docforge-cli 0.5.2__tar.gz → 0.6.0__tar.gz

{docforge_cli-0.5.2 → docforge_cli-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docforge-cli
-Version: 0.5.2
+Version: 0.6.0
 Summary: Forge searchable context from Confluence and git repos for AI coding assistants
 License: MIT
 Project-URL: Homepage, https://GranatenUdo.github.io/docforge/

{docforge_cli-0.5.2 → docforge_cli-0.6.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "docforge-cli"
-version = "0.5.2"
+version = "0.6.0"
 description = "Forge searchable context from Confluence and git repos for AI coding assistants"
 readme = "README.md"
 license = {text = "MIT"}

{docforge_cli-0.5.2 → docforge_cli-0.6.0}/src/docforge/ingest.py

@@ -173,14 +173,15 @@ async def _ingest_confluence_source(
                 await conn.execute(
                     """
                     INSERT INTO chunks (source_id, chunk_index, text,
-                                        embedding, section_title)
-                    VALUES ($1, $2, $3, $4, $5)
+                                        embedding, section_title, title)
+                    VALUES ($1, $2, $3, $4, $5, $6)
                     """,
                     source_id,
                     chunk.chunk_index,
                     chunk.text,
                     np.array(embedding, dtype=np.float32),
                     chunk.section_title,
+                    source.title,
                 )
 
     logger.info("Stored %d chunks for: %s", len(chunks), source.title)
@@ -263,14 +264,15 @@ async def _ingest_git_source(
                     await conn.execute(
                         """
                         INSERT INTO chunks (source_id, chunk_index, text,
-                                            embedding, section_title)
-                        VALUES ($1, $2, $3, $4, $5)
+                                            embedding, section_title, title)
+                        VALUES ($1, $2, $3, $4, $5, $6)
                         """,
                         source_id,
                         chunk.chunk_index,
                         chunk.text,
                         np.array(embedding, dtype=np.float32),
                         chunk.section_title,
+                        file.title,
                     )
 
         logger.info("Stored %d chunks for: %s/%s", len(chunks), source.title, file.title)

docforge_cli-0.6.0/src/docforge/sql/migrations/008_title_weighted_tsv.sql

@@ -0,0 +1,42 @@
+-- Migration 008: weighted text_tsv with title and section_title.
+--
+-- Replaces migration 007's plain to_tsvector('english', text) with a
+-- three-tier weighted variant: title='A', section_title='B', text='D'.
+-- ts_rank_cd respects setweight using its default weights array
+-- {A: 1.0, B: 0.4, C: 0.2, D: 0.1}, so title tokens contribute ~10x a
+-- body token in BM25-style ranking.
+--
+-- Postgres GENERATED ALWAYS expressions cannot be modified in place;
+-- the column is dropped and re-created. Lock window is roughly 15-90s
+-- on ~20k chunks (mostly the ADD COLUMN ... STORED step recomputing
+-- three to_tsvector calls per row under AccessExclusiveLock). Acceptable
+-- for low-volume production; revisit if corpus grows past ~1M chunks.
+--
+-- Idempotency: best-effort via IF [NOT] EXISTS qualifiers. Re-running
+-- causes an unnecessary drop+recreate of text_tsv but doesn't break
+-- anything. The migration runs once per release in practice.
+
+-- Step 1: add the title column (idempotent).
+ALTER TABLE chunks ADD COLUMN IF NOT EXISTS title TEXT NOT NULL DEFAULT '';
+
+-- Step 2: backfill title from sources via JOIN UPDATE.
+-- Only updates rows where title is still the empty default — protects
+-- against repeat runs that would otherwise rewrite the same data.
+UPDATE chunks
+SET title = s.title
+FROM sources s
+WHERE s.id = chunks.source_id AND chunks.title = '';
+
+-- Step 3: drop the v0.5.0 text_tsv (plain to_tsvector('english', text)).
+ALTER TABLE chunks DROP COLUMN IF EXISTS text_tsv;
+
+-- Step 4: re-add text_tsv with the three-tier weighted expression.
+ALTER TABLE chunks ADD COLUMN IF NOT EXISTS text_tsv tsvector
+    GENERATED ALWAYS AS (
+        setweight(to_tsvector('english', title), 'A') ||
+        setweight(to_tsvector('english', coalesce(section_title, '')), 'B') ||
+        setweight(to_tsvector('english', text), 'D')
+    ) STORED;
+
+-- Step 5: re-create the GIN index (was dropped with the old column).
+CREATE INDEX IF NOT EXISTS chunks_text_tsv_idx ON chunks USING GIN (text_tsv);

{docforge_cli-0.5.2 → docforge_cli-0.6.0}/src/docforge_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docforge-cli
-Version: 0.5.2
+Version: 0.6.0
 Summary: Forge searchable context from Confluence and git repos for AI coding assistants
 License: MIT
 Project-URL: Homepage, https://GranatenUdo.github.io/docforge/

{docforge_cli-0.5.2 → docforge_cli-0.6.0}/src/docforge_cli.egg-info/SOURCES.txt

@@ -33,6 +33,7 @@ src/docforge/sql/migrations/004_add_query_log.sql
 src/docforge/sql/migrations/005_add_query_log_user_oid.sql
 src/docforge/sql/migrations/006_add_query_log_request_ms.sql
 src/docforge/sql/migrations/007_add_chunks_text_tsv.sql
+src/docforge/sql/migrations/008_title_weighted_tsv.sql
 src/docforge/templates/docforge.yml
 src/docforge/templates/docker-compose.yml
 src/docforge/templates/mcp_client.py