bar-fmt: lua codemod, workspace integration, LLM type-triage (rollup) - #57
Draft
keithharvey wants to merge 34 commits into
Draft
bar-fmt: lua codemod, workspace integration, LLM type-triage (rollup)#57keithharvey wants to merge 34 commits into
keithharvey wants to merge 34 commits into
Conversation
bar-lua-codemod parses BAR Lua with emmylua_parser (the same front-end family as emmylua_check/emmylua_ls) at Lua51, doc-comment analysis off. Transforms collect byte-range edits over the rowan CST and splice them into the original source, so untouched bytes are untouched by construction. cst.rs carries the shared helpers: parse gate on SyntaxError-kind diagnostics only, bracketed-string-key extraction that tolerates trivia, and the function-declaration-name exclusion. codemod-cargo.sh runs cargo inside bar-dev, where the toolchain lives. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
x["y"] -> x.y and ["y"] = -> y = for string keys that are valid, non-reserved identifiers. Raw quoted content only — escapes and long strings never convert. Injects a space when ] abutted a word character so .identifier doesn't merge with what follows. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Spring.GetMyTeamID / GetMyAllyTeamID / GetMyPlayerID -> the canonical GetLocal* names, wherever the prefix is the bare Spring global. Bracket access and non-Spring prefixes are left alone. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Spring.I18N / Utilities / Debug / Lava / GetModOptionsCopy — BAR's own helpers riding the Spring table — move to the BAR namespace. Handles the _G.Spring.Module spelling too; function-declaration names stay put. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
bar::check / bar::check-errors run emmylua_check inside bar-dev against BAR's .emmyrc.json; check-errors re-derives its exit code from the error count so warnings don't fail CI. Adds the devtools .emmyrc.json, points the VS Code template at the exported emmylua wrappers, extracts Sim/Units/Scripts docs in lua::library, and makes distrobox a hard setup requirement — the whole toolchain lives in the container. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
bar::codemod-build / codemod-test wrap codemod-cargo.sh. Migrations get their own bar::migrate:: namespace — adoption-dated entries listed in order, retired to [private] ~6 months after adoption. The 2026-07 stylua-cleanup entry replays the transform set (bracket-to-dot, rename-aliases, detach-bar-modules) plus stylua onto the local BAR checkout — the idempotent 'transform your branch, then merge' path the README callout documents. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The codemod scanned everything stylua ignores: .lux and
recoil-lua-library tool dirs, and mapgenerator — whose
mapinfo_template.lua is a ${}-placeholder template, not valid Lua.
Old runs never hit it because a local skip-worktree bit kept the file
off disk; upstream touched the file, checkout rewrote the index entry,
the bit dropped, and the template resurfaced as a parse failure.
Explicit excludes replace the hidden state. No output change: the
baseline transform chain has zero conversions under mapgenerator.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
llm-type-triage.sh chunks emmylua_check errors and dispatches parallel LLM workers (claude --print, or the openai worker via BACKEND=openai — hence python3-pip in bar-dev) that fix annotations per file. The codemod-prereq SKILL.md is the categorization rulebook: error pattern -> idempotent fix recipe, first match wins. claude/claude.md carries the cross-repo agent context. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
generate-branches.sh deterministically rebuilds fmt, the mig-* leaves, the mig rollup, and the fmt-llm capstone (env layer + gen(llm) commit), with PR-body generation, the museum table, and stacked-PR base overrides. bar::migrate::stylua-cleanup-generate fronts it ([private] — generation stays out of the listing), wrapping execution in $CREDENTIALS_RUNNER so the triage workers inherit LLM-backend credentials; the tracking-issue template ties the PR stack together. Generated PR bodies now point at the migrate:: names, so the next --update-prs run rewrites the published bodies to match. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Same discovery as the replay recipe: mapgenerator/mapinfo_template.lua
(a ${}-placeholder template, not valid Lua) was kept off disk by a
local skip-worktree bit, which upstream's touch of the file silently
dropped — resurfacing it as a parse failure that aborts generation.
CODEMOD_EXCLUDES mirrors .styluaignore (vendored utils, .lux,
recoil-lua-library, mapgenerator) in the three transform runners and
their PR-body command docs. Verified: both the full_moon baseline
binary and the emmylua binary report 2,166 scanned / 0 errors with the
skip set — the baseline run's exact scan count.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
link_gh_stack used plain gh, which the bar-dev container doesn't have — the extension check always failed inside distrobox and stack linking silently skipped. Route it through resolve_host_gh/host_exec like every other gh call (the 916e302 hardening missed this one). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GitHub's native stacked PRs reject fork-headed PRs, and the whole fmt -> mig -> fmt-llm-source -> fmt-llm chain was fork-headed. The branches already live on the canonical repo, so the chain PRs are recreated same-repo (#8395/#8396/#8397/#8398) and the fork originals (#7199/#7229/#7447/#8235) closed with cross-links. Constants follow. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
No fork-headed PRs anywhere — native stacked PRs reject them and reviewers live upstream. The fork keeps mirror pushes only. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Completes the no-fork-PRs convention: bracket/rename/detach leaves and the two carried leaves recreated with heads on beyond-all-reason (#8401/#8402/#8403/#8404/#8405); fork originals (#7287/#7288/#7289/ #7437/#7438) closed with cross-links. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GitHub rejects --base on PRs in a native stack even as a no-op — the stack owns the base. edit_pr checks the current base and only sends --base when it needs to move. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Chain links move to #8395–#8398 (the fmt-llm line still pointed at the pre-#8235 capstone). The tooling line links the three-layer devtools stack with org beyond-all-reason#17 as the entry point — keithharvey has no push access to beyond-all-reason/BAR-Devtools, so the devtools stack hosts on the fork and beyond-all-reason#17 stays the org-visible umbrella. DEVTOOLS_PR points at the generation layer (3/3). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 23, 2026
keithharvey
force-pushed
the
bar-fmt-03-llm
branch
from
July 23, 2026 23:31
2110efe to
a542849
Compare
… follows the harness bootstrap
The hash was recorded outside the commit branch, so a transform whose content arrived via a prereq cherry-pick filed the previous commit's SHA under its own name — three of six entries were wrong. Match subjects against origin/master..HEAD instead, and warn when the count is short. BLAME_TRANSFORMS narrows the file to mechanical rewrites: the gen(hand) pair is a hand restructure and a pure addition, where --ignore-revs has nothing earlier to attribute to.
The LLM pass shipped gui_chat.lua in a state that would not load: a partial rename, a dropped table field, a stripped namespace prefix, and fifteen lines of shadowing recomputation the analyzer cannot see.
The old hook ran stylua over the whole repo and never re-staged, so the unformatted version committed anyway while ~30 unrelated files were left dirty. It checked nothing. Now it checks only the staged .lua and refuses, filtering .styluaignore itself because stylua honours it only for discovered paths, not for ones passed explicitly. setup-hooks refuses to install while .stylua.toml and .gitattributes disagree about line endings — a --check hook would reject every correctly-formatted file.
From review on #8398: the pass commented out a ${} placeholder to make a
parse error go away, and repaired a nil set-entry to an object that
happened to be in scope rather than to the true the read sites imply.
Checking working-tree files ties the gate to .gitattributes eol=crlf vs stylua's line_endings, and those disagree. Staged content is always LF, so checking the blob is both what git will record and immune to the question. Drops the config-mismatch guard, which is now moot.
keithharvey
commented
Jul 29, 2026
| @@ -0,0 +1,185 @@ | |||
| # Type Triage Worker Prompt (OpenAI variant) | |||
Contributor
Author
There was a problem hiding this comment.
I do think this file is good to include, mostly as a blueprint for how to do this sort of subagent/orchestrator script in the future. This PR does new things so having the some sub-agent stuff in claude/ provides helpful rails for future folks trying the same thing.
keithharvey
commented
Jul 29, 2026
| @@ -0,0 +1,1084 @@ | |||
| --- | |||
Contributor
Author
There was a problem hiding this comment.
This is the skill file so subagents bail if they don't hit a known rule while walking through EmmyLua errors.
git_bar routes git through host_exec, so a commit that creates files writes them on the host while stylua reads them from inside bar-dev. The container sees the dirents but not the contents, and the pass dies with "failed to read ./types/Debug.lua: No such file or directory" — intermittently, and only for the one prereq that adds files. Resolves a host stylua the same way resolve_host_gh does, with an in-container fallback that warns.
target/ is unanchored so it covers bar-mission-kit as well as bar-lua-codemod, and the packaged mission-kit server binary is ignored by path — it is a build output, not a source artifact.
keithharvey
force-pushed
the
bar-fmt-03-llm
branch
from
July 29, 2026 21:38
a542849 to
64d3fb4
Compare
A branch pushed at the same tip as its base reads to GitHub as merged: it closes the PR, credits the pusher, and deletes the branch. A merged PR cannot be reopened, so #8403 is gone for good. It happens when a run dies between 'checkout -B <leaf> fmt' and the transform commit and the empty leaf is pushed anyway. Refuse instead.
Tooling stack is one PR now, not three. Adds the mission-kit PR as its own checklist item, the pre-commit hook to the command list, and .git-blame-ignore-revs to what the stack delivers — with the caveat that git blame aborts on branches predating it, which contributors will hit before the capstone lands. Error count tracks the measured baseline and says so, since it moves with master.
9 tasks
The rebase logic was never BAR-specific. It now lives in repos.just taking the repo as an argument, so two copies would be two ways to do one thing.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The bar-fmt migration as one PR (19 commits at the stack tip): bar-lua-codemod tooling and tests, workspace integration (check recipes, editor wiring,
migrate::stylua-cleanup, the genericbar::restackstack rebaser), the LLM type-triage layer with the generation capstone, and the migrate:: moves that host every pipeline branch and PR on the canonical repo (#8395–#8410).