[harleypig]

Summary

The tidy-ups from #44 that were still open on master, in three commits:

1. fix(bin)
   • set-submodule-branch: a submodule whose path is missing now makes the
     run fail. Before, it was skipped and the script still exited 0. The
     gitlink that the branch checkout moves is now committed together with
     .gitmodules, so nothing is left uncommitted.
   • mvsubmodule: the 2btested target is gone, because .vim/2btested
     no longer exists.
   • update-coc-extensions: it now adds the installed extensions to the
     .vimrc list and no longer rebuilds the list from scratch. Before, a
     machine that lacked an extension dropped it from the list for every
     machine.
2. fix(settings)
   • coc.vim: the <Tab>/<C-Space> branches for a completion plugin
     that is not installed are removed. CheckBackspace is script-local.
     The long <CR> mapping is wrapped.
   • help.vim: the autocmds are in an augroup.
   • maps.vim: <Leader>pp is now nnoremap. In Visual mode it raised
     E481: No range allowed. Two commented-out blocks with no stated
     reason are removed.
   • xdg-settings.vim: CleanOldFiles is script-local. 'spellfile' now
     points into the spell directory this file already created. Before,
     zg wrote into the checkout's .vim.
   • ftplugins: yaml.vim adds yamllint only once. isort's poetry option
     and shfmt's options are buffer-local.
3. fix(config)
   • languageserver.terraform no longer lists the tf filetype (TinyFugue
     in Vim; vim-polyglot gives every *.tf the terraform filetype).
   • docs/hashicorp.md says where terraform and terraform-ls actually come
     from.
   • .gitmodules now uses the same indentation on every line.
   • The fugitive-azure-devops link in docs/plugins-to-try.md points to
     its web page.

Depends on #150 for two wording changes. after/ftplugin/vim.vim now
says the early ale.vim sets g:ale_disable_lsp, and docs/json.md no
longer says the late ale.vim sets the JSONC fixer. Both statements become
true once #150 removes the .vimrc line and the late ale.vim fixer. Until
then, the vim.vim comment is still true, because the early ale.vim does
set the flag. The json.md change describes the state after #150.

Refs #44

#44 checklist

#	Item	Disposition
1	envsubstitute help missing closing quote	Done in #121 (script removed under #45)
2	Old shell-lint: SC2319 envsubstitute, SC1091 dump-maps, shfmt in check-setup/post-clone	Done earlier (#59 retired dump-maps, #121 removed envsubstitute, shfmt clean since #95). Re-checked: shfmt -d bin/ and shellcheck bin/* are clean
3	docs/proselint.md config path	Done in #121
4	envsubstitute help prints a FIXME	Done in #121 (script removed)
5	docs/about-docs.md line 152 over 78 columns	Already fixed on master. The only long lines left are table rows, which are exempt, and markdownlint passes
6	Late coc.vim long <CR> mapping	Done here
7	.vimrc line 6 long (URL)	Dropped: a URL cannot be broken, so the long line is allowed
8	addsubmodule substring duplicate check	Done in #122
9	update-coc-extensions malformed-JSON message	Done in #141 ("Cannot parse")
10	mvsubmodule 2btested target	Done here
10b	mvsubmodule "and updated readme" message	Done in #141
11	.gitmodules tab vs spaces	Done here
12	fugitive-azure-devops .git link	Done here
13	envsubstitute warn prints a literal '%s'	Done in #121 (script removed)
14	envsubstitute SC2319	Done in #121 (script removed)
15	post-clone pipx substring match	Done in #128
16	post-clone nvm install has no failure check	Done in #141
17	post-clone "did not complete" shown in green	Done in #128
18	envsubstitute unused debug()	Done in #121 (script removed)
19	g:ale_disable_lsp set twice	Other PR this round (#150)
20	g:ale_fixers.jsonc set twice	Other PR this round (#150)
21	yaml.vim appends yamllint again each time	Done here
22	Global isort_auto_poetry / shfmt_options	Done here (buffer-local; ALE reads both through ale#Var)
23	docs/ansible.md / post-clone say pipx ansible brings the linters	Done in #140 (#104). The page and post-clone now say it brings neither
24	vim.vim stray comment over "let b:ale_fixers	Done in #139
25	ALEVirtualText* highlights	Done in #143
26	.eslintrc.json	Done in #143
27	Spell directory created but 'spellfile' never set	Done here (made to work: 'spellfile' now points into it)
28	<leader>pp uses a recursive map in every mode	Done here
29	mvsubmodule/rmsubmodule messages and unexplained commented-out lines	Done in #141
30	update-coc-extensions ignores XDG_CONFIG_HOME	Done in #141
31	post-clone nodejs startup file, dpkg -l	Done in #141
32	update-coc-extensions.bats shfmt drift	Done in #141
33	maps.vim commented-out diagnostic-navigation and terminal-launcher blocks	Done here (removed. CoC diagnostics reach the quickfix list through ALE, and c[/c] and <leader>q[/q] move through it)
34	coc.vim branches for an uninstalled completion plugin	Done here
35	tests/vim/oldfiles.bats name and header	Other PR this round
36a	help.vim autocmds outside an augroup	Done here
36b	after/plugin/statusline.vim autocmds	Other PR this round
37a	CheckBackspace, CleanOldFiles global	Done here
37b	CustomFoldText global	Not done. tests/vim/oldfiles.bats asserts foldtext=CustomFoldText(), and another PR this round is changing that file. The fix is a one-line pair, s:CustomFoldText() in fold.vim plus that assertion, and should land after that PR
37c	ToggleLineNumbers global	Not done. It is in plugin/gutter.vim, which is outside this PR's files
38	coc-settings.json terraform tf filetype	Done here
39	Late ale.vim Markdown TODO	Other PR this round (#150)
40	mvsubmodule 2btested (repeat of 10)	Done here
41	set-submodule-branch exits 0 on a missing path, leaves the gitlink uncommitted	Done here
42	update-coc-extensions drops entries that are not installed	Done here
43	.claude/TESTS.md does not say where helpers live	Other PR this round
44	docs/hashicorp.md "via package manager"	Done here

Items 37b and 37c are still open, so this PR says Refs #44, not
Closes.

Decisions

• set-submodule-branch commits the gitlink. bin/ fixes: terraform install, submodule scripts, shared fixture, tidy-ups #141 left it for the user
  to commit. Round 2 of Small tidy-ups: envsubstitute help quote, old shell-lint warnings, proselint config path #44 asked for it to be committed. The commit names
  only .gitmodules and the submodule's path, so anything else that is
  staged stays uncommitted, as before.
• update-coc-extensions adds and never removes. The list says what
  every machine installs, so an extension missing from one machine is not a
  reason to drop it. Removing an extension is now done by hand, and the
  usage text and docs/scripts.md say so. The list is still sorted, so on
  the current .vimrc the first run moves @yaegassy/coc-ansible to the
  top, the same as the old script did. Only lines of the form
  \ 'name', are read. A commented-out entry is not taken as an extension.
• Spell directory: made to work, not removed. tests/vim/xdg.bats
  expects the directory, and leaving 'spellfile' empty sends zg into the
  checkout.
• maps.vim blocks: removed, not annotated. Diagnostic navigation is
  already covered by the quickfix mappings. The terminal-launcher block
  needed an external tool, and nothing referred to it.

Testing evaluation

Run by the author, to the qa-tester's standard (no team-managed: sentinel
here).

1. Verdict. Every behaviour change has a test. The comment and doc
   wording, the .gitmodules whitespace and the link change need none.
   The coc-settings.json filetype list has no test: the file is JSONC,
   and the suite does not test CoC configuration values (TESTS.md, grey
   area).
2. Automatable: yes. bats: tests/shell/set-submodule-branch.bats,
   mvsubmodule.bats, update-coc-extensions.bats and the new
   tests/vim/tidy.bats.
3. Falsifiers. Each test was seen red before its fix and green after:
   • set-submodule-branch "reports a missing submodule path, carries on,
     and fails": red while the missing path is just continued.
   • "commits the gitlink the checkout moved": red while the commit names
     only .gitmodules.
   • mvsubmodule "has no special 2btested target": red with the
     .vim/2btested branch.
   • update-coc-extensions "keeps a listed extension that is not installed
     here" and "adds the installed extensions…": red against the rebuild.
   • tidy.bats (all eight): red against master. That covers the
     uninstalled-plugin branches and the global CheckBackspace, a
     re-source that stacks the help autocmds, <Leader>pp in Visual mode,
     global CleanOldFiles, an empty 'spellfile', a duplicate yamllint,
     and the global isort/shfmt options. The backup-cleanup assertion was
     also made red on purpose by raising the age to 900 days, then
     restored.
4. Disposition: written in this PR.
5. Entry-point reach: the scripts were run as commands. The settings
   were exercised through Vim started on .vimrc.

Manual verification

1. Where: a checkout of this branch, with Vim started on its .vimrc.
2. What: open a YAML file and run :set ft=yaml, then :echo b:ale_linters, :echo &spellfile, :map <Leader>pp and :help help.
3. Success: ['yamllint'] once. A spellfile under
   $XDG_DATA_HOME/vim/spell/en.utf-8.add. <Leader>pp listed for n
   only. Help opens in a vertical split.
4. Failure: ['yamllint', 'yamllint'], an empty spellfile, or an
   entry for every mode.
5. Result (run headless with a throwaway HOME and XDG directories):
   • b:ale_linters: ['yamllint'].
   • spellfile: <tmp>/d/vim/spell/en.utf-8.add.
   • :map <Leader>pp: n \pp * :set paste!<CR>.
   • :help help: two windows, help filetype.
   • bin/mvsubmodule --help and bin/set-submodule-branch --help print
     the new usage text.
   • bin/update-coc-extensions on a copy of the real .vimrc, with only
     coc-json installed: exit 0, every entry kept, and the list sorted.

Test plan

• bash -c "trap '' PIPE; exec bats tests/shell": 100 pass.
• bats tests/vim: 83 pass.
• shfmt -d bin/ and shellcheck bin/*: clean.
• markdownlint on every tracked *.md: clean.

Review

opt-out: no named review skill ran. The diff is a set of small tidy-ups,
and none of them can block or refuse work; the author's own read of the
diff stands in its place.

Points worth a second look:

• the perl merge in update-coc-extensions;
• the -- .gitmodules "$path" commit in set-submodule-branch;
• the <SID> call in the <Tab> <expr> mapping, also covered by the
  existing coc.bats test that inserts a Tab at the start of a line.

Completion report

1. Functionality works: yes. The suites and the manual run above show it.
2. Access model: not applicable (a personal editor configuration).
3. Permissions: not applicable.
4. Audit logging: not applicable.
5. Validation: not applicable. No new input is read.
6. Errors handled: yes. set-submodule-branch now fails on a missing path,
   and update-coc-extensions keeps its existing failure messages.
7. Testing complete: yes, see Testing evaluation.
8. Documentation: docs/scripts.md, docs/hashicorp.md, docs/python.md,
   docs/json.md and learning.txt are updated. docs/coc-ale-setup.md
   still describes the removed <Tab>/<C-Space> branches. It belongs to
   another PR this round, so it is not edited here.
9. AI Completion Report: this section.

Co-Authored-By: AI Agent noreply@harleypig.com