Multiple fixes for broken master by line0 · Pull Request #26 · TypesettingTools/DependencyControl

line0 · 2026-05-23T19:05:03Z

This PR fixes several issues, largely caused by the incomplete refactoring some 10 years ago:

SemanticVersioning: this class was missing entirely and has been ported over from the sqlite branch
ModuleLoader: Fixed a regression where __depCtrlInit hooks were called again on - already-initialized modules, causing errors in modules that mutate their exported state on - first call (e.g., BadMutex).,
UpdateFeed: Fixed a regression where legacy boolean true/false script type arguments were - no longer accepted after the ScriptType enum migration.,
UpdateFeed: Fixed a regression where DownloadManager was instantiated too late, causing the - update process to fail.
Common: The capitalize() function no longer uses unsupported string indexing and now works - for the first time ever. Before the refactor, this was previously masked by it not actually - being called anywhere.
Updater: Fixed a potential issue where a multi-assignment statement could corrupt record - fields after an unsuccessful update.

Additionally, i have added a unit test suite for DepCtrl (only 1 tiny test for now) and made all files use LF + EOF Newline to ensure consistent file hashes.

Copilot

Pull request overview

This PR aims to restore a working “master” by reintroducing missing version-handling functionality, fixing several regressions in module loading and update/feed handling, and adding initial unit-test and line-ending consistency infrastructure.

Changes:

Added SemanticVersioning and migrated version parsing/formatting to use it across updater/feed paths.
Fixed/adjusted updater + module-loader behaviors related to initialization and update flows, and added a minimal DepCtrl unit test.
Updated the project update feed metadata, bumped versions, and enforced LF + final newline consistency.

Reviewed changes

Copilot reviewed 12 out of 15 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
modules/DependencyControl/Updater.moon	Migrates updater version handling to `SemanticVersioning` and adjusts update logging/changelog display.
modules/DependencyControl/UpdateFeed.moon	Adds legacy scriptType compatibility + new invalid scriptType messaging; moves `DownloadManager` init earlier.
modules/DependencyControl/UnitTestSuite.moon	No functional change (newline normalization).
modules/DependencyControl/Tests.moon	Adds a first unit test for `Common.terms.capitalize`.
modules/DependencyControl/SemanticVersioning.moon	Introduces semantic version parsing/formatting/comparison utilities.
modules/DependencyControl/Record.moon	Switches record version parsing to `SemanticVersioning` and adds compatibility helpers.
modules/DependencyControl/ModuleLoader.moon	Updates init-hook guard logic and switches version formatting/parsing to `SemanticVersioning`.
modules/DependencyControl/FileOps.moon	No functional change (newline normalization).
modules/DependencyControl/ConfigHandler.moon	No functional change (newline normalization).
modules/DependencyControl/Common.moon	Fixes `capitalize()` implementation to avoid unsupported string indexing.
modules/DependencyControl.moon	Bumps DepCtrl version, exposes `SemanticVersioning`, and adds a Moonscript minimum-version guard.
macros/l0.DependencyControl.Toolbox.moon	Bumps toolbox version; updates version formatting and script-type handling; registers DepCtrl test suite.
DependencyControl.json	Updates feed versions/hashes and adds new files (including tests) to the module feed.
.vscode/settings.json	Enforces final newline on save in VS Code.
.gitattributes	Enforces LF line endings in git.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot

Pull request overview

Copilot reviewed 12 out of 15 changed files in this pull request and generated 5 comments.

…alized modules

… function

Copilot

Pull request overview

Copilot reviewed 12 out of 15 changed files in this pull request and generated 5 comments.

+        maxVer = SemanticVersioning\toNumber @version
+        minVer = SemanticVersioning\toNumber minVer

        changelog = {}
        for ver, entry in pairs @changelog
-            ver = DependencyControl\parseVersion ver
-            verStr = DependencyControl\getVersionString ver
+            ver = SemanticVersioning\toNumber ver
+            verStr = SemanticVersioning\toString ver
            if ver >= minVer and ver <= maxVer


number segments over 255 and non-dot separate values

…r string is passed

…is virtual

nothing inside DepCtrl is supplying this parameter, so the only effect this bug would have had, is removing the default guard against modules trying to load themselves

- assert now passes through all varargs on success (allows chaining) - new assertNotNil method - fails only on nil, unlike assert which also fails on false - dump/dumpToString methods accept a maxDepth to cap recursive table output - new static Logger.describeType method - returns Lua type name but renders MoonScript class instances as "ClassName object" - fix: guard log/format calls against non-string message templates

…e up-to-date path require's up-to-date-but-not-updated branch called ModuleLoader.loadModule with task.record.namespace as the module-spec argument. loadModule is a fat-arrow static, so the dot-call binds that string as `mdl` and immediately does `with mdl` -> indexing a string -> crash. Pass task.record as the spec, matching the other load sites, and assert the call arguments in the test. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

performUpdate gated its unmanaged-record handling on @record.updateRecordType, a field that exists nowhere (the field is recordType). The check was always false, so an unmanaged module never got its dummy ref + version registered before dependency loading, and finish never removed the dummy ref on failure. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

An empty or separator-only segment made the flatten callback return nil, which left a hole in the flattened array; the ipairs walk then stopped at the hole, so joinPath("a", "", "b") returned "a". Return an empty array for such segments so they contribute nothing without breaking iteration. Also wires the three pre-existing joinPath_segments* tests into _order (they were defined but never ran) and adds an empty-segment regression. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

remove passed its (nil) recurse straight to rmdir, whose own default is true, so `remove somePath` on a path that was unexpectedly a directory silently deleted the whole tree — the opposite of the documented default. Default remove's recurse to false; rmdir keeps its own true default for the callers that rely on it (e.g. temp-dir cleanup). A directory passed without recurse now fails cleanly. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The BFS seeded every config-derived feed and fetched it unconditionally; the block/policy gate applied only to advertised children. So a blocked feed that was also a configured root (an extraFeed or a package-declared feed) got fetched on every crawl, and untrusted roots were fetched even under fetchUntrustedFeeds = never. Gate each root through feedTrust.shouldFetch, exactly as children are: a blocked root is never fetched and an untrusted root honors the policy, while the root still appears in the inventory as recorded-but-not-followed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

io.open's result was used unchecked, so a missing log dir, a permissions error, or a full disk turned every file-logging call into an 'index a nil value' crash — and nothing created ?user/log on a fresh Aegisub profile. Create the log dir best-effort before opening, and on open failure disable file logging (noting it once in the window) instead of crashing. The message still reaches the window. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…d of crashing getChangelog ran each changelog key through SemanticVersioning.toNumber, which returns false (not nil) for an invalid version; toString(false) and false >= minVer then raised. Changelog keys come from unvalidated feed data, so one bad key crashed getChangelog — called right after a successful install. Skip keys that don't parse. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

A key whose value cycled back to an ancestor made _equals return true for the whole comparison, so two cyclic tables were declared equal even when a sibling key differed (e.g. {x=1, self=self} vs {x=2, self=self}) — and the early return also skipped the depth decrement, corrupting later sibling comparisons. Treat a cyclic key as matched and continue checking the remaining keys instead. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The item tracker was a boolean set, so a value appearing more than once was recorded once and matched once: itemsEqual({1,1}, {1,1}) returned false, and assertItemsEqual reported spurious failures on arrays with repeated scalars. Track occurrences as a multiset and match by decrementing the count. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…e dir The temp download path was built with table.concat and no separator, gluing the downloadPath directory onto the file name so the file landed in ?temp/ itself rather than the ..._feedCache subdir that trimFiles scans — so temp feed downloads accumulated there unbounded. Build the path with FileOps.joinPath so it lands inside downloadPath and the maxFiles-20 trim actually applies. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

@get

@get scanned @@handlers comparing the raw filePath, but entries are keyed by the validated path, so an unnormalized spelling (a path token, mixed separators, a /./ segment) missed the cache and built a second handler for the same file — leaving two live handlers with divergent in-memory config. Validate first, then do a direct lookup by the canonical path (also replacing the linear scan). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

setFile reassigned @__configHandler but never added the view to the handler's view set, so the handler's whole-file refreshes (view\refresh! for view in views) skipped it — after another view triggered a load that replaced the hive tables, the setFile'd view pointed at a dead table and its writes were lost. Register the view with the new handler and detach it from the old; apply the same to unsetFile. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…stance GC Collecting one handle to a named semaphore called sem_unlink immediately, so an ephemeral handle being GC'd while another holder still owned the name removed the name from the kernel — a later sem_open then created a *separate* semaphore (value 1), letting a fresh handle acquire the 'same' lock while the first holder still held it (split-brain). Register names for unlink and defer it to a single module-level canary that runs at Lua-state teardown (≈ process exit), anchored on the class so it isn't collected — and fire the unlink — early. Windows is unaffected (kernel refcount self-heals). Verified on POSIX (WSL): the added regression test reproduces the split-brain on the old code and passes on the fix. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…itively block stored opts.matchMode verbatim (a bogus value only coerced to prefix later, on read) and deduped by a case-sensitive URL compare while matchesBlockEntry matches case-insensitively — so https://Bad/ and https://bad/ stored as two equivalent blocks. Validate the mode at the mutation boundary (unknown -> prefix) and compare URLs case-insensitively when deduping. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

__loadOfficial cached its result even when DepCtrl's own feed failed to load, so an offline first run stuck the official trust/block lists at the minimal fallback for the whole session — never picking them up even after the feed became reachable (e.g. once the updater fetched it into the cache). Cache only on a successful load; a failure returns the fallback uncached so a later call retries. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

@Private

- Enum.describe: doc param order matched to the signature (values, pattern, join) — a doc-following caller passed a separator where a formatter function goes. - Updater.scheduleUpdate: return is always an UpdateStatus (never boolean), plus the ProtectedInstall entryPath second value now documented. - scriptType annotations: integer -> ScriptType across ScriptTargetFilter, ScriptUpdateRecord and UpdateFeed (left over from the string-Enum migration). - ConfigView.getSectionHandler: doc says the hive path is file-root-relative (every caller relies on that), not relative to the view. - UpdateTask.refreshRecord: drop the ---@Private (it's called cross-class from Updater) and document it. - AGENTS.md: ScriptType/RecordType are Enums now, not bare tables. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…grate old configs @depConf.scriptFields still listed the pre-0.7 boolean `unmanaged` (with a lone -- REMOVE marker) and omitted `recordType`, so an unmanaged record's type was never written to config — it round-tripped back as managed. Persist/import `recordType`, and migrate a pre-0.7 config that stored only `unmanaged` so those records don't silently become managed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

- Per-test durations used os.clock (CPU time) but are reported as ms wall-time, so tests that sleep or wait (downloads, timers) under-reported. Use the monotonic Timer clock instead.

…asing it to the caller (releaseLock), or it's held until orphanTimeout.

The one-time config migration now rewrites each record's legacy boolean `unmanaged` flag into its recordType (and drops the flag), rather than Record.loadConfig re-deriving recordType from the stale flag on every load. Because migration runs once and persists, the obsolete key is cleared from the config file instead of lingering indefinitely. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…ction to skip tests;

Mechanical rename of the module, its test suite, and all ~180 references ahead of making the class instantiable as a version value. No behavior change. Safe because the class was only split out during the (unreleased) 0.7.0 refactor, so there is no external API to keep compatible. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

… bumping Construct a version from a string or numeric components (or via fromPacked); instances compare with </<=/==, stringify, bump immutably, and satisfy ranges. The static helpers accept an instance anywhere they take a version, and the static formerly named toNumber is renamed toPacked to match the instance method. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

A new primitive for declaring get/set computed properties on a class: Accessors.property{get, set} marks a property in the class body and Accessors.install(class) — called once after the body — rewrites the instance metatable so reads/writes of those keys dispatch to the getter/ setter, falling through to normal method/field lookup otherwise. A setter-less property is read-only (a write raises); a subclass that runs install inherits its parent's properties; every property is recorded on class.__accessors for inspection tooling. Centralizes the metatable manipulation in one audited place. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

A record's canonical version is now a SemanticVersion value object stored as @semanticVersion, with record.version a packed-integer accessor over it, so existing packed-int reads/writes/comparisons are unchanged. The version is persisted to the config as a "major.minor.patch" string (readable at a glance), migrated from the old packed-integer form on first load; @semanticVersion is initialized up front so a record exposed mid-construction never reads a nil. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

install adds a __pairs metamethod so a readable computed property appears in pairs(instance) with its getter value, like a raw field (under 52compat). Since version then shows up as a packed int while the config stores a string, it's kept out of Record's generic scriptFields copy and serialized explicitly. Inline comments across the rework are tightened per the guidelines. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Contribution conventions for DependencyControl: annotation/comment style, file naming, MoonScript gotchas, the public-API and test-exposure patterns, computed properties via Accessors, module-private markers, and changelog rules. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The "build macro and module lists" and "create and run the update tasks" comments only restated the immediately following buildDlgList / buildInstalledDlgList / runUpdaterTask calls, which the variable names already convey. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Convert getProgress -> progress (Downloader), getState -> state (Lock), getFailures -> failures (UnitTestSuite) and isPrivate (Host) from methods into read-only computed properties via the Accessors module, updating their call sites: the DownloadManager shim, the depctrl CLI runner, and the tests. Host.isPrivate memoizes into a backing field so repeated reads (and pairs) stay cheap after the one-time resolution. Add LockState/LockScope LuaCATS aliases so the enum-typed field and returns carry named types instead of a bare integer/string with a prose hint, and refresh the now-stale getFailures example in AGENTS.md. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

fix: add missing SemanticVersioning module

cd03290

line0 requested a review from Copilot May 23, 2026 19:05

Copilot started reviewing on behalf of line0 May 23, 2026 19:05 View session

Copilot AI reviewed May 23, 2026

View reviewed changes

line0 force-pushed the add-missing-semver-module branch 2 times, most recently from 7f7ad69 to 9fee44a Compare May 23, 2026 22:10

line0 requested a review from Copilot May 23, 2026 22:10

Copilot started reviewing on behalf of line0 May 23, 2026 22:11 View session

Copilot AI reviewed May 23, 2026

View reviewed changes

line0 added 8 commits May 24, 2026 00:22

fix: broken script type handling in update feeds

f147a45

fix: module loader running depctrl initializer hooks on already initi…

bf66595

…alized modules

fix: update failing due to unsupported string index in capitalization…

f2e1fa8

… function

test: add first test

977f37a

ci: enforce LF + EOF newline for consistent hash calculation

75b6c43

build: bump version; update feed

7f2530a

fix: typo in ModuleLoader breaking optional module presence checks

ff8495a

refactor: tighten version number validation in updater

d0cbeee

line0 force-pushed the add-missing-semver-module branch from 9fee44a to ad4a5ec Compare May 23, 2026 22:46

line0 requested a review from Copilot May 23, 2026 22:49

Copilot started reviewing on behalf of line0 May 23, 2026 22:49 View session

Copilot AI reviewed May 23, 2026

View reviewed changes

line0 added 4 commits May 24, 2026 00:58

fix: broken version number display in version errors

d16fd64

fix: typos

167d427

fix: semver parsers allowing invalid values

a1fc3f8

number segments over 255 and non-dot separate values

fix: semantic version toString roundtrip exception when invalid semve…

5bc3f45

…r string is passed

line0 force-pushed the add-missing-semver-module branch from ad4a5ec to 5bc3f45 Compare May 23, 2026 23:14

line0 added 4 commits May 24, 2026 08:14

fix: update errors missing information about whether or not a record …

c08a79c

…is virtual

fix: broken skip lists in module loader

80ead7c

nothing inside DepCtrl is supplying this parameter, so the only effect this bug would have had, is removing the default guard against modules trying to load themselves

fix: nil dereference crash in feed template variable expansion

5a582f0

line0 and others added 28 commits July 11, 2026 15:31

fix(UnitTestSuite): measure test durations with a wall clock

5408c77

- Per-test durations used os.clock (CPU time) but are reported as ms wall-time, so tests that sleep or wait (downloads, timers) under-reported. Use the monotonic Timer clock instead.

doc(UpdateTask): run acquires the global updater lock and leaves rele…

8ec2b81

…asing it to the caller (releaseLock), or it's held until orphanTimeout.

refactor(UnitTestSuite): manual _order is no longer required; add fun…

b0717cf

…ction to skip tests;

line0 force-pushed the add-missing-semver-module branch from cceac44 to 9c245f0 Compare July 11, 2026 13:34

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Multiple fixes for broken master#26

Multiple fixes for broken master#26
line0 wants to merge 122 commits into
masterfrom
add-missing-semver-module

line0 commented May 23, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

line0 commented May 23, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants