Add MiniMax as a new LLM provider using the Anthropic-compatible API.
Changes:
- Add MiniMax to LLMProvider type and MiniMaxConfig interface
- Add MiniMax chat model creation via ChatAnthropic with custom base URL
- Add MiniMax settings persistence and model list (MiniMax-M2.5, MiniMax-M2.5-highspeed)
- Add MiniMax provider UI in SettingsPanel with API key and model selection
1. Replace fragile regex in buildExportedTypeMapFromGraph with
extractReturnTypeName() for consistent generic unwrapping.
2. Fix gap detection precision: check upstream.has(binding.exportedName)
instead of exportedTypeMap.has(binding.sourcePath) to avoid
over-counting files that import only unrelated symbols.
All 4 test blocks now share one DB lifecycle to avoid cross-block
"Database is closed" errors caused by LadybugDB's shared global DB
in a single vitest fork. Staleness detection (which triggers closeLbug)
runs last to avoid invalidating connections for other blocks.
11/11 tests pass on macOS, Ubuntu, and Windows.
Critical fixes:
- Re-resolution pass now actually re-resolves CALLS edges by calling
processCalls with importedBindingsMap (was building typeEnv but
discarding it without producing edges)
- Worker path populates ExportedTypeMap via buildExportedTypeMapFromGraph
using graph node isExported + SymbolTable returnType/declaredType
(was dead parameter in processCallsFromExtracted)
Important fixes:
- Skip threshold denominator uses totalFiles (was exportedTypeMap.size +
filesWithGaps which made threshold nearly useless)
- processCalls accepts importedBindingsMap parameter to thread cross-file
bindings into buildTypeEnv during re-resolution
All 3454 tests pass.
Add ExportedTypeMap infrastructure to propagate resolved type bindings
across file boundaries. When file A exports `const user = getUser()`
(resolved to `User`), file B importing `user` now gets seeded with
`user → User`, enabling `user.save()` to produce CALLS edges.
Key components:
- `importedBindings` option on BuildTypeEnvOptions with scopeEnv seeding
AFTER walk() to respect first-writer-wins (local declarations win)
- `collectExportedBindings()` in call-processor using graph node
isExported flag (no SymbolDefinition changes needed)
- Inline Kahn's algorithm topological sort with level grouping for
parallel-safe file ordering and cycle detection
- Re-resolution pass in pipeline.ts: topological order, 3% skip
threshold, path validation, per-file export caps (500)
- 32 new tests: 11 topological sort, 6 seeding, 15 integration
(simple cross-file, re-export chain, circular imports)
All 3454 tests pass (32 net new, 0 regressions).
Address code review findings from PR #392 senior compiler review:
- Fix Java "Yes" → "No" in optional-param-arity matrix (Java has no defaults)
- Simplify Kotlin hasDefaultValue while-as-if to direct const/if check
- Update OPTIONAL_PARAM_TYPES comment to include Ruby
- Replace per-declaration Set allocation with size-based Map iteration skip
- Add 11 unit tests for multi-declarator type association and constructorTypeMap
Add requiredParameterCount to SymbolDefinition and MethodSignature,
enabling range-based arity filtering in filterCallableCandidates.
Calls with omitted optional/default arguments now resolve correctly.
Supported: TS, Python, Kotlin, C#, C++, PHP, Ruby (7 languages).
Detection via OPTIONAL_PARAM_TYPES set + hasDefaultValue helper.
9 integration tests added across all 7 languages.
- Watchdog timer now exempts in-flight queries via activeQueryCount,
preventing premature stdout restoration during long queries (>1s)
- Stale detection uses reinitPromises Map to prevent TOCTOU race where
concurrent callers double-close the connection pool
- Throttle meta.json staleness checks to once per 5s per repo
- Add null guard for i.id in IN-clause construction
- Enrichment queries run in parallel on non-arm64 platforms to preserve
performance; sequential only on arm64 macOS where SIGSEGV occurs
- Updated AGENTS.md and CLAUDE.md to reflect new indexing metrics.
- Enhanced call-processor.ts to support cross-file inheritance tracking and improved virtual dispatch resolution.
- Added support for TypeScript overload signatures in tree-sitter queries.
- Improved type extraction for C++, C#, and Kotlin to handle smart pointers and constructor types.
- Introduced inferLiteralType for overload disambiguation across multiple languages.
- Added tests for C++ smart pointer dispatch and Kotlin virtual dispatch scenarios.
- Updated type-resolution-roadmap.md to reflect completion of phases P.1 to P.3 and outline future work on covariant return types.
- Add AbortSignal.timeout(30s) on all fetch calls
- Add retry with backoff for 429/5xx (core: 2 retries, MCP: 1 retry)
- Guard initEmbedder() and getEmbedder() to throw in HTTP mode
- Discard cached embeddings on dimension mismatch during incremental re-index
- Add MCP embedQuery retry for transient failures
- Add 16 unit tests covering both core and MCP HTTP paths
- Fix README: concise, accurate env var docs
Fixes three related issues that cause SIGSEGV crashes and stale data:
1. Impact enrichment queries (Promise.all → sequential await)
The impact() method ran 3 enrichment queries concurrently via
Promise.all against the same LadybugDB connection pool. On arm64
macOS, concurrent native DB access triggers SIGSEGV. Changed to
sequential await. Also caps IN-clause to 100 IDs to prevent
oversized queries. (#285, #290, #292)
2. Silence stdout during query execution
silenceStdout()/restoreStdout() only wrapped createConnection() and
initLbug(). Now also wraps executeQuery() and executeParameterized()
to prevent native stdout writes from corrupting the MCP stdio
stream during all DB operations. (#285)
3. Stale data after re-index
ensureInitialized() checked pool existence but never verified whether
the underlying index was rebuilt. Now reads meta.json's indexedAt
timestamp on each call and closes/re-opens the pool when the index
has changed. (#297)
- schema.ts: FLOAT[${EMBEDDING_DIMS}] reads from GITNEXUS_EMBEDDING_DIMS env
- embedding-pipeline.ts: vector search CAST uses actual query vector length
- mcp/core/embedder.ts: HTTP embedding support for MCP query-time search
Without this, using a 1024d model (e.g. bge-large) fails with
'Expected: 384, Actual: 1024' on LadybugDB vector insert.
Adds support for OpenAI-compatible embedding endpoints as an alternative
to the local transformers.js pipeline. Enables using self-hosted servers
(Infinity, vLLM, TEI, llama.cpp) over Tailscale/VPN, or any cloud
endpoint — with higher-quality models like bge-large-en-v1.5 (1024d).
Configuration via environment variables:
GITNEXUS_EMBEDDING_URL=http://your-server:8080/v1
GITNEXUS_EMBEDDING_MODEL=BAAI/bge-large-en-v1.5
GITNEXUS_EMBEDDING_API_KEY=your-key (default: 'unused')
GITNEXUS_EMBEDDING_DIMS=1024 (auto-detected if omitted)
When env vars are set:
- initEmbedder() skips local model download entirely
- embedText() and embedBatch() call the HTTP endpoint
- Dimensions auto-detected from first response
- Batches in groups of 64
When env vars are NOT set:
- Existing local transformers.js behavior is completely unchanged
Build: tsc clean
Tests: 1776 passed, 0 failed
Add constructorTypeMap to buildTypeEnv — populated during walk when a
declaration has both a type annotation and a constructor initializer.
Add isSubclassOf helper (BFS, depth-5, cycle-safe).
In call-processor, consult constructorTypeMap to override receiver type
when constructor creates a known subclass (same-file only).
Add inferLiteralType to LanguageTypeConfig for Java, Kotlin, C#, C++.
In resolveCallTarget, when multiple candidates survive arity filtering,
lazily infer argument literal types and filter by parameterTypes match.
Worker path falls through gracefully (no AST available).
Add parameterTypes?: string[] to SymbolDefinition and MethodSignature.
Extract per-parameter type names via extractSimpleTypeName during
parsing for overload disambiguation (Java, Kotlin, C#, C++).
Thread through both sequential (parsing-processor) and worker
(parse-worker) paths.
The fileIndex Map stored SymbolDefinition per name, silently dropping
earlier overloads via Map.set(). Changed to SymbolDefinition[] so all
same-name methods (e.g., Java overloads) survive in same-file resolution.
Added lookupExactAll() for resolution-context to pass all same-file
candidates through to candidate filtering.
- Remove dead replayPendingItems array and inert if-block in type-env.ts
- Add nullable_type fallback in extractKotlinDeclaration for val x: User? local vars
- Tighten isCSharpNullableDecl to avoid substring false positives on type names
- Add missing scope boundaries: function_expression (TS), constructor_declaration/
local_function_statement/lambda_expression (C#) in null-check narrowing walkers
- Extend null-check narrowing fixtures and add 4 integration tests covering:
Kotlin local variable nullable, C# constructor + lambda, TS function expression
Adds 17 new fixture directories and 23 new describe blocks covering every
feature in Milestone D (Phases A, B, C) with full cross-language integration
test coverage:
Phase A — Fixpoint Completeness:
- TS/JS object destructuring (const { field } = obj → fieldAccess resolution)
- TS/JS post-fixpoint for-loop replay (iterable var resolved by fixpoint)
- Rust struct_pattern destructuring (let Point { x, y } = p)
Phase B — Inheritance & Receivers:
- Grandparent MRO (depth-2 C→B→A) for all 9 OOP languages:
TS, Kotlin, C#, C++, Java, PHP, Python, Ruby, JS
- Go inc/dec write access (obj.Field++/-- emit ACCESSES write edges)
Phase C — Branch-Sensitive Narrowing:
- Null-check narrowing for TS (!==null, !=null, !==undefined),
C# (!=null, is not null), and Kotlin (!=null)
Bug fix — Kotlin null-check narrowing (3 issues in jvm.ts):
1. patternBindingNodeTypes registered 'comparison_expression' but
tree-sitter-kotlin produces 'equality_expression' for !=
2. Handler checked for 'null_literal' named child but 'null' is an
anonymous node in the Kotlin grammar
3. extractKotlinParameter only searched for 'user_type' direct child,
missing 'nullable_type' wrapper (so x: User? never got a base binding)
17 fixtures, 23 describe blocks, 705 new lines of test code, 0 failures.
Review follow-ups from compiler front-end review (#379):
- Rust extractPendingAssignment now calls unwrapAwait() on value before
type checks, so `let user = get_user().await` resolves correctly
- type-resolution-system.md: removed "no fixpoint inference" from
limitations, updated "Single-pass" to "Walk + fixpoint", replaced
stale single-pass Tier 2 description with fixpoint loop explanation
- type-resolution-roadmap.md: Phase 9 body updated — 9C is delivered,
9B walk-order dependency documented (for-loop Tier 0b runs before
fixpoint, so fixpoint-resolved types can't update loop variables)
- Added this/self/$this fixpoint gap footnote to feature matrix
Replace the sequential Tier 2b/2a propagation with a unified fixpoint
loop that handles four binding kinds: callResult, copy, fieldAccess,
and methodCallResult. The loop iterates until no new bindings are
produced (max 10 iterations), enabling arbitrary-depth mixed chains:
const user = getUser(); // callResult → User
const addr = user.address; // fieldAccess → Address
const city = addr.getCity(); // methodCallResult → City
city.save(); // resolves to City#save
Infrastructure:
- PendingAssignment union extended with fieldAccess and methodCallResult
- resolveFieldType helper: typeName → class nodeId → lookupFieldByOwner
- resolveMethodReturnType helper: typeName → class nodeId → lookupFuzzyCallable filtered by ownerId
- Fixpoint also resolves reverse-order copy chains that single-pass missed
Languages: TS, JS, Java, Kotlin, C#, Go, Rust, Python, PHP, Ruby, C++.
Each gets field access and/or method-call-with-receiver detection in
extractPendingAssignment, plus method-chain-binding test fixtures.
Activate the dormant Tier 2b pendingCallResults infrastructure in
type-env.ts by extending each language's extractPendingAssignment to
emit { kind: 'callResult', lhs, callee } when the RHS of an untyped
variable declaration is a simple function call.
This enables `var user = getUser(); user.save()` to resolve at TypeEnv
build time. Tier 2b now runs before Tier 2a copy-propagation, enabling
mixed chains like `const user = getUser(); const alias = user;
alias.save()`.
Languages: TS, JS, Java, Kotlin, C#, Go, Rust, Python, PHP, Ruby, C++.
Swift excluded. Each language gets a call-result-binding test fixture
and integration tests.
Conservative: only simple calls (no method calls with receivers), only
when exactly one callable matches, first-writer-wins.
* feat: upgrade @ladybugdb/core to 0.15.2 and remove segfault workarounds
The upstream fix (ladybug-nodejs#1) resolves the child QueryResult lifetime
segfault, making .close() safe on all platforms. This removes 6 workaround
sites:
- Remove `dangerouslyIgnoreUnhandledErrors` from vitest config
- Remove platform-conditional .close() guards in global-setup and test helper
- Delete test/setup.ts (process._getActiveHandles unref hack)
- Replace no-op cleanup in test-indexed-db.ts with real adapter close
- Fix pool adapter closeOne() to properly close connections with shared
Database refcount guard and orphaned connection handling in checkin()
- Update segfault-related comments across the codebase
Also bumps @ladybugdb/wasm-core to ^0.15.2 in gitnexus-web for consistency.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: keep dangerouslyIgnoreUnhandledErrors for macOS N-API exit crash
The N-API destructor ordering crash during worker fork exit on macOS is
independent of the QueryResult lifetime fix in 0.15.2. Tests pass, but
the exit triggers a crash. Keep the flag with an updated comment
explaining the actual cause. Can be removed once LadybugDB fixes all
destructor ordering issues upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* ci: unify test run for single-pass coverage
- Update `npm test` to run all tests (unit + integration + lbug-db)
via `vitest run` instead of `vitest run test/unit`
- Add `test:unit` script for running unit tests only
- Remove `ci-integration.yml` — the per-file lbug-db process isolation
is no longer needed with `dangerouslyIgnoreUnhandledErrors` and
`fileParallelism: false` handling fork exit issues
- Update `ci-unit-tests.yml` to run all tests with build + coverage
- Simplify `ci.yml` gate (two jobs: quality + tests)
- Simplify `ci-report.yml` (single coverage artifact, no merge step)
* fix: update cli-commands test for renamed test:all → test:unit script
* fix: set USERPROFILE in setup-skills test for Windows compatibility
os.homedir() checks USERPROFILE on Windows, not HOME.
* fix: add isolate: false to lbug-db project to prevent fork crashes
On macOS, N-API destructors crash fork workers on exit. With
isolate: true (default), vitest recycles the fork between files,
triggering the crash after each file. After several crashes, the
remaining lbug-db files never execute.
isolate: false keeps all 8 lbug-db files in a single fork — the
fork only exits once after all files complete, and that single exit
crash is caught by dangerouslyIgnoreUnhandledErrors.
* fix: add unique sequence.groupOrder to vitest projects
Vitest v4 requires unique groupOrder when projects have different
maxWorkers (lbug-db has fileParallelism: false → maxWorkers: 1).
* fix: await async close() in global-setup and remove isolate: false
global-setup.ts called conn.close() and db.close() without await —
these return Promise<void> in @ladybugdb/core 0.15.2. The setup
function returned before the DB was fully closed, so vitest forks
hit a stale file lock when opening the same DB path, crashing the
lbug-db worker before any test ran.
isolate: false caused native state corruption after 2-3 open/close
cycles in the same fork (vitest-specific, not reproducible in plain
Node.js). Without it, each file gets its own module scope and the
N-API destructor crash at fork exit is caught by
dangerouslyIgnoreUnhandledErrors.
Also fixes fire-and-forget close() calls in the pool adapter —
try/catch around an async close() never catches rejections; changed
to .catch(() => {}) for proper unhandled-rejection prevention.
Before: 0/8 lbug-db files ran on macOS CI (fork crash).
After: 8/8 pass, 84 files, 3077 tests, zero errors.
* fix: update project index references in AGENTS.md and CLAUDE.md to reflect correct symbol counts and relationships
* feat: enhance lbug adapter with external database support and write operation validation
* feat: create ci-tests workflow for comprehensive test coverage across platforms
* ci: move PR report inline to ci.yml, delete ci-report.yml
The old ci-report.yml used workflow_run which always runs code from
the default branch (main). This meant the PR comment used main's
stale report template that still referenced the old unit/integration
split architecture — causing "Merge coverage reports" failures.
Moving the report inline to ci.yml means it runs from the PR branch
and uses the current report template. The report now shows:
- per-platform status (Ubuntu/Windows/macOS columns)
- unified test counts from the single vitest run
- coverage with base branch (main) delta comparison
- commit SHA for traceability
Also removes the save-pr-meta job since the report no longer needs
a separate workflow_run trigger.
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Gergo Magyar <gergomagyar@icloud.com>
