Skip to content

Releases

1.1.0 — Chinese ranking, dispatch confidence, faster impact

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 08:31
d2c710a

Local Codebase Indexer 1.1.0 improves Chinese search, separates guessed program paths from real ones, and speeds up impact analysis.

Improvements

  • Chinese (CJK) queries rank much better in hybrid search. The symbol-name and vector layers are scaled by the share of non-CJK characters in the query, so a Chinese query is ranked by the keyword layer with synonym expansion. On 20 Chinese navigation questions in a 3.4k-file C# project, MRR went from 0.48 to 0.84; English benchmark suites are unchanged.
  • Shared-pipeline fan-out is uncertain. In data-access programs and impact, a path that reaches a method only through an interface or base method with several implementations is marked uncertain (interface dispatch, one of N implementations). Use data-access --confidence probable to keep only programs that do not depend on a runtime choice.
  • Faster and more faithful impact analysis. Historical snapshots start from a copy of the current index and use the project's .ai_context/ignore, so only files that differ are parsed and the base revision honours the same exclusions. First runs were 2-2.5x faster on a 3.4k-file project.

Upgrade notes

  • No index format change.
  • Programs that were probable only because of multi-implementation dispatch are now uncertain.

See CHANGELOG.

1.0.4 — MCP initialize no longer waits for the index

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 07:06
5463dca

Local Codebase Indexer 1.0.4 keeps MCP clients from timing out while the index is being opened.

Fixes

  • The MCP server answers initialize right away. initialize, ping and tools/list no longer open the index; the first tools/call does. When an upgrade adds indexes to a large existing database, building them on a cold cache could exceed a client's connection timeout (Claude Code gives up after 30 s). A failure to open the index is now a tool error, and the next call retries.

Upgrade notes

  • No index format change. A missing project root still fails at startup.

See CHANGELOG.

1.0.3 — Faster incremental updates

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 06:50
8de57ce

Local Codebase Indexer 1.0.3 makes incremental updates and doctor faster on large projects.

Performance

  • Faster incremental updates. Resolving EF context types looked up method text and qualified names without an index, once per EF access in the whole project, on every update with changes. New indexes make those lookups direct, and file dependencies read references with a table scan instead of a per-row lookup. On a 3.4k-file project an update of 28 changed files went from 73 s to 23 s, with identical results.
  • doctor no longer reopens unchanged indexed files.

Upgrade notes

  • No index format change. Two small indexes are created automatically the first time an existing index is opened.

See CHANGELOG.

1.0.2 — Faster first calls on large projects

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 05:43
c73d2db

Local Codebase Indexer 1.0.2 makes the first call after a server starts much faster on large projects.

Performance

  • Scanning no longer reopens indexed files. A file already indexed with an unchanged modification time is known to be text, so the binary check is skipped; new and modified files are still checked. On a 3.4k-file project this removes about 3,400 file opens from every status call and pre-query refresh.
  • status consistency checks read indexes instead of whole tables. Rowid sets are compared with EXCEPT over covering indexes, the full-text side uses the FTS5 docsize shadow table, and orphaned reference paths are found by index seeks. On a 420 MB index one check pass reads about 65 MB instead of about 460 MB. Corruption outside an indexing run is still detected.

Upgrade notes

  • No index format change. A small idx_vss_rowid index is created automatically the first time an existing index is opened (about 15 s on a 42k-chunk index).

See CHANGELOG.

1.0.1 — Pipeline dispatch through base classes and callee tables in impact

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 04:24
32123b9

Local Codebase Indexer 1.0.1 fixes two gaps in program tracing and impact analysis.

Fixes

  • Follow pipelines through abstract base classes. When an interface is implemented by an abstract base class, overrides in its subclasses are now reached from callers of the interface or base method (override dispatch). Data-access programs and impact no longer list such steps under no_known_caller.
  • Impact lists tables touched by called methods. tables now include accesses reached through confirmed / probable calls from the changed code (up to 4 hops), each with depth and via_chain. A callee_depth_limit coverage warning marks a cutoff.

Upgrade notes

  • No index format change; existing indexes keep working.
  • Dispatch through a shared interface cannot tell which implementation a caller selects at runtime, so every caller of that interface is reported as a probable program for each implementation.

See CHANGELOG.

1.0.0 — SQL evidence, stable MCP contracts, and project setup

Choose a tag to compare

@sfy257000 sfy257000 released this 20 Nov 01:27
acc6609

Local Codebase Indexer 1.0.0 brings scoped SQL tracing, explainable dependency paths, stable MCP contracts, and simpler project setup.

Highlights

  • Keep SQL objects separate across servers, databases and schemas. Qualified names, explicit USE context, EF schema mappings and quoted procedure names retain their identities. Ambiguous short names return candidates.
  • Follow the evidence. Trace SQL access through C# methods, controller actions and Razor pages with source locations, confidence labels and visible traversal limits. Export offline HTML/SVG, Mermaid or JSON reports.
  • Set up your client from the CLI. init --client claude and init --client codex create project-local MCP configuration while preserving unrelated settings. Indexing progress is available with index --progress.
  • Consume stable structured results. The seven MCP tools retain their text responses and add versioned JSON schemas, protocol negotiation, pagination, bounded summaries and retryable busy errors.
  • Choose retrieval for your task. Lexical search skips vectors; a bounded vector cache helps long-running processes. Fixes cover same-line C# methods/overloads and Unicode SQL diagnostics.
  • Read a shorter README. English, Traditional Chinese and Simplified Chinese documentation now focus on setup, commands and reproducible tool comparisons. Detailed evaluation records remain in the technical documentation.

Upgrade

python -m pip install --upgrade local-codebase-indexer

Index format 11 automatically rebuilds older parsed indexes on the first update. Allow time for that refresh; symbol IDs and cursors are tied to the index revision. Existing source files are unaffected.

Static analysis remains approximate. Check confidence labels and coverage warnings; unsupported dispatch and dynamic SQL can require source verification. The vector cache defaults to 128 MiB and can be disabled with INDEXER_VECTOR_CACHE_MB=0.

Validation

240 tests pass, with CI covering Linux and Windows on Python 3.10–3.13, installed packages, type/lint checks, MCP contracts and Claude client transport. Frozen retrieval benchmarks retain gains and losses; no hosted-model accuracy, token-cost or production-runtime guarantee is implied.

Changelog · CLI reference · Measured results and limitations

0.9.1 — SQL provenance fixes and reproducible tool comparisons

Choose a tag to compare

@sfy257000 sfy257000 released this 13 Nov 06:41
1535f89

This patch release corrects SQL access attribution and adds reproducible comparisons explaining when Local Codebase Indexer helps with SQL Server / C# maintenance.

Correctness fixes

  • Limit sp_executesql analysis to its SQL argument and statement, including quoted procedure names. Unrelated literals, parameter values, and similarly named columns no longer create false accesses.
  • Scope CTE names and UPDATE/DELETE aliases to their consuming statement, including omitted semicolons and UNION / INSERT SELECT patterns.
  • Recognize GO only as a batch separator line, preserving comma-joined sources after predicates such as a.go = b.id.
  • Keep SQL object definitions and suggestions restricted to SQL source files; application functions retain normal symbol navigation.
  • Respect C# / VB6 parameter and local shadowing, invalidate overwritten SQL bindings, and avoid treating query-parameter strings as executed SQL. Preserve earlier valid execution evidence.

Reproducible tool comparisons

English, Traditional Chinese, and Simplified Chinese READMEs now include measured results, trade-offs, and links to all raw query records. A new optional runner compares the unchanged retrieval questions with actual ripgrep candidate search plus standard BM25 ranking.

Retrieval suite rg + BM25 file Recall@5 Indexer Difference
MES-style synthetic, 42 questions 81.0% 91.7% +10.7 percentage points
Same questions with distractors 81.0% 90.5% +9.5 percentage points
Python standard library, 12 questions 100.0% 79.2% -20.8 percentage points

The indexer retains its default 5 result chunks; the baseline returns 5 files. Both are scored at file level. These are 54 unique project-authored questions and 96 executions including the repeated noisy suite, not independent production-accuracy measurements.

  • In the separate bundled synthetic navigation task, required evidence-file coverage increases from 3/5 to 5/5 (60% to 100%) versus one literal table-name search. Controller reachability remains probable, not runtime proof.
  • On noisy/stdlib retrieval suites, rendered output uses 94.9% / 96.9% fewer Unicode characters than full reads of the same selected files. Tiny synthetic files produce 2.3% more characters. These are not measured Claude-token savings or reductions against Grep snippets.
  • Richer navigation costs more time: recorded warm CLI medians are 244.5 ms versus 8.3 ms for a literal rg call, with 266.1 ms for the initial indexer call including indexing. Timings are machine-specific.
  • Serena and Claude Context features are documented with pinned upstream sources; no end-to-end Claude, Serena, or Claude Context superiority claim is made.

See the protocol, recorded summary, and all 96 query records. Recorded measurements retain their pre-release package-version label and source fingerprint.

Upgrade

python -m pip install --upgrade local-codebase-indexer==0.9.1

Index format increases from 9 to 10. The first index update automatically rebuilds older parsed data; existing projects do not need to delete their index manually.

Validation

187 tests pass, including 33 SQL/binding regressions and 7 benchmark-protocol tests added since 0.9.0. CI covers Linux and Windows on Python 3.10, 3.12, and 3.13, plus lint/type checks, installed-package validation, project evaluation, and comparison smoke runs. Retrieval questions, gold answers, and thresholds were not changed.

Changes: #4 and #5. Full changelog.

0.9.0 — SQL correctness, VB6 data access, and schema impact

Choose a tag to compare

@sfy257000 sfy257000 released this 10 Nov 03:40
120529c

This release fixes missing and incorrect SQL data-access answers, adds conservative VB6 SQL support, and makes schema changes visible without mixing them with row writes.

python -m pip install --upgrade local-codebase-indexer
local-codebase-indexer demo

SQL correctness

  • Resolve UPDATE/DELETE aliases named a, an, or the; support parameterized TOP and UPDATE table hints.
  • Preserve bracketed/double-quoted names, spaces, escaped delimiters, quoted dots, and linked-server qualifiers.
  • Recognize legacy NOLOCK table hints and comma sources following JOIN predicates.
  • Exclude cursor FETCH and scalar TRIM operands from table reads; mask nested block comments correctly.
  • Analyze literal SQL inside OPENQUERY as uncertain, preserving the need to verify remote execution.
  • Attribute constant C# SQL to each executing method, without sharing same-named constants across unrelated classes.

Legacy sources and schema impact

  • BOM-tagged UTF-16/UTF-32 work in indexing, source context, and diagnostics. Actual binary files no longer remain permanently pending as new files.
  • VB6 .frm, .bas, and .cls SQL strings have enclosing method owners. Complete SQL at known ADO/DAO receivers can be confirmed as a static call site; unknown receivers and incomplete expressions remain probable/uncertain. VB6 symbols do not become SQL functions or C# call targets.
  • Table/index DDL appears under schema_changes with confidence filtering and pagination. Git impact includes file-level SQL, before/after evidence, and correct GO batch boundaries.
  • Explicit negations in .ai_context/ignore can include source kept under bin/packages/build/dist. VCS and index-state directories remain excluded.

Maintenance and compatibility

  • Fix first-run Windows lock-file contention and serialize index ignore-file creation.

  • Preserve the seven MCP tools, CLI commands, and public imports while separating relation, context, and rendering code from the query facade.

  • Resolve the existing 32 mypy errors and add Ruff/mypy CI and release checks.

  • Test Linux and Windows on Python 3.10, 3.12, and 3.13.

  • Use pyproject.toml as the dependency source and AGENTS.md as the canonical assistant instructions.

Upgrade note: index format 9 triggers one full rebuild on the first update after upgrading. Subsequent updates remain incremental. Restart existing MCP server processes after upgrading to load the new parser. VB6 support is static string analysis, and schema tracking does not prove migration safety or resolve every runtime SQL pattern.

Validation

147 tests, including 38 new regression cases; installed-wheel demo and Unicode/VB6 smoke checks; extracted source-distribution tests; Ruff/mypy; package/version/checksum validation; and official MCP manifest validation. The three search benchmark Recall@5 results remain approximately 0.92 / 0.91 / 0.62.

See the support and upgrade notes and changelog.

0.8.0 — SQL Server to C# navigation, first-run demo, and MIT license

Choose a tag to compare

@sfy257000 sfy257000 released this 09 Nov 10:14
59bd057

Trace a SQL Server table to the C# code and pages that reach it, starting with a database-free example.

Try the new first-run demo

python -m pip install --upgrade local-codebase-indexer
local-codebase-indexer demo

The bundled synthetic source tree connects a table, stored procedure, repository, service interface, controller actions, and a Razor page. The demo checks four expected relationships and reports source locations, confidence labels, exact-reference counts, and local timings. Use demo --json for the complete report or demo --output demo-orders to keep a copy in a new or empty directory. Existing files are never overwritten.

What's included

  • MIT license with SPDX package metadata and the license file in distribution artifacts.
  • English, Traditional Chinese, and Simplified Chinese READMEs focused on the SQL Server / C# maintenance workflow, with a three-step first run and a 32-second actual-output walkthrough.
  • A reproducible, source-backed case study and raw JSON/text evidence. Detailed configuration and limitations remain in docs/reference*.md.
  • Contribution guidance and issue forms for missing or incorrect navigation results.
  • Automatic official MCP Registry publication after PyPI succeeds, using GitHub OIDC and a pinned, signature-verified official publisher.

The example is synthetic. Its four checks do not establish production accuracy, runtime behavior, or measured developer time savings. Interface-based controller reachability is labeled probable. Dynamic SQL and some C# dispatch patterns still require source verification.

Validation

109 tests passed locally and against extracted source-distribution tests using the installed wheel. The isolated installed-wheel demo passes all four checks and includes all sample files and MIT metadata. Build, twine, release-version/checksum validation, official MCP manifest validation, and workflow lint pass. The 15-question project search evaluation remains Recall@5 0.533 and MRR 0.328.

See the case study and full changelog.

0.7.0 — Stable navigation, explainable search, and project evaluation

Choose a tag to compare

@sfy257000 sfy257000 released this 09 Nov 09:37
aa69655

0.7.0: Stable navigation, explainable search, and project evaluation

This release makes navigation more consistent across pages, Git revisions, model changes, and multiple local processes. It also adds inspectable retrieval results, project-owned quality evaluations, and automated release attachments.

Fixed

  • Search ranking no longer changes with page size or offset. Continuation cursors are bound to the query, filters, synonym configuration, embedding model, and completed index revision.
  • Explicit Git ranges analyze both requested snapshots instead of mixing in symbols from HEAD or a dirty working tree.
  • ONNX identities fingerprint model files, tokenizer, external weights, and inference settings. Changes to loaded artifacts are rejected so that incompatible vectors are not silently mixed.
  • Index initialization and writes are coordinated across processes. Readers use consistent SQLite snapshots; contention produces bounded, actionable errors, and interrupted batches force recovery.
  • Context clipping keeps focus lines and reports omitted or individually clipped lines.
  • Project synonyms reload on subsequent queries, including file replacement or removal. Invalid settings warn once per content version and recover after correction.

Added

  • search --explain / MCP explain=true: matched query terms, expansion provenance, retrieval-layer ranks and contributions, plus SQL evidence for routed data-access hits.
  • search --cursor <next_cursor> / MCP cursor: stable continuation with explicit rejection when the index or settings change.
  • evaluate --queries <file>: per-question answers, Recall@k, hit rate, MRR, p50/p95 latency, estimated output tokens, baseline comparisons, and enforceable quality/performance budgets.
  • A curated set of 15 navigation questions about this repository, with CI-generated evaluation reports.
  • A Release pipeline that tests and builds distributions once, validates versions, generates SHA256SUMS, and reuses the same artifacts for PyPI and GitHub attachments. Upload retries skip identical files and refuse to overwrite conflicting assets.
  • Expanded cross-platform CI and updated English, Traditional Chinese, and Simplified Chinese documentation.

Compatibility and usage

  • An invalid or stale search cursor requires a new search; legacy offsets remain supported without revision checks.
  • CLI exit status 4 indicates index contention. INDEXER_LOCK_TIMEOUT and INDEXER_BUSY_TIMEOUT configure bounded waits.
  • New ONNX identities trigger a full re-index on the first indexing run after upgrading. Use a dedicated model directory and restart the server after changing model artifacts.
  • In historical comparisons, snapshot=after refers to the target commit. The revisions object identifies the compared commits.
  • Evaluation tokens are estimated from rendered character counts, not a model-specific tokenizer. Empty datasets fail, and breached budgets return exit status 3.

Validation and known limitations

  • The local Linux/Python 3.12 suite and source-distribution suite pass 105 tests, including ONNX, multi-process, historical-range, pagination, and release-integrity regressions.
  • Installed-wheel smoke checks, distribution metadata checks, and actionlint validation passed.
  • GitHub CI passed on Linux/Python 3.10, 3.12, 3.13 and Windows/Python 3.12, including distribution installation and the navigation report job.
  • The 15 curated repository-navigation questions achieved Recall@5 0.533, below the experimental 0.70 target. The failure report was preserved. These are navigation examples, not collected user questions or an independent semantic benchmark; the CI report job does not enforce a recall threshold.
  • Existing static-analysis and dynamic-SQL limitations remain. Historical analysis does not reconstruct external extra_roots.

Upgrade

pip install --upgrade local-codebase-indexer==0.7.0