[brettheap]

How to review

• The change is four commits on 84a09c2e. Review it as git diff 84a09c2e 2572a717 (9 test files, +629/-118). 40531e19 is the change. 60836f5c and bcf6952e are fix round 1, which answers lane 3's review RV869 (comment); review the round alone as git diff 40531e19 bcf6952e (3 test files, +101/-23). 2572a717 is fix round 2, which answers the two optional NITs of lane 3's re-check RV869b (comment); review it alone as git diff bcf6952e 2572a717 (3 test files, +72/-4). GitHub's diff against fix/834-oserror-skips also shows the two merges below.
• Merge 1, e387a3c2: tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's current head a1498632. tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's head 8f6a38e3 is stacked on tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's older head 76ae6419. Merging main straight onto it conflicts in one place: the from conftest import line of tests/test_adopt_git_reads.py (tests/test_doctor_legs_members_next.py and 24 other files: tests append what they tell git to the GIT_CONFIG_COUNT entries the environment holds, so a pasted line keeps the session's maintenance settings and the Windows job's two (#758) #818's appended_to_command_scope beside tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's skip_where_a_name_cannot_be_not_utf_8). tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's own merge of e62b5e17 already resolves that line, keeping both names, so taking tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's head first makes the next merge clean.
• Merge 2, 84a09c2e: origin/main 3fb30a8f. It brings adopt-project.py: _default_branch reads origin/HEAD and HEAD by full name and whole, and reads the source only at a branch of its own, so a git clone -b source is no longer git-failed with no remedy, and a HEAD on origin/x, on main<U+00A0> or beside a tag main no longer adopts another branch's commit (#797) #830 (51414d2, adopt-project.py: _default_branch reads origin/HEAD and HEAD by short name, so a source cloned with git clone -b <branch> is refused git-failed with no remedy, and a HEAD on a local origin/x or on a name ending in U+00A0 adopts another branch's commit #797), which tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853's second line waits for. adopt-project.py: _default_branch reads origin/HEAD and HEAD by full name and whole, and reads the source only at a branch of its own, so a git clone -b source is no longer git-failed with no remedy, and a HEAD on origin/x, on main<U+00A0> or beside a tag main no longer adopts another branch's commit (#797) #830 rewrote with_default_branch and added remembered_on. Both merges were clean, and their trees match the merge-tree predictions (88a781dc, f7147379).
• Landing order. This PR is stacked on lane 5's tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845, which is stacked on tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806. It lands after both, by lane 3's three-merge restack.

Closes #853.
Closes #854.

Tests only. No tool, template, pinned file or doc changes. Beyond the two issues' text, the PR also makes six more tests in tests/test_adopt_template_values.py skip under a C locale, where they raised UnicodeEncodeError (see "Beyond #853 and #854" below).

What changed

#853, line 1: three tests no longer skip when make_source_repo fails

refused_then_moved (tests/test_adopt_template_values.py), test_a_utf8_clone_of_a_source_whose_path_is_not_utf8_adopts (tests/test_adopt_git_reads.py) and test_a_source_directory_that_is_not_utf8_is_shown_spelled (tests/test_adopt_plan_escapes.py) each wrapped make_source_repo(parent / "Thing") in except AssertionError: pytest.skip("git will not work under ..."). All three try blocks are gone.

• The answers that mean "this system cannot carry the name" are unchanged. tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's skip_where_a_name_cannot_be_not_utf_8 still asks them before the mkdir.
• Git works under any directory the system carries, so if git fails after that question, the test fails. A disk that fills after the mkdir is the case this catches.
• refused_then_moved was worked from tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's version, which had added skip_where_git_fails (true for the byte name, false for U+202E). With the try gone, the parameter has no use, so it is removed, together with the U+202E caller's skip_where_git_fails=False.
• tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's control test test_a_failure_of_git_under_the_byte_directory_still_skips_the_test, in tests/test_non_utf8_oserror_sites.py, asserted this skip as a known limit. It is removed, and that file's docstring points to the new tests.
• Windows, held for each of the three (fix round 1). tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853's Verify asks that Windows' os.fsencode still skips. A new test, test_where_fsencode_makes_no_byte_the_byte_directory_test_skips_first, has one row for each of the three tests. With windows_fsencode in place, each test must skip on the os.fsencode reason, and nothing may be made under tmp_path. RV869's mutant asks the question after the template-values test's mkdir. It passed every row before, and it fails this one now.

#853, line 2, and the sixth site: three branch helpers no longer skip when git fails

with_default_branch, moved_to_a_branch_that_is_not_utf8 and remembered_on (tests/test_adopt_git_reads.py) caught except (AssertionError, UnicodeError, OSError) around the git calls that make the branch.

• remembered_on is the sixth site. adopt-project.py: _default_branch reads origin/HEAD and HEAD by full name and whole, and reads the source only at a branch of its own, so a git clone -b source is no longer git-failed with no remedy, and a HEAD on origin/x, on main<U+00A0> or beside a tag main no longer adopts another branch's commit (#797) #830 added it after the issue's scan of main at fc867207. The same AST scan, run on this branch after merge 2, found it at tests/test_adopt_git_reads.py:730. It is called with ma, U+2028 and in (text) and with caf and byte 0xE9 (not UTF-8).
• Each helper now calls a new function, skip_where_no_ref_can_be_named(source, branch), before its first git call, and the except is gone. That function asks:
  • for a name that is not UTF-8: skip_where_a_name_cannot_be_not_utf_8(source.parent), as the issue says. That asks Windows' os.fsencode first, then macOS's EILSEQ, on the disk the ref files go to.
  • for a name that is text: only the file name encoding. This is one step past the issue's "for a UTF-8 name, ask nothing". subprocess puts every argument through os.fsencode, so under LC_ALL=C the git call for main plus U+00A0, or for U+2028, raised UnicodeEncodeError in the test. On the base those rows skipped through the broad except, as no branch here is named b'main\xc2\xa0': UnicodeEncodeError(...). Asking nothing would have turned those skips into failures under the C locale. Under UTF-8 the question never skips.
• moved_to_a_branch_that_is_not_utf8 asks before it makes anything, so a skip costs no make_source_repo and no plan run.
• The skip after git symbolic-ref, in with_default_branch, stays. Whether git gives the name back as it was written is the system's answer, and no exception is behind it.

#854: the U+202E test skips under an ASCII file name encoding

test_execute_refuses_a_local_source_directory_holding_a_bidi_control now calls skip_where_the_environment_cannot_carry(parent.name) before parent.mkdir(). Nothing else in the test is guarded.

• The helper moved. It is tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's, and it moved from tests/test_non_utf8_git_reads.py to tests/conftest.py together with its callers: tests/test_non_utf8_git_reads.py (3 calls), tests/test_non_utf8_name_callers.py (3) and tests/test_non_utf8_oserror_sites.py (1). Without the move, test_adopt_template_values.py, which three other test files import, would have had to import test_non_utf8_git_reads.py and, through it, test_bump_leg.py and test_commit_trailers.py. It now sits after skip_message, away from update-shape.py, family.py and bump-leg.py: the branch HEAD is on is read as the bytes git wrote, so a branch ending in U+00A0 is no longer read as its neighbour, and a name that is not UTF-8 is refused by name before anything is written (#765) #849's hunk in conftest.py.
• Its reason now says "cannot spell", where it said "cannot carry". The skip reads "this system's file name encoding (ascii) cannot spell '\u202ecaf'", worded as make_file's encoding skip is since tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's fix round. The existing tests assert only "file name encoding", so tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's three callers keep passing. Their reasons change by that one word.
• tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's two U+202E tests in tests/test_non_utf8_oserror_sites.py (the disk sweep and the git failure) now ask the same question first. Without it, under the C locale, the bidi test would skip before the disk or git is reached, and both tests would fail on a skip.

Beyond #853 and #854: six more tests skip under a C locale (fix round 1)

Under PYTHONUTF8=0 LC_ALL=C PYTHONCOERCECLOCALE=0 LANG=C LC_CTYPE=C, six tests in tests/test_adopt_template_values.py raised UnicodeEncodeError in the test body. Neither issue names them: #854 names only the U+202E test, and #853 is about AssertionError skips. They are the same kind of failure as #854, in a file this PR already changes, so this PR takes them (the review's NIT 2) and they need no issue of their own.

• Five raised at subprocess.run in adopted. In ASCII it cannot put the value on plan's command line. They are test_plan_refuses_a_flag_no_file_can_hold[--elected-by='Jos\\u2028e'], [--elected-by='\\u202eevil'], [--reference='ref \\u2066x\\u2069 y'], test_a_valid_non_ascii_elector_still_adopts[accented] and [zero-width-non-joiner]. Each now calls skip_where_the_environment_cannot_carry with its value (the elector, in the second test) as its first line, before make_source_repo and adopted.
• The sixth, test_a_utf8_source_directory_still_adopts_with_local_remotes, raised at the mkdir of caf and U+00E9 in make_source_repo (tests/conftest.py:916). It now asks about that directory's name before make_source_repo.
• Measured, under C.
  • At 40531e19, the six: 6 failed, 3 passed, 69 deselected. The 3 that pass are the flag test's surrogate and TAB rows, which ASCII with surrogateescape can carry.
  • The same six failed on the base 84a09c2e and on origin/main (RV869's runs 3 and 5).
  • At bcf6952e: 3 passed, 6 skipped. Each of the six skips on "this system's file name encoding (ascii) cannot spell ...".
• Under UTF-8, all six run and pass. Nothing in the 11-file run below is skipped.
• CI holds the six too (fix round 2). CI runs UTF-8, where a test that asked its question after make_source_repo or adopted still passed. Six emulated rows in tests/test_non_utf8_git_failures.py now hold them (see "Fix round 2" below).

New: tests/test_non_utf8_git_failures.py (42 tests)

Each place is called as a plain function, as tests/test_non_utf8_oserror_sites.py does. The tests reuse #845's GIT_FAILED, a_source_git_cannot_make, the_error_of and BIDI_TEST, and #806's fixtures, refusing and ascii_fsencode.

group	what is emulated	the caller must	tests
line 1	make_source_repo raises AssertionError("... No space left on device") after the byte directory is made	fail with that error	3
line 1, Windows	os.fsencode gives a surrogate three bytes, and make_source_repo fails	skip on os.fsencode before the mkdir or git, with nothing made under tmp_path	3
line 2, not UTF-8	the module's git raises it	fail with it	4
line 2, text	the same	fail with it	2
macOS	EILSEQ on the probe	skip on the helper's reason before git, with nothing made under tmp_path	4
Windows	os.fsencode gives a surrogate three bytes	skip on os.fsencode before the disk or git, with nothing made	4
text asks nothing about the byte	both of the above, with git failing	fail with git's error, not skip	2
ASCII encoding, text branch	os.fsencode is ASCII	skip on "file name encoding" before git	2
#854	os.fsencode is ASCII, and os.mkdir raises UnicodeEncodeError as the real one does there	skip on "this system's file name encoding ... cannot spell", with nothing made	1
#854, the six template-values tests (fix round 2)	os.fsencode is ASCII, and make_source_repo fails	skip on "file name encoding" before make_source_repo, with nothing made under tmp_path	6
the scan	every tests/*.py, read as #853's scan reads it, and wider	no except that catches an AssertionError holds a pytest.skip. That covers one that names AssertionError, Exception or BaseException (alone or in a tuple), and a bare except	1
the scan's own rows	found: except AssertionError alone, in a tuple, or with the skip behind a raise; except Exception; except (OSError, BaseException); a bare except. Not found: except AssertionError or except Exception with no skip; one for another error; the symbolic-ref kind of skip	as listed	10

The states the helper answers still skip for the three line-1 tests:

• tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's rows in tests/test_non_utf8_name_callers.py (EILSEQ and EACCES on the probe, for adopt-git-reads, adopt-template-values and adopt-plan-escapes-directory) hold macOS's answer and the refusal.
• The new line-1 Windows rows hold Windows' os.fsencode.

All of them pass in the runs below.

AST scan after the change. I re-ran #853's scan, an except that names AssertionError and holds a pytest.skip, over tests/*.py. It finds no site. Before the change it found 6 on the merged tree: the issue's 5 and remembered_on.

Fix round 1 widened the scan (the review's NIT 3).

• skips_on_an_assertion_error is now skips_that_catch_an_assertion_error. It also finds a bare except, an except Exception or an except BaseException that holds a pytest.skip. Each of those catches a failed git call without naming it.
• Before the widening, every open PR's head was scanned for such a handler (see "Merge checks"). None holds one, so the wider scan fails no later landing.

Tests

Linux, Python 3.12.3. Every run had a stub gh first on PATH (it exits 97), GH_TOKEN=bogus, an empty GH_CONFIG_DIR and /usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/bt-<name> -o tmp_path_retention_policy=failed -rs <files>. The C locale adds PYTHONUTF8=0 LC_ALL=C PYTHONCOERCECLOCALE=0 LANG=C LC_CTYPE=C, and sys.getfilesystemencoding() printed ascii there.

UTF-8, on the change's files. The 11 files are tests/test_adopt_template_values.py, tests/test_adopt_git_reads.py, tests/test_adopt_plan_escapes.py, tests/test_non_utf8_name_helpers.py, tests/test_repo_hygiene.py, tests/test_non_utf8_oserror_sites.py, tests/test_non_utf8_name_callers.py, tests/test_non_utf8_git_reads.py, tests/test_non_utf8_git_failures.py, tests/test_family_init_asks_every_value.py (it imports test_non_utf8_git_reads.py) and tests/test_no_auto_maintenance.py (it scans every test file):

550 passed in 184.60s (0:03:04)

Nothing skipped. The new file alone: 36 passed in 9.36s. At 40531e19 the same 11 files gave 543 passed. The 7 new ones are the 3 line-1 Windows rows and the 4 new rows of the scan.

C locale, on the 8 changed files. These are files 1-4 and 6-9 of the 11 above: the 7 changed test files and tests/test_non_utf8_name_helpers.py. test_repo_hygiene.py is left out.

302 passed, 32 skipped in 121.47s (0:02:01)

• No failure, and no error at collection or setup.
• At 40531e19 the same run was 6 failed, 295 passed, 26 skipped. The 6 are the tests under "Beyond tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853 and tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854" above, which skip now, and the 7 new tests pass.
• The U+202E test SKIPS, on the encoding's reason: SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell '\u202ecaf'.
• The base, 84a09c2e, on its 7 files (the 8 less the new one), under the same environment: 7 failed, 272 passed, 20 skipped in 131.11s (0:02:11). The 7 were those six plus the U+202E test, which failed with UnicodeEncodeError: 'ascii' codec can't encode character '\u202e'.

The skip reasons of the C-locale run (-rs):

SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'Jos\u2028e'
SKIPPED [2] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell '\u202eevil'
SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'ref \u2066x\u2069 y'
SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell '\u202ecaf'
SKIPPED [3] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'Jos\xe9'
SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'Ali\u200cReza'
SKIPPED [1] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'caf\xe9'
SKIPPED [6] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'main\xa0'
SKIPPED [3] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'ma\u2028in'
SKIPPED [2] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell '\u202e'
SKIPPED [3] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'caf\xe9.yml'
SKIPPED [2] tests/test_non_utf8_git_reads.py:111: this system's file name encoding cannot spell 'caf\xe9.yml': 'ascii' codec can't encode character '\\xe9' in position 145: ordinal not in range(128)
SKIPPED [4] tests/test_non_utf8_git_reads.py:111: this system's file name encoding cannot spell 'caf\xe9.yml': 'ascii' codec can't encode character '\\xe9' in position 117: ordinal not in range(128)
SKIPPED [2] tests/conftest.py:1219: this system's file name encoding (ascii) cannot spell 'Mehr\u200cdad'

Mutants

Each mutant was applied alone in a scratch worktree, and tests/test_non_utf8_git_failures.py was run. All 13 are caught:

mutant	result	caught by
line 1: the try/skip back in refused_then_moved	2 failed, 27 passed	test_a_failure_of_git_under_the_byte_directory_fails_the_test[adopt-template-values], the scan
the same in the git-reads test	2 failed, 27 passed	[adopt-git-reads], the scan
the same in the plan-escapes test	2 failed, 27 passed	[adopt-plan-escapes], the scan
line 2: the except back in with_default_branch	5 failed, 24 passed	both not-UTF-8 rows, both NBSP rows, the scan
the same in moved_to_a_branch_that_is_not_utf8	2 failed, 27 passed	its git-failure row, the scan
the same in remembered_on (the sixth site)	4 failed, 25 passed	its not-UTF-8 row, both U+2028 rows, the scan
skip_where_no_ref_can_be_named asks nothing for a name that is not UTF-8	8 failed, 21 passed	the 4 EILSEQ rows and the 4 Windows rows
it asks nothing for a name that is text	2 failed, 27 passed	both ASCII-encoding rows
a text name asks the byte helper too	2 failed, 27 passed	both "asks nothing of a byte" rows
moved_to_a_branch_that_is_not_utf8 asks after make_source_repo	2 failed, 27 passed	its EILSEQ and Windows rows (tmp_path not empty)
#854: the encoding check dropped from the U+202E test	1 failed, 28 passed	test_where_the_file_name_encoding_cannot_spell_u202e_the_bidi_test_skips
#854: the check after the mkdir	1 failed, 28 passed	the same
the helper's reason back to "cannot carry"	1 failed, 28 passed	the same

With a real git failure. A stub git first on PATH, in the scratch worktree only, failed one subcommand as a full disk does (error: unable to write file .git/objects/xx: No space left on device, exit 128) and ran /usr/bin/git for everything else:

• git commit fails, test_execute_refuses_a_local_source_directory_that_is_not_utf8:
  • this branch: 1 failed in 0.39s (AssertionError: error: unable to write file ...);
  • the line-1 mutant: 1 skipped in 0.38s, SKIPPED [1] tests/test_adopt_template_values.py:266: git will not work under 'caf\udce9' here: error: unable to write file .git/objects/xx: No space left on device. This is the test that passes where it should fail.
• git branch -m fails, test_check_finds_a_plans_branch_gone_for_one_that_is_not_utf8:
  • this branch: 1 failed in 0.98s;
  • the moved_to_a_branch_that_is_not_utf8 mutant: 1 skipped in 0.95s ("no branch here is named b'caf\xe9'").
• git update-ref fails, test_the_way_through_spells_a_name_no_command_line_carries (the sixth site):
  • this branch: 2 failed in 0.63s;
  • the remembered_on mutant: 2 skipped in 0.79s.

#854 in the real C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control:

• this branch: 1 skipped in 0.15s, on the reason above;
• the mutant without the check: 1 failed in 0.26s, with UnicodeEncodeError: 'ascii' codec can't encode character '\u202e' in position 115 at the mkdir, as on the base;
• the same mutant under UTF-8: 1 passed.

Fix round 1. Each mutant was applied alone in a scratch worktree:

mutant	run	result	caught by
RV869's survivor: the template-values byte test asks skip_where_a_name_cannot_be_not_utf_8 after its mkdir	the four test_non_utf8_* files (git_failures, name_callers, oserror_sites, name_helpers), UTF-8	1 failed, 107 passed	test_where_fsencode_makes_no_byte_the_byte_directory_test_skips_first[adopt-template-values]
test_a_valid_non_ascii_elector_still_adopts asks after adopted	the test, C	2 failed (UnicodeEncodeError for U+00E9 and for U+200C)	both of its rows
test_plan_refuses_a_flag_no_file_can_hold asks after adopted	the test, C	3 failed, 3 passed	its U+2028, U+202E and U+2066 rows
test_a_utf8_source_directory_still_adopts_with_local_remotes asks after make_source_repo	the test, C	1 failed (UnicodeEncodeError at the mkdir)	the test
RV869's own1: except Exception as exc: pytest.skip(...) around with_default_branch's git calls	tests/test_non_utf8_git_failures.py, UTF-8	5 failed, 31 passed	test_no_test_skips_on_an_assertion_error ({'test_adopt_git_reads.py': [567]}) and the 4 behaviour rows of with_default_branch. With the scan of 40531e19 in its place, the scan test passes on own1: 1 passed

• How the first mutant fails. It fails because windows_fsencode takes no Path, at the test's own os.fsencode(tmp_path). I also made windows_fsencode take one (os.fspath, in the scratch tree only). Then the same row fails on assert list(tmp_path.iterdir()) == [], with the directory caf\udce9 left there. Either way the row fails.
• The three C mutants pass under UTF-8, which is CI's locale (2 passed, 6 passed, 1 passed). At bcf6952e a C-locale run like the one above held those skips, and no emulated row did. Fix round 2 adds the rows (below).

Fix round 2 (2572a717): RV869b's two NITs

NIT 1: windows_fsencode takes a path (tests/conftest.py:1187). It calls os.fspath(text) first, as the real os.fsencode does, so a test that asks os.fsencode(tmp_path) under the emulation no longer raises AttributeError: 'PosixPath' object has no attribute 'encode' before the question it holds. A new row, test_windows_fsencode_takes_a_path_as_os_fsencode_does (tests/test_non_utf8_name_helpers.py:93), asserts windows_fsencode(Path("caf\udcff")) == b"caf\xed\xb3\xbf". Its non-ASCII is written as escapes.

NIT 2: six emulated rows for the six C-locale skips (tests/test_non_utf8_git_failures.py:299-347, WHERE_ASCII_CANNOT_SPELL and test_where_the_file_name_encoding_is_ascii_the_template_values_tests_skip). Each row sets os.fsencode to ascii_fsencode and make_source_repo to a_source_git_cannot_make, as the line-1 rows do. It then asserts that the test skips on "file name encoding" and that tmp_path is empty. The rows are flag-u2028, flag-u202e, flag-u2066, elector-accented, elector-zwnj and cafe-directory. A test that asked after make_source_repo fails on git's error under UTF-8, which is CI's locale.

Runs (stub gh first on PATH, as above):

• UTF-8, the 7 files that use windows_fsencode or ascii_fsencode (tests/test_bootstrap_branch_bytes.py, tests/test_bootstrap_non_utf8_branch.py, tests/test_non_utf8_git_failures.py, tests/test_non_utf8_git_reads.py, tests/test_non_utf8_name_callers.py, tests/test_non_utf8_name_helpers.py, tests/test_non_utf8_oserror_sites.py), plus tests/test_adopt_template_values.py and tests/test_repo_hygiene.py: 411 passed in 111.76s (0:01:51), nothing skipped.
• C locale, tests/test_non_utf8_git_failures.py and tests/test_adopt_template_values.py: 109 passed, 11 skipped in 44.32s. The 11 skips are the 6 template-values tests, the U+202E test and the 4 text-branch rows, each on "file name encoding ... cannot spell".

Mutants, under UTF-8, each alone in a scratch worktree, tests/test_non_utf8_git_failures.py:

mutant	result	caught by
RV869b's m2b: the flag test asks after adopted	3 failed, 39 passed	[flag-u2028], [flag-u202e], [flag-u2066]
RV869b's m2c: the caf and U+00E9 test asks after make_source_repo	1 failed, 41 passed	[cafe-directory]
the flag test asks about flag, not value	3 failed, 39 passed	the same three flag rows
the elector test asks after make_source_repo	2 failed, 40 passed	[elector-accented], [elector-zwnj]
NIT 1: os.fspath taken out of windows_fsencode	1 failed, 60 passed (with tests/test_non_utf8_name_helpers.py)	test_windows_fsencode_takes_a_path_as_os_fsencode_does, by the AttributeError
RV869's m1: the template-values byte test asks after its mkdir	1 failed, 114 passed (the four test_non_utf8_* files)	[adopt-template-values] of the byte-directory Windows rows, on assert list(tmp_path.iterdir()) == [] and no longer on an AttributeError

Merge checks

git merge-tree --write-tree of bcf6952e:

• against origin/main (d6475f96): clean (tree 963aff8b);
• against tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 (8f6a38e3): clean. The result is the head's own tree, 840b7794;
• against tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806: clean. Its head moved during this round, from a1498632 to 22238b47, which merges origin/main d6475f96 into it. a1498632 gives the head's own tree, 840b7794, and 22238b47 gives 963aff8b, the same tree as the merge with origin/main. This round made no merge: tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869 restacks after tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 lands;
• against family.py, adopt-project.py and the four tools' --org: init asks GitHub's limits on the topic, the --org and the holder's name before any repository exists, check finds what execute refuses from one list, and an --org GitHub gives no organisation (., a trailing hyphen, --) is refused where GitHub is asked (#863) #871 (66fc8f6b, which touches tests/test_non_utf8_git_reads.py): clean;
• against update-shape.py, family.py and bump-leg.py: the branch HEAD is on is read as the bytes git wrote, so a branch ending in U+00A0 is no longer read as its neighbour, and a name that is not UTF-8 is refused by name before anything is written (#765) #849 (fix/765-branch-read-bytes, d98df568, which touches tests/conftest.py):
  • tests/conftest.py merges clean.
  • It conflicts in tests/test_bump_leg_branch_and_push_line.py and tests/test_family_tracking_branch.py, files this PR does not touch, as it did at 40531e19.
  • Those conflicts are inherited: merge-tree origin/main d98df568, at d6475f96, conflicts in the same two files.

gh pr list --state open --limit 200 --json number,files shows 34 open PRs. Only #806, #845, #849 and #871 touch these 8 files.

The open PRs' tests/, for the widened scan.

• Fix round 1 fetched every open PR's head (gh pr list --state open --limit 200 --json number,headRefOid, 34 open, this one included). It scanned each head's tests/ with the widened scan.
• No head holds a bare except, an except Exception or an except BaseException that holds a pytest.skip.
• Every other head holds only the known except AssertionError sites (5 or 6 a head), which it keeps from main before this PR.
• The new head holds none, and neither does its merge with origin/main.

Left out

• No Windows or macOS run. CI is Linux only (ci: run the test suite on Linux only #597). The Windows and macOS answers are emulated.
• The six C-locale skips are held by emulated rows (fix round 2). The rows hold each test's question, which it asks through os.fsencode. They do not change what subprocess does with an argument, so make_source_repo is replaced by a function that fails: a test that went on past its question fails on git's error under UTF-8. A real C-locale run is still what finds a new test of this kind.

Lane: openreposhape-3 (openRepoShape-3)
🤖 Generated with Claude Code

Summary by Sourcery

Make non-UTF-8 test skips depend on explicit system capability checks while allowing unexpected Git and filesystem failures to fail the tests.

New Features:

• Add coverage for failures and platform-specific encoding behavior in non-UTF-8 Git and template-value tests.

Bug Fixes:

• Ensure genuine Git and filesystem failures fail tests instead of being incorrectly converted into skips.
• Skip tests before filesystem or subprocess operations when the environment cannot encode the required names or values, including U+202E cases under ASCII locales.

Enhancements:

• Centralize environment encoding checks and apply them consistently across affected test helpers.
• Add static checks preventing exception handlers from hiding failed Git calls behind pytest skips.

Tests:

• Add comprehensive emulated Windows, macOS, and C-locale coverage for encoding probes, Git failures, skip ordering, and exception-handler scanning.

[brettheap]

(lane openRepoShape-3, 2026-10-18T03:55:44Z) Lane 3's independent review of its own #869 (#853, #854) at 40531e19, by RV869 (opus): READY. No MUST. One SHOULD: line 1 has no Windows (windows_fsencode) row at the caller level, so a mutant that asks after the mkdir survives. Four NITs. CI is green on all four checks. A fix round starts now (W869f). It takes the SHOULD and NITs 1, 3 and 4. It also takes NIT 2's six C-locale failures in tests/test_adopt_template_values.py, a file this PR already changes, so they need no issue of their own.

• To lane openRepoShape-5, on tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806 and tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845: tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869 still merges clean with both heads. A later round of yours conflicts with tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869 as text only if it edits the helper body this PR moves to tests/conftest.py, the three call sites in test_non_utf8_name_callers.py, refused_then_moved's signature or tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's control test (check 2b below). The old module path test_non_utf8_git_reads.skip_where_the_environment_cannot_carry still resolves.

---

Verdict: READY. There is no MUST. There is one SHOULD and four NITs. The tests CI job finished green on 40531e19 while I reviewed, so nothing is still pending.

Review of #869, head 40531e19. The reviewed change is git diff 84a09c2e 40531e19: 8 files, +475/−110, one commit that touches exactly those 8 files. Every number below is my own run.

The checks the brief asked for

1. Both merges are exact and clean.

• e387a3c2 (parents 8f6a38e3 and a1498632): merge-tree --write-tree exits 0 and gives 88a781dc, which is the commit's tree.
• 84a09c2e (parents e387a3c2 and 3fb30a8f): exits 0 and gives f7147379, which is the commit's tree. 3fb30a8f is on origin/main.

2a. A text branch name asking the file name encoding is right.

• I made a mutant where a text name asks nothing, as tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853 says, and ran it under the real C locale. The U+00A0 rows (test_plan_refuses_a_default_branch_ending_in_white_space and test_plan_reads_..._whole, both where values) and the U+2028 row (test_the_way_through_..[u2028]) gave 5 failed, 1 passed in 3.86s.
• The 5 failures are 4 × UnicodeEncodeError for U+00A0 and 1 × for U+2028. They come from subprocess encoding the arguments, in the test, before git starts.
• Under UTF-8, and on Windows and macOS, os.fsencode cannot fail on text, so the question adds no skip there. The UTF-8 run had 543 passed and 0 skipped.
• It is narrower than the old except, and it follows tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's rule for environment values. The issue's "ask nothing" did not allow for the C locale, and the PR body records the departure.

2b. Moving the helper and rewording its reason breaks nothing.

• Nothing asserts "cannot carry". The git grep on the head, on tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's head and on tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's head finds only "file name encoding" in reason.

• test_non_utf8_git_reads.skip_where_the_environment_cannot_carry still resolves, through that module's import from conftest. An identity check returned True. So a later tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806 or tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 round that calls the old module path still works.

• merge-tree of the head with tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806 and with tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 is clean. Both give the head's own tree 57507cdb, because both are already in it.

• A later round would conflict as text only if it edits one of these:

  • the helper body this PR deletes from test_non_utf8_git_reads.py;
  • the three call sites rewritten in test_non_utf8_name_callers.py;
  • refused_then_moved's signature;
  • tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's control test.

  The restack settles any of these.

• "cannot spell" is the wording tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854 asks for, and it matches tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's make_file reason.

2c. The control test's purpose is kept, the other way round.

• tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's control held two things: that the byte test skips when git fails, and that skip_where_git_fails defaults to true. With the parameter gone, the second has nothing left to hold.
• The first is now held as "fails" by test_a_failure_of_git_under_the_byte_directory_fails_the_test[adopt-template-values]. It runs the same test with the same monkeypatch and the same probe fixture, and asserts GIT_FAILED.
• With a real git failure, the base skips and the head fails (runs 7 and 8 below).

3. The scan, with my own script, independent of the writer's.

Tree	except naming AssertionError that holds a skip
base 84a09c2e	6: test_adopt_git_reads.py 472, 549, 569, 730; test_adopt_plan_escapes.py 408; test_adopt_template_values.py 265
head	0
head merged with origin/main 4dea85ee (tree b39ecb29)	0
origin/main	6

• On the head there are also 0 bare, Exception or BaseException handlers holding a skip.
• tests/ has no subfolders, so glob("*.py") covers every file.

4. remembered_on's skip is the narrowest one that is right.

• For the byte name it asks os.fsencode (Windows), then the EILSEQ probe on source.parent (macOS).
• That is the same disk as .git/refs, and it also works in the new rows, where source does not exist yet.
• For U+2028 it asks only the encoding.
• The old "or alters the name" in its docstring had no read-back behind it, so nothing is lost.

5. The 6 C-locale failures are unchanged and were already there.

• They have the same names and the same error lines on the base and on the head.
• origin/main 4dea85ee fails the same 6 plus the U+202E test.
• Five of them raise at subprocess.run in adopted (tests/test_adopt_template_values.py:100).
• The sixth, test_a_utf8_source_directory_still_adopts_with_local_remotes, raises at the mkdir of caf+U+00E9 in make_source_repo (tests/conftest.py:916). That is exactly tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854's kind of failure.
• Neither issue covers them. tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854 names only the U+202E test. tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853 is about AssertionError skips, and its Verify says the files pass on Linux, which is CI's UTF-8.

7. Windows.

• Windows' os.fsencode is UTF-8 with surrogatepass, so skip_where_the_environment_cannot_carry never skips there. That is correct: on Windows, subprocess and mkdir pass text as UTF-16.
• skip_where_no_ref_can_be_named sends a name that is not UTF-8 through skip_where_fsencode_gives_a_surrogate_no_byte before the disk or git is touched.
• Every new-file row that uses real git or a real probe carries a_system_that_carries_the_byte, so it skips on Windows.
• The gap is line 1; see SHOULD 1.

8. SonarCloud analysed 40531e19:

• quality gate OK, 0 issues, 0 hotspots;
• 0 new duplicated lines and 0 code smells in each of test_non_utf8_git_failures.py (363 new lines), test_adopt_git_reads.py, conftest.py and test_non_utf8_oserror_sites.py;
• no composite assert a and b among the added lines.

9. Merges. See the merge results below.

Other hygiene:

• The diff adds no non-ASCII character, no tab and no line over 79 columns.
• pyflakes shows only the noqa: F401 fixture-import patterns that the base also has.

SHOULD

SHOULD 1: line 1 has no Windows (windows_fsencode) row at the caller level.

• Where: tests/test_non_utf8_git_failures.py:109-120. The tests concerned are tests/test_adopt_template_values.py:300, tests/test_adopt_git_reads.py:467 and tests/test_adopt_plan_escapes.py:403.

• Scenario: tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853's Verify asks that "the states the helper answers (EILSEQ on the probe, and os.fsencode of Windows through windows_fsencode) still skip". Line 2 has Windows rows that also assert nothing is made. Line 1 relies only on tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's EILSEQ and EACCES rows, and the PR body names only those.

• Measurement: I made a mutant that moves skip_where_a_name_cannot_be_not_utf_8(tmp_path) after parent.mkdir() in the template-values byte test. It survives all four non_utf8 files (git_failures, name_callers, oserror_sites, name_helpers): 101 passed in 6.32s. On real Windows that order raises at os.fsdecode(... b"caf\xe9") instead of skipping.

• Why it is not a MUST: the order comes from tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806, and this PR does not change it.

• Fix: add a 3-row test over UNDER_A_BYTE_DIRECTORY that:

  1. sets os.fsencode to windows_fsencode;
  2. takes reason = skip_message(call, tmp_path, adopter);
  3. asserts "os.fsencode" in reason;
  4. asserts list(tmp_path.iterdir()) == [].

  My prototype passes on the head (3 passed in 0.31s) and kills the mutant (1 failed, 2 passed).

NITs

1. A stale docstring sentence. tests/test_non_utf8_oserror_sites.py:28-29 says "The U+202E test makes its directory, and its source repository, with no guard." The test now asks the file name encoding first (tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854). Fix: add "but the file name encoding's question (tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854)".
2. Open a follow-up issue for the six C-locale failures listed under check 5. Fix: call skip_where_the_environment_cannot_carry(value) before adopted in the five, and before the make_source_repo of caf+U+00E9 in the sixth.
3. The scan sees only an except that names AssertionError.
   • My mutant own1, except Exception: pytest.skip around with_default_branch's git calls, is invisible to the scan.
   • The behaviour rows catch it only at the six known sites.
   • Head has no such handlers today, so widening skips_on_an_assertion_error (tests/test_non_utf8_git_failures.py:293-300) to bare, Exception and BaseException handlers costs nothing.
   • This is optional: the Verify is met as written.
4. The PR body's file list is slightly off. "The first 4 and the last 4 of the files above" is really files 1-4 and 6-9: the 7 changed test files plus test_non_utf8_name_helpers.py, 327 tests. The counts reproduce exactly.

Test runs

Every run had a stub gh first on PATH (exits 97), GH_TOKEN=bogus, an empty GH_CONFIG_DIR, and timeout 2400 /usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/bt-RV869-<name> -o tmp_path_retention_policy=failed -rs <files>. "C" means PYTHONUTF8=0 LC_ALL=C PYTHONCOERCECLOCALE=0 LANG=C LC_CTYPE=C, where the file name encoding printed ascii.

#	Tree, environment, files	Result
1	head, UTF-8, the writer's 11 files	543 passed in 244.42s, 0 skipped
2	head, C, 8 files (7 changed + name_helpers)	6 failed, 295 passed, 26 skipped in 183.15s
3	base 84a09c2e, C, 7 files	7 failed, 272 passed, 20 skipped in 161.28s; only the U+202E test differs from run 2
4	head merged with origin/main, UTF-8, 8 files + test_repo_hygiene.py	405 passed in 199.56s
5	origin/main 4dea85ee, C, test_adopt_template_values.py	7 failed, 71 passed in 65.01s
6	head, a plugin that makes os.mkdir answer EILSEQ to a name that is not UTF-8 (new file + 3 adopt files)	183 passed, 20 skipped in 117.82s
7	real git failing commit with an ENOSPC message, the 3 line-1 tests	head: each 1 failed with that AssertionError; base: each 1 skipped, "git will not work under ..."
8	real git failing switch, branch and update-ref, -k "default_branch or branch_gone or way_through"	head 14 failed, 23 deselected in 6.67s; base 14 skipped, 23 deselected in 4.83s
9	#854's Verify, real U+202E test	head under C: 1 skipped in 0.19s, reason "this system's file name encoding (ascii) cannot spell" U+202E+caf. Check removed: C 1 failed in 0.32s (UnicodeEncodeError at mkdir), UTF-8 1 passed in 6.82s
10	CI on 40531e19	tests success (02:46:47Z to 03:53:04Z), tests-git-bash success, package success, SonarCloud pass

Mutants

Each was run against tests/test_non_utf8_git_failures.py under UTF-8 unless noted.

Mutant	Result	Killed by
Writer's: the try/skip back in refused_then_moved	2 failed, 27 passed	[adopt-template-values] and the scan (matches the writer)
Writer's: the except back in remembered_on, helper kept	4 failed, 25 passed	its not-UTF-8 row, both U+2028 rows, the scan (matches). With the helper call also removed, as on the base: 7 failed
Writer's: moved_to_... asks after make_source_repo	2 failed, 27 passed	its EILSEQ and Windows rows (matches)
Writer's: #854 check after the mkdir	1 failed, 28 passed	the U+202E skip test (matches)
Writer's: a text name asks nothing	2 failed, 27 passed	both ASCII rows (matches); under the real C locale, 5 real rows fail (check 2a)
Own own1: except Exception: pytest.skip in with_default_branch	4 failed	the behaviour rows only; the scan cannot see it
Own own2: except Exception: pytest.skip in refused_then_moved	1 failed	[adopt-template-values]
Own own3: a text name asks the byte probe instead of the encoding	4 failed	both "asks nothing" rows and both ASCII rows
Own: line-1 byte test asks after its mkdir	101 passed	survives; SHOULD 1

Merge results

• merge-tree --write-tree of the head:
  • with origin/main 4dea85ee: clean (b39ecb29);
  • with tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806 a1498632: clean;
  • with tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 8f6a38e3: clean.
• With update-shape.py, family.py and bump-leg.py: the branch HEAD is on is read as the bytes git wrote, so a branch ending in U+00A0 is no longer read as its neighbour, and a name that is not UTF-8 is refused by name before anything is written (#765) #849 d98df568, conftest.py merges clean. It conflicts in tests/test_bump_leg_branch_and_push_line.py and tests/test_family_tracking_branch.py. Those conflicts are inherited: the same two files conflict the same way against origin/main and against 84a09c2e, and the content matches once the conflict markers are normalised.
• Against the other 33 open PRs:
  • 28 merge clean, and my scan finds 0 sites in each merged tree;
  • siblings.py and shape-doctor.py: a Windows junction to nothing at ../<Project> is NOT A CLONE, as a link to nothing is, so the dry run no longer says would clone and the doctor's members row no longer calls the slot absent (#793) #794, repo_shape.py, setup-project.py, shape-doctor.py, README and the handbook: under Git Bash (MSYSTEM set) the lines that follow the host are quoted for sh, through one question, host_shell_dialect, so the doctor's lines, setup's family lines and family.py's NEXT, switch -c, add and cd lines paste whole in the shell that printed them, and README's Windows sentence says so #821, adopt-project.py: a --source, or a plan's source.local_path, that names no repository's top level is refused by name before git reads anything else in it, so a file, a directory inside a repository, an unknown ~user, an empty value, owner/.git and a lost blob no longer end in a traceback, plan a subtree or reach gh (#796) #831 and adopt-project.py: the clone plan and check make is claimed with the run's own mkdir and removed only while it is still that directory, and a --work-dir another user can rename in is refused first, so a swapped source-readonly no longer turns the removal onto another directory (#803) #843 conflict exactly as they do against main;
  • shape-doctor.py: the the way in rows say a checkout's own contracts/repository-naming.yaml that does not load, or raises, is that checkout's, an n/a environment row, where a tab line was FINDING the way in refused and a pattern that does not compile ended the doctor in a traceback; and the EXIT CODES list says what INVALID and manifest-kinds name (#707) #778 conflicts in tests/test_doctor_validator_verdicts.py as it does against 84a09c2e, and is clean against main. That is tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's known import-line conflict.
• No open PR head adds a new except AssertionError + pytest.skip site, so the new scan test fails no later landing as things stand.

Scratch paths

All under /workspace/projects/openRepoShape-worktrees/_lane3-scratch/:

• clone: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869
• worktrees:
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-head (40531e19)
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-base (84a09c2e)
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-mut (left clean)
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-mainmerge (tree b39ecb29)
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-main (4dea85ee)
• logs: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-logs/
• prototype row: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-logs/test_rv869_probe_row.py.txt
• scripts:
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-run.sh
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-scan.py
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-mutate.py
• git stubs:
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-failgit/git
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-failgit2/git
• EILSEQ plugin: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/RV869-plugin/rv869_eilseq.py
• basetemps: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869-scratch/bt-RV869-*
• open PR heads: fetched into private refs refs/rv869/pr<N> in the clone.

No Windows or macOS run was possible here; those answers are emulated. Nothing was pushed or posted, and no git config was changed.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-18T04:13:32Z) Fix round 1 (W869f) of lane 3's #869, answering RV869: pushed. The head is now bcf6952e (60836f5c and bcf6952e on top of 40531e19; the round alone is git diff 40531e19 bcf6952e, 3 test files, +101/-23). The body is updated. No merge was made: #806's head moved during the round, from a1498632 to 22238b47 (main d6475f96 merged in). #869 is clean with both, and with #845 8f6a38e3 and origin/main d6475f96. It restacks after #845 lands.

• SHOULD 1, taken. test_where_fsencode_makes_no_byte_the_byte_directory_test_skips_first (tests/test_non_utf8_git_failures.py:128) has 3 rows over UNDER_A_BYTE_DIRECTORY. With windows_fsencode in place, each test must skip on the os.fsencode reason and make nothing under tmp_path. make_source_repo also fails there, so a test that went on fails fast. RV869's survivor (the template-values byte test asking after its mkdir) now gives 1 failed, 107 passed over the four test_non_utf8_* files.

  • The kill comes first from windows_fsencode taking no Path.
  • With it taking one (scratch only), the row fails on the empty-tmp_path assertion instead.

• NIT 1, taken. tests/test_non_utf8_oserror_sites.py:28-29 now says the U+202E test has "no guard but the file name encoding's question, asked before its mkdir (tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854)".

• NIT 2, taken here, not as a follow-up. The six C-locale failures in tests/test_adopt_template_values.py each call skip_where_the_environment_cannot_carry first:

  • the flag test and the elector test, before make_source_repo and adopted;
  • the caf + U+00E9 test, before its make_source_repo.

  Under C: at 40531e19 the six gave 6 failed, 3 passed; at bcf6952e they give 3 passed, 6 skipped, each on the encoding's reason. The 8-file C run is 302 passed, 32 skipped, 0 failed (it was 6 failed). Under UTF-8 the 11-file run is 550 passed, 0 skipped. The mutants that ask too late fail under C (2, 3 and 1 failed) and pass under UTF-8, so these skips are held by a C-locale run, not by CI. The body records that the PR goes beyond tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853's and tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854's text here.

• NIT 3, taken. skips_on_an_assertion_error is now skips_that_catch_an_assertion_error. It also finds a bare except, except Exception and except BaseException that hold a pytest.skip, with 4 new scan rows. RV869's own1 now fails the scan ({'test_adopt_git_reads.py': [567]}; with 40531e19's scan the same test passes).

  • Before widening, all 34 open PR heads were scanned. None holds such a handler: only the known narrow except AssertionError sites kept from older main. So the wider scan fails no later landing.

• NIT 4, taken. The body's C-locale file list now reads "files 1-4 and 6-9 of the 11 above".

An independent re-check of this round (RV869b, opus) starts now.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-18T04:37:54Z) Lane 3's independent re-check of #869's fix round 1 (40531e19..bcf6952e), by RV869b (opus): READY. No MUST and no SHOULD. All five of RV869's items are fixed and measured, and all 33 open PR heads pass the wider scan, as do their predicted merges. Four optional NITs. Lane 3 takes NIT 1 (windows_fsencode takes a path, as the real os.fsencode does) and NIT 2 (emulated rows for the six C-locale skips, so CI holds them) in a small round 2 now. It leaves NIT 3 (scan edge cases that occur in no tree), and takes NIT 4 in the squash message. Restack: #806 landed as d6071c7, and lane 5 has merged it into #845 (fc3cb5e9). #869 merges clean with that head, and restacks after #845 lands.

---

RV869b: independent re-check of fix round 1 on lane 3's #869 (closes #853 and #854). Head bcf6952e: READY.

There is no MUST and no SHOULD, and there are four optional NITs. All five of RV869's items are fixed, and I measured each fix myself. All 33 open PR heads pass the wider scan, and so do their predicted merges. The head is clean with everything the brief named except a plain merge with current main (5 files). That conflict comes only from #806's squash landing at 04:17:02Z, after this round; merged through #845's new head it is clean (merge results below).

What moved while I worked (the PR head did not; ls-remote read bcf6952e again at 04:34:45Z):

• tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806 landed as d6071c7d. Its tree 9f94a180 is the same as its final head 22238b47.
• Lane 5 restacked tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 to fc3cb5e9 (04:17:50Z): one merge of 22238b47, then one of d6071c7d.
• main moved again to 75d70502 (repo_shape.py strips only YAML's space and tab, so a value a Unicode space opens reads as PyYAML reads it, and validate-manifest, bootstrap, bump-leg and update-shape refuse a branch holding a character that does not print or is white space (#715) #823).
• family.py, adopt-project.py and the four tools' --org: init asks GitHub's limits on the topic, the --org and the holder's name before any repository exists, check finds what execute refuses from one list, and an --org GitHub gives no organisation (., a trailing hyphen, --) is refused where GitHub is asked (#863) #871 moved to 99afbc3d.

On CI, bcf6952e has SonarCloud, package and tests-git-bash passing. tests was still pending at 04:34Z. SonarCloud's API for PR 869 shows commit bcf6952e, gate OK, 0 new issues, 0 code smells, 0 new duplicated lines and 0 hotspots. None of the added lines is a composite assert.

The checks

1. SHOULD 1: the Windows rows at tests/test_non_utf8_git_failures.py:128-141.

• The writer's claim holds. Each row asserts three things: a skip happened, its reason contains os.fsencode, and tmp_path is empty.
  • On RV869's survivor (the template-values byte test asking after its mkdir), the first failure is AttributeError: 'PosixPath' object has no attribute 'encode'.
  • That error comes from the test's own os.fsencode(tmp_path) at tests/test_adopt_template_values.py:305. windows_fsencode (tests/conftest.py:1180-1188) takes only str | bytes.
  • The same happens for the other two tests (tests/test_adopt_git_reads.py:468, tests/test_adopt_plan_escapes.py:404).
• Measured both ways, on the 4 test_non_utf8_* files for the template-values mutant and on test_non_utf8_git_failures.py for the other two:
  • windows_fsencode takes a path (text = os.fspath(text) first). All three mutants then fail on the row's own assert list(tmp_path.iterdir()) == [], with caf\udce9 left there: 1 failed, 107 passed, then 1 failed, 35 passed for each of the other two. The head is unchanged across all six files that use windows_fsencode: 221 passed in 52.94s.
  • The row wraps it (lambda value: windows_fsencode(os.fspath(value))). The same assertion fails: 1 failed, 107 passed.
  • The faithful emulation without the empty-tmp_path assertion. The mutant survives (108 passed). With the head's emulation and no assertion, it still dies, but only on the AttributeError.
• Verdict on check 1: the row tests what it says. The empty-tmp_path assertion is the part that matters, and the AttributeError is a side effect of the emulation that hits first. That is NIT 1 below.

2. NIT 2, taken here: the six C-locale tests.

• Placement. Each call comes before the first thing that would raise:
  • the flag test's call is its first line (:205), before make_source_repo and adopted;
  • the elector test's call is its first line (:658);
  • the caf+U+00E9 test builds parent (no encoding happens yet), then asks (:682), then calls make_source_repo (:683).
• Under C:
  • head: 3 passed, 6 skipped, 69 deselected in 1.86s, each skip on "this system's file name encoding (ascii) cannot spell ...";
  • 40531e19: 6 failed, 3 passed, 69 deselected in 3.70s.
• Under UTF-8: 9 passed, 69 deselected in 20.13s, and the 11-file run skips nothing.
• Going beyond the issues. It is acceptable: lane 3 ruled it at the top of RV869's comment, and it is tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854's kind of failure in a file the PR already changes. The body says it plainly, both near the top ("Beyond the two issues' text,") and in its own section. The title does not (NIT 4).

3. NIT 3: the wider scan.

• The forms the brief named. I ran the head's own skips_that_catch_an_assertion_error (taken from the file by AST). All of these are found:
  • except (OSError, Exception), except Exception as e, except (Exception) as e;
  • a nested tuple, except builtins.Exception, except* Exception;
  • a skip in a nested def or behind an if.
• Correctly not found: OSError or subprocess.CalledProcessError with a skip, and except Exception with pytest.fail or pytest.xfail.
• Edge cases (NIT 3): one false positive and two misses, none occurring in any tree today.
• Every open PR head (33, fetched into refs/rv869b/pr<N> in my clone):
  • none holds a bare, Exception or BaseException handler with a skip;
  • each holds only the known narrow sites from older main: 5, or 6 where remembered_on exists;
  • tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869's head holds 0.
• Predicted merges. I merged each of the other 32 PRs into the predicted post-landing tree 963aff8b (main d6071c7d + tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 + tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869). The scan finds 0 in every one of them. The conflicted files (all inherited) parse on neither side's markers, and each side scans clean.

4. NITs 1 and 4.

• tests/test_non_utf8_oserror_sites.py:28-29 now reads "no guard but the file name encoding's question, asked before its mkdir (tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854)".
• The body says "files 1-4 and 6-9 of the 11 above".

5. Every hunk read.

• The round is exactly 3 files, +101/-23:
  • test_adopt_template_values.py: 3 hunks;
  • test_non_utf8_git_failures.py: docstring, the 3-row test, the wider scan, and 4 new scan rows (10 in all);
  • test_non_utf8_oserror_sites.py: 1 docstring line.
• The new file has 36 tests: own1's run counts 5 failed, 31 passed.
• The added lines have no non-ASCII character, no tab and no line over 79 columns. One raw é is removed (now \u00e9).
• pyflakes shows only the known fixture-import patterns.
• No doc or pinned file changes. No reference to the old name skips_on_an_assertion_error remains.
• Windows: the six new calls never skip there, because os.fsencode is UTF-8 with surrogatepass.

NITs (all optional)

NIT 1: windows_fsencode takes no path, so the new rows kill on an emulation artifact first.

• Where: tests/conftest.py:1186. It affects tests/test_non_utf8_git_failures.py:128-141, whose docstring at :134-135 says the mutant "would leave the directory here".
• Scenario: under the head's emulation, a test that asks after its mkdir fails with a misleading AttributeError before it makes anything.
• Measured: see check 1.
• Fix: add text = os.fspath(text) as the first line of windows_fsencode, and add a row in tests/test_non_utf8_name_helpers.py such as windows_fsencode(Path("caf\udcff")) == b"caf\xed\xb3\xbf".
  • This matches the real os.fsencode, and the sibling ascii_fsencode already does it (tests/test_non_utf8_name_callers.py:258).
  • tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's own row already notes this artifact (tests/test_non_utf8_name_callers.py:206, "or fail on os.fsencode of a path under the emulation").
  • Prefer it over the row-local lambda, because it fixes all 7 files that use the emulation.
  • Cost: 221 of 221 still pass, and a scratch commit of it still merges tests/conftest.py clean with update-shape.py, family.py and bump-leg.py: the branch HEAD is on is read as the bytes git wrote, so a branch ending in U+00A0 is no longer read as its neighbour, and a name that is not UTF-8 is refused by name before anything is written (#765) #849 (d98df568).
  • repo_shape.py strips only YAML's space and tab, so a value a Unicode space opens reads as PyYAML reads it, and validate-manifest, bootstrap, bump-leg and update-shape refuse a branch holding a character that does not print or is white space (#715) #823 (now on main) changed tests/test_bootstrap_branch_bytes.py, which uses it: re-run that file if this is taken.

NIT 2: the body's "Left out" says the six C-locale skips cannot get an emulated row. They can.

• Where: the body's "Left out", second bullet.
• Scenario: an order mutant of the six passes in CI, which runs UTF-8.
• Measured: my prototype /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-logs/test_rv869b_c_rows.py.txt has 6 rows. Each sets os.fsencode = ascii_fsencode and make_source_repo = a_source_git_cannot_make, as the line-1 rows do, then asserts a skip on "file name encoding" with tmp_path empty.
  • head, UTF-8: 6 passed in 5.40s; real C: 6 passed in 0.27s;
  • m2b: 3 failed, 3 passed; m2c: 1 failed, 5 passed; own-flag: 3 failed, 3 passed. All of these are under UTF-8.
• Fix: add the rows, or reword the bullet to "not added".

NIT 3: the scan has one false positive and two misses.

• Where: names_in (tests/test_non_utf8_git_failures.py:302) reads an attribute by its last name.
• Scenario:
  • except pytest.skip.Exception: pytest.skip(...) is flagged as an Exception handler;
  • skip(...) imported with from pytest import skip is missed;
  • raise pytest.skip.Exception(...) inside except Exception is missed.
• Measured: none occurs anywhere.
  • The head has no from pytest import, no import pytest as and no raise pytest.skip.Exception.
  • Its two except pytest.skip.Exception handlers (conftest.py:1199, test_non_utf8_name_helpers.py:60) hold no skip.
  • My scan of the PR heads found no such alias either.
• Fix: count Exception and BaseException only as an ast.Name or builtins.<name>, or leave it as is.

NIT 4: the title does not name the six extra skips.

• Where: the PR title. It describes tests: five adopt tests skip on any failure of make_source_repo, or of the git call that makes a branch whose name is not UTF-8, so a full disk or a refused write reads as SKIPPED #853 and tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854 only.
• Fix: name the six C-locale skips in the explicit squash message when landing.

Test runs

All runs used a stub gh first on PATH (exits 97), GH_TOKEN=bogus, an empty GH_CONFIG_DIR, GIT_ALLOW_PROTOCOL unset, and /usr/bin/python3 -m pytest -q -p no:cacheprovider --basetemp=<scratch>/bt-RV869b-<name> -o tmp_path_retention_policy=failed -rs. "C" means PYTHONUTF8=0 LC_ALL=C PYTHONCOERCECLOCALE=0 LANG=C LC_CTYPE=C, where the file name encoding printed ascii.

#	Tree, locale, files	Result
1	head, UTF-8, the body's 11 files	550 passed in 165.42s, 0 skipped
2	head, C, the 8 files (1-4, 6-9)	302 passed, 32 skipped in 142.53s; skip reasons match the body's list
3	head, C, test_adopt_template_values.py -k the six tests' functions	3 passed, 6 skipped, 69 deselected in 1.86s; UTF-8 9 passed ... in 20.13s; 40531e19 C 6 failed, 3 passed ...
4	predicted restack 963aff8b (head + #845 fc3cb5e9), UTF-8, the 11 files + test_stdout_encoding.py	597 passed, 1 skipped in 256.09s; the skip is the Windows-only test at test_stdout_encoding.py:285
5	the same tree, C, the 8 files	302 passed, 32 skipped in 117.35s
6	head + windows_fsencode taking a path, the 6 files that use it	221 passed in 52.94s

Mutants

Under UTF-8 unless marked C.

Mutant	Run	Result	Killed by
m1, the writer's: template-values byte test asks after mkdir	4 non_utf8 files	1 failed, 107 passed	[adopt-template-values] row, by the AttributeError
m1 with windows_fsencode taking a path	same	1 failed, 107 passed	same row, by the empty-tmp_path assertion
m1 with the row wrapping os.fspath	same	1 failed, 107 passed	same, by the empty-tmp_path assertion
m1, path-taking emulation, row without its empty assertion	same	108 passed	survives (the assertion is the part that holds it)
m1, head emulation, row without its empty assertion	same	1 failed, 107 passed	AttributeError only
Own: git-reads test asks after mkdir (head emulation / path-taking)	test_non_utf8_git_failures.py	1 failed, 35 passed each	[adopt-git-reads]: AttributeError / empty assertion
Own: plan-escapes test asks after mkdir (head emulation / path-taking)	same	1 failed, 35 passed each	[adopt-plan-escapes]: AttributeError / empty assertion
own1, the writer's: except Exception as exc: pytest.skip in with_default_branch	same	5 failed, 31 passed	the 4 with_default_branch rows, and the scan {'test_adopt_git_reads.py': [567]}. With 40531e19's scan: 1 passed, 28 deselected (invisible), as the writer says
m2b, the writer's: flag test asks after adopted	C	3 failed, 3 passed	U+2028, U+202E and U+2066 rows (UnicodeEncodeError)
m2c, the writer's: caf+U+00E9 test asks after make_source_repo	C / UTF-8	1 failed / 1 passed	the test (UnicodeEncodeError at conftest.py:916 mkdir); passes under UTF-8 (CI)
Own: flag test asks about flag, not value	C / UTF-8	3 failed, 3 passed / 6 passed	the same 3 rows, under C only
Own: Exception dropped from CATCH_AN_ASSERTION_ERROR	test_non_utf8_git_failures.py	1 failed, 35 passed	scan row [exception]
Own: bare except dropped from catches_an_assertion_error	same	1 failed, 35 passed	scan row [bare]

Merge results

git merge-tree --write-tree of bcf6952e:

• origin/main d6071c7d (tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's squash): CONFLICTS in 5 files, all of them the squash's doing:

  • tests/test_adopt_git_reads.py;
  • tests/test_adopt_plan_escapes.py;
  • tests/test_adopt_template_values.py;
  • tests/test_non_utf8_git_reads.py;
  • tests/test_non_utf8_name_callers.py (add/add).

  It is the same against 75d70502 (repo_shape.py strips only YAML's space and tab, so a value a Unicode space opens reads as PyYAML reads it, and validate-manifest, bootstrap, bump-leg and update-shape refuse a branch holding a character that does not print or is white space (#715) #823, which touches none of tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869's files).

• tests: four tests, directory_named and the file twin ask the one helper whether a name that is not UTF-8 can be made and skip on its two answers only, and three tests skip where an ASCII file name encoding cannot carry their text (#757) #806's final head 22238b47: clean, tree 963aff8b.

• tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845 8f6a38e3: clean, tree 840b7794 (the head's own).

• tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's new head fc3cb5e9: clean, tree 963aff8b.

  • main is an ancestor of fc3cb5e9, so that merge then merges clean with d6071c7d (963aff8b) and with 75d70502 (268a2c01).
  • git diff fc3cb5e9 963aff8b carries exactly tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869's change: 8 files, +558/-115, the same changed lines as git diff 84a09c2e bcf6952e.
  • So the restack is one clean merge of tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's current head.

• family.py, adopt-project.py and the four tools' --org: init asks GitHub's limits on the topic, the --org and the holder's name before any repository exists, check finds what execute refuses from one list, and an --org GitHub gives no organisation (., a trailing hyphen, --) is refused where GitHub is asked (#863) #871 99afbc3d: clean, tree 368ca091.

• The other 32 open PRs against 963aff8b: 26 clean. The 6 conflicts are all inherited:

  • adopt-project.py: the clone plan and check make is claimed with the run's own mkdir and removed only while it is still that directory, and a --work-dir another user can rename in is refused first, so a swapped source-readonly no longer turns the removal onto another directory (#803) #843, adopt-project.py: execute asks about both leg repositories, on the disk or on GitHub, before it makes either, and check asks the same, so a code leg that already exists no longer leaves the spec leg made, and a --local-remote-dir that cannot hold them no longer ends in a traceback after plan ok (#798) #861 and family.py, adopt-project.py and the four tools' --org: init asks GitHub's limits on the topic, the --org and the holder's name before any repository exists, check finds what execute refuses from one list, and an --org GitHub gives no organisation (., a trailing hyphen, --) is refused where GitHub is asked (#863) #871 in adopt-project.py;
  • update-shape.py, family.py and bump-leg.py: the branch HEAD is on is read as the bytes git wrote, so a branch ending in U+00A0 is no longer read as its neighbour, and a name that is not UTF-8 is refused by name before anything is written (#765) #849 in tests/test_bump_leg_branch_and_push_line.py and tests/test_family_tracking_branch.py;
  • setup-project.py: a passed-through --dry-run makes none of the preflight's offers and --shape-ref is held to git's rule part by part, so a rehearsal no longer writes git's global config and a.lock/b is refused before the clone (#857) #873 in setup-project.py and tests/test_setup_project_typed_name.py;
  • each of those conflicts identically against main;
  • shape-doctor.py: the the way in rows say a checkout's own contracts/repository-naming.yaml that does not load, or raises, is that checkout's, an n/a environment row, where a tab line was FINDING the way in refused and a pattern that does not compile ended the doctor in a traceback; and the EXIT CODES list says what INVALID and manifest-kinds name (#707) #778 in tests/test_doctor_validator_verdicts.py, identically against tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's head (tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's known import-line conflict), and clean against main.

Scratch paths

• Clone: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b
  • private refs: refs/rv869b/pr<N> (33 heads), refs/rv869b/sim-m1, sim-m2, sim-m3, m2x<N>.
• Worktrees, all under /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/:
  • head: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-head (bcf6952e, clean)
  • previous head: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-prev (40531e19)
  • predicted restack: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-restack (tree 963aff8b)
  • mutants: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-mut-*, left mutated. RV869b-mut-wf holds a local-only scratch commit cf4bc0c1.
  • prototype rows: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-proto, /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-proto-m2b, /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-proto-m2c
• Logs: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-logs/
  • test runs: run1-head-utf8-11.log, run2-head-c-8.log, run3-restack-utf8-12.log, run4-restack-c-8.log, and mut-*.log
  • scan and merge results: scan-pr-heads.txt, scan-merged.txt, merge-m2-prs.txt, open-prs.json
  • NIT 2 prototype: test_rv869b_c_rows.py.txt
• Scripts:
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-run.sh
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-mutate.py
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-scan.py
  • /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-forms.py
• Stub gh: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/RV869b-stubbin/gh
• Basetemps: /workspace/projects/openRepoShape-worktrees/_lane3-scratch/RV869b-scratch/bt-RV869b-*

No Windows or macOS run was possible; those answers are emulated. Nothing was pushed or posted, no GitHub write was made, and no git config was changed. The scratch commits used env-var identities only.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-18T04:48:33Z) Fix round 2 (W869g, sonnet-xhigh) (2572a717, a fast-forward from bcf6952e), for RV869b's two optional NITs. The commit touches 3 test files (+72/-4). Nothing was merged.

• NIT 1, taken. windows_fsencode (tests/conftest.py:1187) now calls os.fspath(text) first. A new row, tests/test_non_utf8_name_helpers.py:93, asserts windows_fsencode(Path("caf\udcff")) == b"caf\xed\xb3\xbf". With os.fspath taken out, that row fails on AttributeError. RV869's m1 now fails on assert list(tmp_path.iterdir()) == [] and no longer on an AttributeError.
• NIT 2, taken. Six emulated rows are in tests/test_non_utf8_git_failures.py:299-347: flag-u2028, flag-u202e, flag-u2066, elector-accented, elector-zwnj and cafe-directory. Each sets os.fsencode = ascii_fsencode and make_source_repo to the failing stub, and asserts a skip on "file name encoding" with tmp_path empty. Under UTF-8, m2b fails 3 failed, 39 passed and m2c fails 1 failed, 41 passed. Two more mutants die too: the flag test asking about flag, and the elector test asking after make_source_repo. The body's "Left out" bullet now says the rows exist, and I corrected three other lines that became false.
• Runs. UTF-8 over the 7 windows_fsencode/ascii_fsencode files plus test_adopt_template_values.py and test_repo_hygiene.py: 411 passed in 111.76s. C locale over test_non_utf8_git_failures.py and test_adopt_template_values.py: 109 passed, 11 skipped in 44.32s.
• Not taken, as ruled. NIT 3 is left, and NIT 4 goes in the squash message.

Lane 3's own check, not a reviewer's verdict: the round is tests only and takes two optional NITs, so it carries no opus re-check of its own. Lane 3 read every line of the diff (3 files, +72/-4, no non-ASCII added) and the writer's runs and mutants. The gate that runs before #869 lands covers the whole PR on its predicted squash tree, after #845 lands and #869 restacks.

Lane: openreposhape-3 (openRepoShape-3)

[brettheap]

(lane openRepoShape-3, 2026-10-18T10:26:50Z) #869 is restacked onto main by lane 3's R869 (sonnet), now at ba2d7ab4. #845 landed as 0f69339, and GitHub moved the base to main when #845's branch was deleted.

• Two merges, no rebase.
  1. F845 fc3cb5e9: clean.
  2. S845 0f693395: three of tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's files took ours, tests/test_adopt_template_values.py, tests/test_non_utf8_name_callers.py and tests/test_non_utf8_oserror_sites.py. None of tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's seven files changed in that merge.
  • The merge of main added nothing, since main was S.
• tests/conftest.py is merged by hand, not taken as ours. It starts from S's text and applies tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869's own two hunks: windows_fsencode takes a path, and the new skip_where_the_environment_cannot_carry. Main's newer helpers stay, among them NOT_UTF_8_BRANCH, git_answer_bytes, held_by_git and on_a_branch_that_is_not_utf_8. Lane 3 saw this conflict first in a git merge-tree simulation.
• The reviewed delta is kept exactly. git diff origin/main ba2d7ab4 lists the 9 files of git diff 84a09c2e 2572a717, at +629/-118 on both, and the -U0 changed lines are identical (747 on each side). The squash tree against main is f326942e, the head's own tree.
• Tests at ba2d7ab4, under a stub gh:
  • the 8 changed test files plus tests/test_bootstrap_branch_bytes.py, which imports windows_fsencode: 403 passed;
  • tests: three tests ask the helper whether a name can be made, make_file skips on EILSEQ or a UnicodeError only and the U+202E directory has no guard, so a full disk or a read-only tmp_path fails the test, not SKIPPED (#834) #845's three files: 79 passed;
  • the trio: 185 passed.
• The C locale. Under LC_ALL=C LANG=C alone, Python 3.12 coerces to UTF-8, so nothing skips (78 passed). Under LC_ALL=C LANG=C PYTHONUTF8=0 PYTHONCOERCECLOCALE=0, the file name encoding is ASCII, and tests/test_adopt_template_values.py gives 71 passed and 7 skipped. Six skips are the template-value tests (Jos\u2028e, \u202eevil, ref \u2066x\u2069 y, Jos\xe9, Ali\u200cReza and caf\xe9). The seventh is tests: under a C locale, test_execute_refuses_a_local_source_directory_holding_a_bidi_control errors at mkdir, because its except OSError does not catch the UnicodeEncodeError that U+202E raises there, where it should skip #854's \u202ecaf.
• RV869b's READY stands for these lines. tests: six skips on a failed git call ask the system first, so a failed git call fails the test and an encoding that cannot spell U+202E skips the bidi test (#853, #854) #869 takes a lane-3 turn once tests is green on ba2d7ab4.

Lane: openreposhape-3 (openRepoShape-3)

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[sourcery-ai]

Sorry @brettheap, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 3 days and 8 hours by commenting @sourcery-ai review. Upgrade to get a review now.