ph-code-graph 0.2.0__py3-none-any.whl → 0.4.0__py3-none-any.whl

ph_code_graph/__init__.py

@@ -42,7 +42,9 @@ answers most of what an agent actually asks and reports where it cannot.
 from __future__ import annotations
 
 import hashlib
+import importlib.util
 import logging
+from collections.abc import Mapping
 from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any, Literal
@@ -59,6 +61,7 @@ from ph.seams._registry import contribute_item
 from ph.seams.changes import tree_state
 from ph.seams.commands import CommandContext, CommandDefinition
 from ph.seams.diagnostics import Diagnostic, contribute
+from ph.seams.fs import FsService
 from ph.seams.skills import discover_skills
 from ph.text import count_of
 from ph.tools.definition import ToolModel, ToolOutput, ToolRunContext, define_tool, text_content
@@ -196,7 +199,8 @@ class CodeGraphSeam:
 
     def report(self) -> list[tuple[str, str]]:
         """`phern doctor`'s section."""
-        import tree_sitter
+        # Not a declared dependency: the row checks for it before this runs.
+        import tree_sitter  # noqa: PLC0415
 
         # No `hasattr` guard: the row declares `inject=[FS]`, so `ctx.fs` is
         # present for as long as it is active — and a `Path.cwd()` fallback
@@ -492,8 +496,6 @@ def offer_skills(ctx: Context) -> None:
 @plugin("code-graph", inject=[TOOLS, FS], config=Config)
 async def apply(ctx: Context, config: Config) -> None:
     """Mount the seam and register both tools."""
-    import importlib.util
-
     for module in ("tree_sitter", "tree_sitter_language_pack"):
         if importlib.util.find_spec(module) is None:
             raise MountRefusal(MISSING)
@@ -535,6 +537,16 @@ async def apply(ctx: Context, config: Config) -> None:
     def store(run: ToolRunContext) -> CodeGraphStore:
         return seam.store_for(ctx.require(FS).root_for(run.agent))
 
+    def _forget_missing(
+        fs: FsService,
+        book: CodeGraphStore,
+        known: Mapping[str, tuple[str, str]],
+        run: ToolRunContext,
+    ) -> int:
+        """Drop every indexed path that is no longer on disk. One thread hop."""
+        gone = [path for path in known if not fs.resolve(path, agent=run.agent).exists()]
+        return book.forget(gone) if gone else 0
+
     async def index_tool(args: IndexArgs, run: ToolRunContext) -> dict[str, Any]:
         fs = ctx.require(FS)
         book = store(run)
@@ -616,10 +628,24 @@ async def apply(ctx: Context, config: Config) -> None:
                 slice_ = await fs.read(
                     path,
                     limit=None,
+                    # `skip_reason` above is this walk's bound, and it is a skip
+                    # rather than a raise on purpose: an indexing pass must not
+                    # fail because it found a minified bundle. `read`'s own
+                    # default would be a second, larger bound that raises.
+                    max_bytes=None,
                     scope=run.scope,
                     agent=run.agent,
                     session=run.session,
                 )
+                # **No prose test here, deliberately** (X6 review). The text
+                # index refuses documents that hold no prose; source code is not
+                # prose, and the same predicate refused a module whose only
+                # statement is a long `__all__`, and a `TABLE = {...}` literal
+                # beside `def lookup()` — so `lookup` would have vanished from
+                # this index with nothing but a skip line to say so. Telling a
+                # minified bundle from a legitimate data module by shape is not
+                # reliable for code, and a false positive here loses symbols
+                # silently. `skip_reason`'s size bound stays this walk's guard.
                 # The content hash stays the authority on *whether* a file
                 # changed — the filter above only decides whether to open it, so
                 # the "content, not clock" guarantee is untouched.
@@ -642,6 +668,25 @@ async def apply(ctx: Context, config: Config) -> None:
                     book.put, path, digest, extraction, state.id_for(path)
                 )
                 indexed += 1
+            if not args.forget:
+                # **Gone from disk is gone from the index** (X3). A deleted file
+                # kept its symbols until somebody ran `forget` by hand, so
+                # `code_graph` went on naming definitions at paths that are not
+                # there — and a rename showed the symbol twice, once under each
+                # name, with nothing saying which was real. A pointer into a file
+                # that does not exist is the one answer this tool must not give,
+                # because the model's next move is to open it.
+                #
+                # **Asked of the filesystem, not of the walk.** "Absent from
+                # `paths`" looks like the same question and is not: `collect`
+                # stops at `max_files` and reports nothing, and the policy screen
+                # prunes whole directories — so on a tree past the cap, or with a
+                # `permissions-fs` rule hiding a subtree, the walk is a *subset*
+                # of the scope and differencing against it would forget
+                # everything it did not reach. It also settles the narrowed walk
+                # for free, where the question is not "did we cover the scope"
+                # but "is this file still there".
+                removed = await anyio.to_thread.run_sync(_forget_missing, fs, book, known, run)
             # Stored last, and only after the loop: a token recorded before the
             # writes would, on a crash between the two, vouch for files this run
             # never actually indexed.

ph_code_graph/_extract.py

@@ -40,6 +40,7 @@ query alone yields two tags and no calls — measured. `INHERITS` is that fact.
 from __future__ import annotations
 
 import logging
+import os
 from collections.abc import Callable, Sequence
 from dataclasses import dataclass, replace
 from functools import cache
@@ -129,7 +130,8 @@ class Extraction:
 
 
 def _pack() -> Any:  # noqa: ANN401
-    import tree_sitter_language_pack
+    # Not a declared dependency: the row checks for it before this runs.
+    import tree_sitter_language_pack  # noqa: PLC0415
 
     return tree_sitter_language_pack
 
@@ -196,8 +198,6 @@ def use_cache(directory: Path) -> Path:
     An operator who set the environment variable deliberately keeps it — their
     spelling wins, which is what makes the variable still mean something.
     """
-    import os
-
     global _configured
 
     override = os.environ.get("TREE_SITTER_LANGUAGE_PACK_CACHE_DIR")
@@ -343,7 +343,8 @@ def _tags_query(language: str) -> Any:  # noqa: ANN401
     for the same reason: a `Query` is read-only once built, and `extract` makes
     a fresh `QueryCursor` per call — which is the mutable half.
     """
-    from tree_sitter import Query
+    # Not a declared dependency: the row checks for it before this runs.
+    from tree_sitter import Query  # noqa: PLC0415
 
     pack = _pack()
     sources = [pack.get_tags_query(one) for one in INHERITS.get(language, ())]
@@ -359,7 +360,8 @@ def extract(path: str, text: str, language: str) -> Extraction:
         should stop the batch or be reported and skipped is its decision, not
         this function's.
     """
-    from tree_sitter import QueryCursor
+    # Not a declared dependency: the row checks for it before this runs.
+    from tree_sitter import QueryCursor  # noqa: PLC0415
 
     pack = _pack()
     config = pack.ProcessConfig(

ph_code_graph/_store.py

@@ -213,6 +213,22 @@ class Hit:
         }
 
 
+def _fts_values(rowid: int, name: str, doc: str | None, path: str) -> tuple[int, str, str, str]:
+    """One `symbols_fts` row, for the insert **and** the delete that undoes it.
+
+    FTS5's `'delete'` command re-derives the postings to remove from the values
+    it is given, so a delete whose tuple differs from the insert's by so much as
+    one NULL removes nothing — and SQLite reports no error for it. `_forget`
+    carried a copy of this expression with a comment saying it mirrored the
+    insert, which is a reader being asked to check two expressions in two
+    methods rather than a compiler being asked to.
+
+    `doc or ""` is the whole substance: `symbols.doc` is nullable and the FTS
+    column is not.
+    """
+    return (rowid, name, doc or "", path)
+
+
 @dataclass(slots=True)
 class CodeGraphStore:
     """The index. **Blocking**; the seam calls it in a worker thread."""
@@ -361,17 +377,33 @@ class CodeGraphStore:
     def _forget(self, connection: sqlite3.Connection, paths: Sequence[str]) -> int:
         gone = 0
         for path in paths:
-            # The FTS rows first, by the id they were inserted under: a
+            # The FTS rows first, **with the values they were inserted under**: a
             # contentless table cannot be asked which rows belong to a path, so
-            # the ids come from `symbols` while it still has them.
-            ids = [
-                row["id"]
-                for row in connection.execute("SELECT id FROM symbols WHERE path = ?", (path,))
+            # they come from `symbols` while it still has them (X1).
+            #
+            # The columns and not just the rowid, because that is what FTS5's
+            # `'delete'` command needs: it re-derives the postings to remove from
+            # the values it is given, and three empty strings therefore remove
+            # *nothing*. SQLite reports no error for it. Every posting of every
+            # re-indexed symbol stayed, and `symbols.id` has no `AUTOINCREMENT` —
+            # so the next insert took the dead row's id and the stale posting
+            # started pointing at a different symbol. `search` joins the rowid
+            # back to `symbols`, so the old name returned the new symbol: worse
+            # than a miss, because it reads like an answer.
+            #
+            # Both tuples come from `_fts_values`, so the agreement the paragraph
+            # above depends on is structural rather than a comment asking a
+            # reader to check two expressions in two methods.
+            stale = [
+                _fts_values(row["id"], row["name"], row["doc"], row["path"])
+                for row in connection.execute(
+                    "SELECT id, name, doc, path FROM symbols WHERE path = ?", (path,)
+                )
             ]
             connection.executemany(
                 "INSERT INTO symbols_fts(symbols_fts, rowid, name, doc, path) "
-                "VALUES ('delete', ?, '', '', '')",
-                [(one,) for one in ids],
+                "VALUES ('delete', ?, ?, ?, ?)",
+                stale,
             )
             connection.execute("DELETE FROM symbols WHERE path = ?", (path,))
             connection.execute("DELETE FROM refs WHERE path = ?", (path,))
@@ -420,7 +452,7 @@ class CodeGraphStore:
                 by_line[definition.start_line] = identifier
                 connection.execute(
                     "INSERT INTO symbols_fts(rowid, name, doc, path) VALUES (?, ?, ?, ?)",
-                    (identifier, definition.name, definition.doc or "", path),
+                    _fts_values(identifier, definition.name, definition.doc, path),
                 )
             # One `{line: tightest definition}` table for the whole file, rather
             # than a scan of every definition per reference: that was

{ph_code_graph-0.2.0.dist-info → ph_code_graph-0.4.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: ph-code-graph
-Version: 0.2.0
+Version: 0.4.0
 Summary: pH plugin: `code_index` and `code_graph`, a Python-native tree-sitter code graph with line-accurate answers.
 Project-URL: Homepage, https://github.com/chastabor/pH
 Project-URL: Repository, https://github.com/chastabor/pH
@@ -21,7 +21,7 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development
 Classifier: Typing :: Typed
 Requires-Python: >=3.12
-Requires-Dist: ph-core==0.2.0
+Requires-Dist: ph-core==0.4.0
 Requires-Dist: tree-sitter-language-pack>=1.16
 Requires-Dist: tree-sitter>=0.25
 Description-Content-Type: text/markdown

ph_code_graph-0.4.0.dist-info/RECORD

@@ -0,0 +1,11 @@
+ph_code_graph/__init__.py,sha256=Cw9l34LPyR7LjinpIIZfjrWn0i8cw8IXwdSpGTJHgVE,38079
+ph_code_graph/_extract.py,sha256=9AprYeNCqIQRZOv5RWtxTQFbU6jlSlPo7ot2fkAcavg,24020
+ph_code_graph/_store.py,sha256=zcxMXE6_DM5DWggQ2iMD_-ds3DfPq0oz4t0O55hyhUM,28785
+ph_code_graph/bundle.yaml,sha256=x3UoJbgQ6QUootM-M6rWZpDQT1sRGSCrtiBh2LFq1ms,1190
+ph_code_graph/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+ph_code_graph/skills/code-graph/SKILL.md,sha256=VNIs0a0U2iveu60E1tQHxz_nRRHOUUHmg9b5_saOMNs,2950
+ph_code_graph-0.4.0.dist-info/METADATA,sha256=Y7RyUptibqyGGbMByzIU4NEhpsMGSg2XS5jxigREeP8,13501
+ph_code_graph-0.4.0.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
+ph_code_graph-0.4.0.dist-info/entry_points.txt,sha256=4DPMS_RTRniZPXb59juSl5mpTpi0fHpwViZPCvffAgo,94
+ph_code_graph-0.4.0.dist-info/licenses/LICENSE,sha256=sN3yhtK25eQpVcewQbaq1k4A-_fWl3RkTf3QfGDh3ko,1070
+ph_code_graph-0.4.0.dist-info/RECORD,,

{ph_code_graph-0.2.0.dist-info → ph_code_graph-0.4.0.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: hatchling 1.32.0
+Generator: hatchling 1.32.4
 Root-Is-Purelib: true
 Tag: py3-none-any

ph_code_graph-0.2.0.dist-info/RECORD

@@ -1,11 +0,0 @@
-ph_code_graph/__init__.py,sha256=SbIRwgUpurM8tGGYS7VYK2HQo4LcDGGaNR5otTXMD-s,35101
-ph_code_graph/_extract.py,sha256=3zBEy3EibEMy4636w3o68lcwNb_wnXXWxXemuhO5WiY,23755
-ph_code_graph/_store.py,sha256=5ZPXvlW_Vt1U-c0ag6fgYBaDp2PVrzPGTI_aPcOFOuU,26964
-ph_code_graph/bundle.yaml,sha256=x3UoJbgQ6QUootM-M6rWZpDQT1sRGSCrtiBh2LFq1ms,1190
-ph_code_graph/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-ph_code_graph/skills/code-graph/SKILL.md,sha256=VNIs0a0U2iveu60E1tQHxz_nRRHOUUHmg9b5_saOMNs,2950
-ph_code_graph-0.2.0.dist-info/METADATA,sha256=xfEy_jtMardrhV8IRKIzo5jz6t9P9_Ua86sFoFZPhi0,13501
-ph_code_graph-0.2.0.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-ph_code_graph-0.2.0.dist-info/entry_points.txt,sha256=4DPMS_RTRniZPXb59juSl5mpTpi0fHpwViZPCvffAgo,94
-ph_code_graph-0.2.0.dist-info/licenses/LICENSE,sha256=sN3yhtK25eQpVcewQbaq1k4A-_fWl3RkTf3QfGDh3ko,1070
-ph_code_graph-0.2.0.dist-info/RECORD,,