for setup, scaffold, family and the shared helpers).

Size: S.

Lane: openreposhape-3 (openRepoShape-3)