family.py add: six states its case-spelling remedies still miss after #828, from a section that names a variant beside a correct gitlink to a Members that is a link to a directory outside the holder #840

Goal: where add refuses a member recorded in another case, no printed command leaves a section of .gitmodules naming another spelling, moves a live checkout out from under its .git file, or fails through a link, in the six states the spelling remedies were not built for.

Filed by lane openRepoShape-5, from the checklist candidates of #761's PR (#828, writer w761: candidates 3 to 5, its "Left out, and why") and of #685's PR (#767, reviewer rv685, round 3: "Already in the body's 'Left out'", candidates 1 to 3; #767 landed as 98f14976). #761 itself and #685 are not reopened. The evidence of each line is said on it: two were typed by a lane 5 writer or reviewer at a named head, the others are read from source, not measured, and none was run again here.

Lands after: lane 5's #761 (PR #828, fix/761-add-move-back-gitmodules, head d4e8c65a), which edits the same two functions, _members_spellings (scripts/family.py:3187-3338) and _spellings_refusal (:3369). path:line is at main 6626a3c5, which holds #767 and not #828.

Lines

• A section that names a case variant of a row's path, with the gitlink at the row's path, is not named by add, which goes through (w761's candidate; typed at bbb961f8: after family.py add: where another member's own gitlink sits at a case variant (Members/<P> for its row members/<P>), the refusal's remedy prints git rm -r --cached or git mv, which takes that member's gitlink away from its row #646's old move back, the gitlink is at members/IRRS and the section still says Members/IRRS; the gate exits 1 with gitmodules-path-unnamed in every mode and add exits 0). family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828 stops its own move back from leaving that state; a hand edit of .gitmodules, or a holder whose earlier family.py add: where another member's own gitlink sits at a case variant (Members/<P> for its row members/<P>), the refusal's remedy prints git rm -r --cached or git mv, which takes that member's gitlink away from its row #646 move back is already committed, still has it. _members_spellings reads index and HEAD entries and the disk (:3187-3338), not what _submodule_paths (:3586) reads from .gitmodules, so no remedy is built for it. The gate's own line is git config -f .gitmodules submodule.<name>.path <row path> and git add .gitmodules (_section_lines). Direction: add refuses it before it commits, with the gate's two lines and a commit of .gitmodules, since add would otherwise commit that .gitmodules beside the new member (_commit_one). Say which refusal code it is, so the gate and add give it the same.
• family.py: add refuses, before the member is read and on every disk, any path under another spelling of members/, in its index, in HEAD or on disk, and names a remedy per entry that never moves the members' checkouts (#362, #363, #364) #574's git mv of a directory also moves an untracked checkout under it (w761's candidate; not measured). RENAME is git mv for what the index holds where it is on disk in its own spelling (:2740-2744, _on_disk_as_named :2943), and git mv of a directory moves the work tree's whole directory, untracked files included. A live submodule checkout made by hand (git submodule add) under it holds a .git file naming its store by a relative path, which a move leaves pointing nowhere: the fault validate-family.py: a staged git rm is named again and fails retyped, the member-gitlink-absent remedy ends in a refused add, a shape-pin row with no path: reads None, and #639's mv orphans a live checkout and is stated by OS #681's N-1 measured for a plain mv, validate-family.py and validate-pins.py: a staged gitlink removal says to commit it, member-gitlink-absent puts the gitlink back by hand, the variant mv is guarded, and a pin row with no path: is shape-pin-row (#681) #747 fixed in the validator's text, and family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828 fixes for add's closing mv. Direction: measure it first: a tracked Members/<P>/ holding a file, with an untracked checkout of another member under it, then the printed git mv typed as printed and git -C of the moved checkout. If it breaks, name the checkout before the git mv, as family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828's CHECKOUT remedy does for one found on disk.
• A case variant of another row's path under members/ itself is not named by add (w761's candidate; read from source, not measured): members/irrs beside the row members/IRRS. w761 did not run it, and it is not re-read here. _members_spellings finds a case variant of the mount path of the member being added (project) and what sits under another spelling of members/; whether a variant of an existing row's path is among them is the question. Direction: read first which of the two a person meets, with a probe holder; if add goes through, name it as family-members-dir-variant names a variant of the mount path.
• An unmerged .gitmodules that the work tree also lacks, beside a gitlink (rv685's round-3 candidate 1; read from source, not measured). _checked_gitmodules_as_staged leaves a missing copy to git for add (:6544, "add LEAVES A MISSING FILE TO GIT"), so it raises nothing there, and the spelling refusal says nothing of the file even though _gitmodules_beside (:3341) answers GITMODULES_UNMERGED from the index alone. Direction: probe it (the rename committed, .gitmodules conflicted in the index and deleted from disk), and have the refusal name the file's state first, as it does for an unmerged copy that is on disk (gitmodules-unstaged).
• A deletion of .gitmodules the holder already staged is committed by any printed commit (rv685's round-3 candidate 2, "as on the base. Nothing here stages it"; read from source, not measured). family.py add: a members/ renamed with git mv is renamed back in two steps through a free name, after what the disk lacks under it is put back, and a printed --sparse says the git it needs, so no remedy fails as typed (#685) #767 puts back a copy deleted from disk before its move back (git checkout -- .gitmodules), and family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828 stages .gitmodules where a section is pointed back; a deletion already staged, with HEAD holding the file, goes in with the commit the remedy prints and takes every member's section out of HEAD. _gitmodules_beside (:3341-3361) answers empty where the index holds no .gitmodules (if not records), which is also the state of a staged deletion with HEAD holding the file, and remove refuses a staged deletion that HEAD still holds (scripts/family.py:6667-6671: "HEAD holds one and its deletion is staged", then git checkout HEAD -- .gitmodules), while _checked_gitmodules_as_staged leaves add to git submodule add's own refusal to write a .gitmodules the work tree lacks, which a printed spelling remedy does not pass through. Direction: name the staged deletion first in the spelling refusal, where a gitlink is beside it, and put the file back with git checkout HEAD -- .gitmodules (scripts/family.py:6671 prints it for remove), as family.py add: a members/ renamed with git mv is renamed back in two steps through a free name, after what the disk lacks under it is put back, and a printed --sparse says the git it needs, so no remedy fails as typed (#685) #767 does for a copy deleted from disk but staged by nobody.
• Members a link to a directory outside the holder: the move back's git rm --cached fails through the link (rv685's round-3 candidate 3; measured by rv685 at bbb961f8, in its third-round table of typed rows: the typed git rm exits 128 not a git repository: Members/IRRS/../../.git/modules/members/IRRS; the gate exits 1 in all three modes, add exits 2). The other link state in that row, members -> Members, is family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828's (h) and is not here. family.py add: a members/ renamed with git mv is renamed back in two steps through a free name, after what the disk lacks under it is put back, and a printed --sparse says the git it needs, so no remedy fails as typed (#685) #767 withholds the rename back where the spelling is a link (_renamed_with_git, :3089) and falls back to family.py add: where another member's own gitlink sits at a case variant (Members/<P> for its row members/<P>), the refusal's remedy prints git rm -r --cached or git mv, which takes that member's gitlink away from its row #646's move back, which has no such check. Direction: name the link first, with the words member-path-unaddable has for members itself (:3654-3690), before any line that goes through it. Prove it on the probe holder at family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828's head, since family.py add: the move back of a member's gitlink points its section of .gitmodules back, sets aside first what the index holds in its way and puts back a member whose removal is staged, its closing mv is guarded and leaves a checkout to delete by hand, and --sparse follows the sparse rules, so no printed line leaves a section naming another spelling or un-registers a member (#761) #828's set-aside part may already answer it.

Verify: tests/test_family_add_spellings_remedies.py (#767; edited by #828), tests/test_family_add_member_variant.py, tests/test_family_members_case_variants.py, tests/test_family_add_refusals.py and tests/test_family_add_move_back_sections.py (new in #828) pass on Linux. One new case per line, typed as printed to the gate in every mode and then to the add that was refused (typed_as_printed, assert_passes_in_every_mode), as #767's and #828's files do; the lines that say "probe first" start with the probe, and a line that the probe shows already answered is closed as such on this issue. The folding disk is emulated by disk_as: CI is Linux only (#597), and no macOS or Windows run is made.

Overlaps. #828 edits scripts/family.py and the three test files above, so this lands after it. #731, #795, #809, #813, #814 and #816 edit scripts/family.py in other functions; #816 (fix/684-family-holder-states) lists the case-variant git mv remedy among its lines, so read its head before starting. scripts/family.py is not a pinned copy, so no upstream-changed row follows; the validator halves are in the validator's issue (#762's PR, #820).

Parent: #459.

Milestone: v1.1.1. Each state needs a hand git mv, a hand-edited .gitmodules, a conflict, or a link where a directory goes, and add or the gate ends in a refusal in all six rather than recording a wrong state. None is a regression and none blocks v1.1.0.

🤖 Generated with Claude Code