dirsql-plugin-embeddings 0.1.8__tar.gz → 0.1.9__tar.gz

{dirsql_plugin_embeddings-0.1.8 → dirsql_plugin_embeddings-0.1.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dirsql-plugin-embeddings
-Version: 0.1.8
+Version: 0.1.9
 Summary: First-party dirsql plugin: semantic search via an OpenAI-compatible embeddings endpoint.
 Requires-Python: >=3.10
 Requires-Dist: cachetta>=0.6.15
@@ -44,17 +44,18 @@ and embedded like any other document. The extension check is case-insensitive
 (`.PDF` is a PDF), though the glob above is not — globset matches case-sensitively,
 so an uppercase-suffixed file needs its own `glob` entry to be picked up at all.
 
-The glob is an allowlist rather than `**/*` for a reason worth knowing: a hook
-that exits non-zero aborts the entire scan
-([dirsql#697](https://github.com/thekevinscott/dirsql/issues/697)), and reading a
-PNG as UTF-8 does exactly that. Matching everything would mean one image anywhere
-under the root produces no index at all, so the list stays limited to what the
-plugin can actually read.
-
-A PDF that cannot be parsed aborts the scan rather than being skipped: the hook
-exits non-zero and dirsql reports the path and the pypdf reason. A *scanned*,
-image-only PDF is not a failure — pypdf yields no text, and the file is indexed
-with an empty `text`, exactly like an empty `.md`.
+The glob is an allowlist rather than `**/*` on cost, not correctness. Every
+matched file costs a hook subprocess, and every file the plugin can decode costs
+a billed embedding call — so pointing `**/*` at a tree containing `node_modules`
+or `.git` makes for a slow and expensive scan. The list is what is worth
+embedding; widening it trades money for recall.
+
+A file the plugin cannot read is skipped, not fatal. The hook exits non-zero,
+dirsql names the file on stderr and carries on indexing the rest, and the run
+exits `23` — "completed, some files skipped". From the SDK the same information
+is on `scan_failures()` / `scanFailures()`. A *scanned*, image-only PDF is not a
+failure at all: pypdf yields no text, and the file is indexed with an empty
+`text`, exactly like an empty `.md`.
 
 ## Configuration
 

{dirsql_plugin_embeddings-0.1.8 → dirsql_plugin_embeddings-0.1.9}/README.md

@@ -35,17 +35,18 @@ and embedded like any other document. The extension check is case-insensitive
 (`.PDF` is a PDF), though the glob above is not — globset matches case-sensitively,
 so an uppercase-suffixed file needs its own `glob` entry to be picked up at all.
 
-The glob is an allowlist rather than `**/*` for a reason worth knowing: a hook
-that exits non-zero aborts the entire scan
-([dirsql#697](https://github.com/thekevinscott/dirsql/issues/697)), and reading a
-PNG as UTF-8 does exactly that. Matching everything would mean one image anywhere
-under the root produces no index at all, so the list stays limited to what the
-plugin can actually read.
-
-A PDF that cannot be parsed aborts the scan rather than being skipped: the hook
-exits non-zero and dirsql reports the path and the pypdf reason. A *scanned*,
-image-only PDF is not a failure — pypdf yields no text, and the file is indexed
-with an empty `text`, exactly like an empty `.md`.
+The glob is an allowlist rather than `**/*` on cost, not correctness. Every
+matched file costs a hook subprocess, and every file the plugin can decode costs
+a billed embedding call — so pointing `**/*` at a tree containing `node_modules`
+or `.git` makes for a slow and expensive scan. The list is what is worth
+embedding; widening it trades money for recall.
+
+A file the plugin cannot read is skipped, not fatal. The hook exits non-zero,
+dirsql names the file on stderr and carries on indexing the rest, and the run
+exits `23` — "completed, some files skipped". From the SDK the same information
+is on `scan_failures()` / `scanFailures()`. A *scanned*, image-only PDF is not a
+failure at all: pypdf yields no text, and the file is indexed with an empty
+`text`, exactly like an empty `.md`.
 
 ## Configuration
 

dirsql_plugin_embeddings-0.1.9/e2e-attestations/claude-725-plugin-glob-comment.json

@@ -0,0 +1,7 @@
+{
+  "command": "uv run --with sqlite-vec --with-editable . --with-editable ../../packages/python python -m pytest tests/e2e -q",
+  "ran_at": 1785680099,
+  "exit_code": 0,
+  "commit": "0030917b47975bfa9b4d7412e7f00ca2b760f292",
+  "branch": "claude/725-plugin-glob-comment"
+}

{dirsql_plugin_embeddings-0.1.8 → dirsql_plugin_embeddings-0.1.9}/src/dirsql_plugin_embeddings/dirsql.toml

@@ -16,11 +16,11 @@ entrypoint = "sqlite3_vec_init"
 # rewrites those to `*`, but only when the braces hold a bare identifier, so a
 # comma-separated list reaches globset intact.
 #
-# An allowlist rather than `**/*`, and not for tidiness: a hook that fails takes
-# the whole scan down (dirsql#697), and reading a PNG as UTF-8 fails. Matching
-# everything would mean one image anywhere under the root yields no index at
-# all. The list is therefore what `read_content` can actually read -- prose, not
-# source -- and widening it is safe only as far as that stays true.
+# An allowlist rather than `**/*` because every matched file costs a hook
+# subprocess and, for anything `read_content` can decode, a billed embedding
+# call. Pointed at a tree holding `node_modules` or `.git`, `**/*` is a slow
+# and expensive scan, not a broken one -- unreadable files are skipped. So the
+# list is what is worth embedding, and widening it trades money for recall.
 [[table]]
 ddl = "CREATE TABLE documents (path TEXT, text TEXT, embedding TEXT)"
 glob = "**/*.{md,markdown,mdx,rst,txt,pdf}"