dirsql 0.3.124__tar.gz → 0.3.125__tar.gz

{dirsql-0.3.124 → dirsql-0.3.125}/Cargo.lock

@@ -500,7 +500,7 @@ dependencies = [
 
 [[package]]
 name = "dirsql-py-ext"
-version = "0.3.124"
+version = "0.3.125"
 dependencies = [
  "dirsql",
  "pyo3",

{dirsql-0.3.124 → dirsql-0.3.125}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dirsql
-Version: 0.3.124
+Version: 0.3.125
 Summary: Ephemeral SQL index over a local directory
 Keywords: sql,filesystem,directory,sqlite,index
 Author: Kevin Scott

{dirsql-0.3.124 → dirsql-0.3.125}/packages/python/Cargo.toml

@@ -4,7 +4,7 @@ name = "dirsql-py-ext"
 # pypi/maturin handler can rewrite it via `write-version` before
 # `maturin build`. `pyproject.toml` declares `dynamic = ["version"]`
 # and maturin reads this field. Mirrors `packages/rust/Cargo.toml`.
-version = "0.3.124"
+version = "0.3.125"
 # abi3 note: the `abi3-py311` pyo3 feature (below) builds ONE stable-ABI
 # `cp311-abi3` wheel per platform that loads on every CPython >= 3.11
 # (matching `pyproject.toml`'s `requires-python = ">=3.11"`), instead of

dirsql-0.3.125/packages/python/changelog.d/2026-07-24-remove-stat-fact-injection.md

@@ -0,0 +1,3 @@
+**Removed**
+
+- **Removed the stat-fact injection layer** (core change, surfaced through the Python SDK). dirsql no longer auto-fills DDL columns named `path`/`basename`/`dir`/`ext`/`size`/`mtime`/`ctime` from filesystem stat. A `Table` now contains exactly what its `on_file` callback returns; a DDL column the callback never emits is `None`/`NULL`, not injected. To keep a stat column populated, emit it from the callback (e.g. derive `path` from the file argument), or query the path directly with a path-table (`FROM './'`). (#634)

dirsql-0.3.125/packages/python/e2e-attestations/feat-634-core-remove-injection.json

@@ -0,0 +1,7 @@
+{
+  "command": "just test-e2e",
+  "ran_at": 1784902229,
+  "exit_code": 0,
+  "commit": "7571e0d010b096821bfc99b1f893db2b320347dc",
+  "branch": "feat/634-core-remove-injection"
+}

dirsql-0.3.125/packages/python/migrations.d/2026-07-24-remove-stat-fact-injection.md

@@ -0,0 +1,51 @@
+### Stat-fact injection removed (#634)
+
+#### Summary
+
+dirsql previously injected filesystem-derived values into any DDL column named
+`path`, `basename`, `dir`, `ext`, `size`, `mtime`, or `ctime` — filling them
+even when a table's `on_file` callback emitted nothing for that column. That
+injection layer (in the shared Rust core) is gone. A `Table` now contains
+**exactly what its `on_file` callback returns**; a declared column the callback
+never emits is `None`/`NULL`. No Python API signature changes — runtime
+behavior only.
+
+#### Required changes
+
+| Before (injected) | After (NULL unless emitted) | Fix |
+| ----------------- | --------------------------- | --- |
+| DDL declares `path TEXT`, callback omits it → column held the root-relative path | Column is `None` | Emit it from the callback: `Table("CREATE TABLE t (path TEXT)", "**/*", on_file=lambda path: [{"path": path}])`. |
+| DDL declares `size INTEGER`, callback omits it → column held the file size | Column is `None` | Emit it: `on_file=lambda path: [{"size": os.path.getsize(path)}]`. |
+| Want stat columns with **no** callback at all | (was injection) | Query the path directly with a path-table: `FROM './'`. |
+
+#### Deprecations removed
+
+_None._
+
+#### Behavior changes without code changes
+
+- A DDL column named `path`/`basename`/`dir`/`ext`/`size`/`mtime`/`ctime` that
+  the `on_file` callback does not emit is now `None`/`NULL` instead of an
+  injected filesystem value.
+- A **strict** table whose DDL declares such a column but whose callback does
+  not emit it now raises a schema-validation error rather than being silently
+  filled.
+
+#### Verification
+
+```python
+from dirsql import DirSQL, Table
+
+# Callback emits only "size"; DDL also declares "path".
+db = DirSQL(root, tables=[Table(
+    ddl="CREATE TABLE files (path TEXT, size INTEGER)",
+    glob="**/*.txt",
+    on_file=lambda _path: [{"size": 42}],
+)])
+rows = db.query("SELECT path, size FROM files")
+# Before: rows[0]["path"] == "a.txt"  (injected)
+# After:  rows[0]["path"] is None      (not emitted)
+
+# Restore by emitting it:
+#   on_file=lambda path: [{"path": path, "size": 42}]
+```

dirsql-0.3.125/packages/rust/changelog.d/2026-07-24-remove-stat-fact-injection.md

@@ -0,0 +1,3 @@
+**Removed**
+
+- **Removed the stat-fact injection layer.** dirsql no longer auto-fills DDL columns named `path`/`basename`/`dir`/`ext`/`size`/`mtime`/`ctime` from filesystem stat. A named table now contains exactly what its `on-file` hook emits; a DDL column the hook never emits is `NULL` (standard SQL), not injected. To keep a stat column populated, emit it from the hook (e.g. derive `path` from the file argument), or query the path directly with a path-table (`FROM './'`). (#634)

dirsql-0.3.125/packages/rust/migrations.d/2026-07-24-remove-stat-fact-injection.md

@@ -0,0 +1,55 @@
+### Stat-fact injection removed (#634)
+
+#### Summary
+
+dirsql previously injected filesystem-derived values into any DDL column named
+`path`, `basename`, `dir`, `ext`, `size`, `mtime`, or `ctime` — filling them
+even when the `on-file` hook emitted nothing for that column. That injection
+layer is gone. A named table now contains **exactly what its hook emits**; a
+declared column the hook never emits is `NULL` (standard SQL). No API signature
+changes — this is a runtime behavior change.
+
+#### Required changes
+
+| Before (injected) | After (NULL unless emitted) | Fix |
+| ----------------- | --------------------------- | --- |
+| DDL declares `path TEXT`, hook emits neither `path` nor a value for it → column held the root-relative path | Column is `NULL` | Emit it from the hook. Config `on-file` shell hook (`$1` = abs path, `$2` = index root): `rel=${1#"$2"/}; jq -n --arg path "$rel" '[{path:$path}]'`. Rust SDK closure: `\|path\| vec![HashMap::from([("path".into(), Value::Text(path.to_string()))])]`. |
+| DDL declares `size INTEGER`, hook emits neither → column held the file size | Column is `NULL` | Emit it from the hook: `jq -n --argjson size "$(stat -c%s "$1")" '[{size:$size}]'`, or in Rust read `std::fs::metadata(path)?.len()`. |
+| Want stat columns with **no** hook code at all | (was injection) | Query the path directly with a path-table: `FROM './'`. |
+
+#### Deprecations removed
+
+_None._
+
+#### Behavior changes without code changes
+
+- A DDL column named `path`/`basename`/`dir`/`ext`/`size`/`mtime`/`ctime` that
+  the `on-file` hook does not emit is now `NULL` instead of an injected
+  filesystem value. Configs and SDK tables that relied on injection must emit
+  the column themselves (see the table above).
+- A **strict** table whose DDL declares such a column but whose hook does not
+  emit it now fails schema validation (`missing columns for table …`) rather
+  than being silently filled.
+
+#### Verification
+
+Given `a.txt` in the index root and a non-strict table whose DDL is
+`CREATE TABLE files (path TEXT, size INTEGER)` and whose hook emits only
+`size`:
+
+```bash
+# Config flags follow the subcommand: `dirsql query <sql> -c <cfg>`.
+
+# Before this change (hook emits only size):
+dirsql query "SELECT path, size FROM files" -c ./.dirsql.toml
+# -> [{"path":"a.txt","size":42}]   (path injected)
+
+# After this change (same config):
+dirsql query "SELECT path, size FROM files" -c ./.dirsql.toml
+# -> [{"path":null,"size":42}]      (path NULL — not emitted)
+
+# Restore it by emitting path from the hook:
+#   on-file = 'rel=${1#"$2"/}; jq -n --arg path "$rel" '\''[{path:$path}]'\'''
+dirsql query "SELECT path, size FROM files" -c ./.dirsql.toml
+# -> [{"path":"a.txt", ...}]        (path populated again)
+```

{dirsql-0.3.124 → dirsql-0.3.125}/packages/rust/src/lib.rs

@@ -547,8 +547,6 @@ impl DirSQL {
             Err(e) => return vec![error_event(Some(table), rel_path, e.to_string())],
         };
 
-        let stat = compute_stat_virtuals(rel_path, abs_path);
-
         let strict = *self.inner.strict_map.get(table).unwrap_or(&false);
 
         // Normalize, snapshot the file's previous rows, and apply the
@@ -559,11 +557,6 @@ impl DirSQL {
                 Ok(g) => g,
                 Err(e) => return vec![error_event(Some(table), rel_path, e.to_string())],
             };
-            let declared_columns = match db.get_table_columns(table) {
-                Ok(cols) => cols,
-                Err(e) => return vec![error_event(Some(table), rel_path, e.to_string())],
-            };
-            let raw_rows = merge_filesystem_facts(raw_rows, &stat, &declared_columns);
             let mut new_rows = Vec::with_capacity(raw_rows.len());
             for raw in &raw_rows {
                 match db.normalize_row(table, raw, strict) {
@@ -840,10 +833,6 @@ impl DirSQL {
                     message: e.to_string(),
                 })?;
 
-            let stat_virtuals = compute_stat_virtuals(&rel_path, &abs_path);
-            let declared_columns = db.get_table_columns(&table_name).map_err(map_db_error)?;
-            let raw_rows = merge_filesystem_facts(raw_rows, &stat_virtuals, &declared_columns);
-
             // When updating an existing file in the persistent cache, drop
             // its old rows before inserting the new ones.
             if persist_ready.is_some() {
@@ -1475,21 +1464,14 @@ fn relative_path(root: &Path, path: &Path) -> String {
 
 /// Build [`Table`] objects from a parsed config.
 ///
-/// A plain config-defined table produces one row per matched file built
-/// entirely from filesystem facts: glob path captures and stat virtuals
-/// (`path`, `basename`, `dir`, `ext`, `size`, `mtime`, `ctime`) are
-/// injected by the core pipeline ([`merge_filesystem_facts`]). Its synthesized
-/// on_file emits a single empty row per file; the fact-injection layer fills it
-/// in.
-///
-/// A table with an `on-file` command instead runs that command once per matched
-/// file (see [`run_on_file`]): the command reads the file and prints a JSON
-/// array of row objects on stdout, which becomes the file's rows (filesystem
-/// facts are still merged on top, user values winning). `config_dir` is the
-/// command's working directory (the config file's parent) and `root` is the
-/// resolved index root exposed as the `{root}` placeholder. `timeout` bounds
-/// each `on-file` run; the caller resolves it from the global
-/// `[dirsql].hook-timeout` key, falling back to
+/// A config-defined table runs its `on-file` command once per matched file
+/// (see [`run_on_file`]): the command reads the file and prints a JSON array of
+/// row objects on stdout, which becomes the file's rows verbatim. The core
+/// injects nothing — a DDL column the hook does not emit is NULL, validated
+/// against the DDL as usual. `config_dir` is the command's working directory
+/// (the config file's parent) and `root` is the resolved index root exposed as
+/// the `{root}` placeholder. `timeout` bounds each `on-file` run; the caller
+/// resolves it from the global `[dirsql].hook-timeout` key, falling back to
 /// [`command::DEFAULT_COMMAND_TIMEOUT`].
 /// Whether this build declared no tables by any route — neither a config file
 /// nor a programmatic table. That is exactly the state in which `files` used to
@@ -1719,33 +1701,6 @@ fn find_capture_column_collision(glob: &str, declared_columns: &[String]) -> Opt
         .find(|name| declared.contains(name.as_str()))
 }
 
-/// Merge filesystem-fact columns (stat virtuals) into each raw row produced by
-/// an on_file closure. Auto-injected keys are filtered to those declared in
-/// `declared_columns`, so a strict-mode table with a minimal DDL is not broken
-/// by virtuals it didn't ask for. User-provided values in `raw_rows` win over
-/// auto-injected values: an on_file that explicitly emits e.g. `path` is
-/// honored.
-fn merge_filesystem_facts(raw_rows: Vec<Row>, stat: &Row, declared_columns: &[String]) -> Vec<Row> {
-    let declared: std::collections::HashSet<&str> =
-        declared_columns.iter().map(String::as_str).collect();
-
-    raw_rows
-        .into_iter()
-        .map(|raw| {
-            let mut merged = Row::new();
-            for (k, v) in stat {
-                if declared.contains(k.as_str()) {
-                    merged.insert(k.clone(), v.clone());
-                }
-            }
-            for (k, v) in raw {
-                merged.insert(k, v);
-            }
-            merged
-        })
-        .collect()
-}
-
 /// Async wrapper around [`DirSQL`] whose constructor returns immediately while
 /// the initial scan runs on a background thread.
 ///
@@ -2157,7 +2112,7 @@ mod internal_tests {
         let db = DirSQL::with_ignore_and_fs(
             &root,
             vec![Table::new(
-                "CREATE TABLE items (name TEXT, path TEXT)",
+                "CREATE TABLE items (name TEXT)",
                 "**/*.txt",
                 |_| {
                     vec![Row::from_iter([(
@@ -2174,10 +2129,9 @@ mod internal_tests {
         let events = db.process_file_event(FileEvent::Created(abs));
         assert_eq!(events.len(), 1, "expected one insert: {events:?}");
         match &events[0] {
-            RowEvent::Insert { row, .. } => {
+            RowEvent::Insert { file_path, .. } => {
                 assert_eq!(
-                    row.get("path"),
-                    Some(&Value::Text("nested/a.txt".to_string())),
+                    file_path, "nested/a.txt",
                     "watch_root prefix must be stripped to a root-relative path"
                 );
             }
@@ -2195,7 +2149,7 @@ mod internal_tests {
         let mut db = DirSQL::with_ignore_and_fs(
             &root,
             vec![Table::new(
-                "CREATE TABLE items (name TEXT, path TEXT)",
+                "CREATE TABLE items (name TEXT)",
                 "**/*.txt",
                 |_| {
                     vec![Row::from_iter([(
@@ -2215,10 +2169,9 @@ mod internal_tests {
         let events = db.process_file_event(FileEvent::Created(abs));
         assert_eq!(events.len(), 1, "expected one insert: {events:?}");
         match &events[0] {
-            RowEvent::Insert { row, .. } => {
+            RowEvent::Insert { file_path, .. } => {
                 assert_eq!(
-                    row.get("path"),
-                    Some(&Value::Text("b.txt".to_string())),
+                    file_path, "b.txt",
                     "root fallback must strip the user-supplied root prefix"
                 );
             }
@@ -3089,37 +3042,6 @@ mod internal_tests {
         );
     }
 
-    #[test]
-    fn merge_filesystem_facts_filters_to_declared_columns() {
-        let mut stat = Row::new();
-        stat.insert("path".to_string(), Value::Text("2024-01/a.txt".into()));
-        stat.insert("size".to_string(), Value::Integer(9));
-        let raw = vec![Row::from_iter([(
-            "name".to_string(),
-            Value::Text("x".into()),
-        )])];
-        let declared = vec!["name".to_string(), "path".to_string()];
-        let merged = merge_filesystem_facts(raw, &stat, &declared);
-        assert_eq!(merged.len(), 1);
-        let row = &merged[0];
-        assert_eq!(row["path"], Value::Text("2024-01/a.txt".into()));
-        assert!(!row.contains_key("size"), "undeclared stat dropped");
-        assert_eq!(row["name"], Value::Text("x".into()));
-    }
-
-    #[test]
-    fn merge_filesystem_facts_raw_value_wins_over_stat() {
-        let mut stat = Row::new();
-        stat.insert("path".to_string(), Value::Text("auto".into()));
-        let raw = vec![Row::from_iter([(
-            "path".to_string(),
-            Value::Text("explicit".into()),
-        )])];
-        let declared = vec!["path".to_string()];
-        let merged = merge_filesystem_facts(raw, &stat, &declared);
-        assert_eq!(merged[0]["path"], Value::Text("explicit".into()));
-    }
-
     #[test]
     fn find_capture_column_collision_flags_a_placeholder_naming_a_column() {
         let declared = vec!["thread_id".to_string(), "basename".to_string()];

dirsql-0.3.124/packages/python/e2e-attestations/feat-634-prep-python-selfemit.json

@@ -1,7 +0,0 @@
-{
-  "command": "just test-e2e",
-  "ran_at": 1784897533,
-  "exit_code": 0,
-  "commit": "6285e8c0753676dbe7333f42d3a66749a2e808b8",
-  "branch": "feat/634-prep-python-selfemit"
-}

dirsql-0.3.124/packages/python/e2e-attestations/feat-662-query-default-cli.json

@@ -1,7 +0,0 @@
-{
-  "command": "just test-e2e",
-  "ran_at": 1784897776,
-  "exit_code": 0,
-  "commit": "93084d576d3080073dc2e0af6c20201de7def162",
-  "branch": "feat/662-query-default-cli"
-}