drydock-cli 3.0.55__tar.gz → 3.0.56__tar.gz

{drydock_cli-3.0.55/drydock_cli.egg-info → drydock_cli-3.0.56}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.55
+Version: 3.0.56
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.55 → drydock_cli-3.0.56}/drydock/tools/__init__.py

@@ -41,7 +41,7 @@ def _resolve_path(path: str, config: dict) -> str:
 SCHEMAS = [
     {
         "name": "Read",
-        "description": "Read a file. Returns content with line numbers. Use limit/offset for large files.",
+        "description": "Read a file. Returns content with line numbers. Use limit/offset for large files. A very large file with no limit returns a STRUCTURE INDEX (key symbols + line numbers) instead of the whole file — then Read the ranges you need with offset/limit.",
         "input_schema": {
             "type": "object",
             "properties": {
@@ -303,18 +303,67 @@ SCHEMAS = [
 
 # ── Tool implementations ──────────────────────────────────────────────────
 
+# Above this many lines, a Read with no explicit window returns a structural
+# INDEX (key symbols + line numbers) instead of dumping the file — semantic
+# chunking so a huge file can't blow the context window. The model then Reads
+# the ranges it needs with offset/limit.
+_BIG_FILE_LINES = 1500
+
+# Structural anchors: a definition → a node the model can jump to. Code anchors
+# apply to every file; markdown headers (anchored at column 0 so indented code
+# COMMENTS aren't mistaken for headings) apply only to doc files.
+_CODE_RE = re.compile(
+    r"^\s*("
+    r"(?:async\s+)?def\s+\w+"            # python functions
+    r"|class\s+\w+"                       # python/js/etc classes
+    r"|(?:export\s+)?(?:async\s+)?function\s+\w+"  # js/ts functions
+    r"|(?:export\s+)?(?:const|let|var)\s+\w+\s*=\s*(?:async\s*)?\("  # js arrow fns
+    r"|func\s+\w+"                        # go
+    r"|fn\s+\w+"                          # rust
+    r")"
+)
+_MD_RE = re.compile(r"^#{1,6}\s+\S")  # markdown headers, column 0 only
+_DOC_EXT = {".md", ".markdown", ".rst", ".txt", ".org"}
+
+
+def _file_index(lines: list[str], fp: str) -> str:
+    """A compact, line-numbered index of a large file's structure."""
+    is_doc = Path(fp).suffix.lower() in _DOC_EXT
+    anchors = [
+        f"  L{i + 1}: {ln.rstrip()[:100]}"
+        for i, ln in enumerate(lines)
+        if _CODE_RE.match(ln) or (is_doc and _MD_RE.match(ln))
+    ]
+    head = (
+        f"{fp} is large ({len(lines)} lines) — showing a STRUCTURE INDEX instead "
+        f"of the whole file (to save context). Read a specific section with "
+        f"offset/limit (e.g. {{\"file_path\": ..., \"offset\": <line-1>, "
+        f"\"limit\": 120}}).\n"
+    )
+    if not anchors:
+        return head + "(no def/class/header anchors found — read ranges with offset/limit.)"
+    body = "\n".join(anchors[:400])
+    if len(anchors) > 400:
+        body += f"\n  [... {len(anchors) - 400} more anchors ...]"
+    return f"{head}\nKey locations ({len(anchors)}):\n{body}"
+
+
 def tool_read(params: dict, config: dict) -> str:
     fp = _resolve_path(params["file_path"], config)
-    limit = params.get("limit", 2000)
+    limit = params.get("limit")  # None = caller didn't specify a window
     offset = params.get("offset", 0)
     try:
         with open(fp, "r", encoding="utf-8", errors="replace") as f:
             lines = f.readlines()
-        selected = lines[offset:offset + limit]
+        # Huge file, no explicit window → index it instead of dumping it.
+        if limit is None and offset == 0 and len(lines) > _BIG_FILE_LINES:
+            return _file_index(lines, fp)
+        eff_limit = 2000 if limit is None else limit
+        selected = lines[offset:offset + eff_limit]
         numbered = [f"{i + offset + 1}\t{line.rstrip()}" for i, line in enumerate(selected)]
         result = "\n".join(numbered)
-        if len(lines) > offset + limit:
-            result += f"\n[... {len(lines) - offset - limit} more lines]"
+        if len(lines) > offset + eff_limit:
+            result += f"\n[... {len(lines) - offset - eff_limit} more lines]"
         return result or "(empty file)"
     except FileNotFoundError:
         return f"Error: file not found: {fp}"

{drydock_cli-3.0.55 → drydock_cli-3.0.56/drydock_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.55
+Version: 3.0.56
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.55 → drydock_cli-3.0.56}/drydock_cli.egg-info/SOURCES.txt

@@ -58,6 +58,7 @@ tests/test_mcp.py
 tests/test_oneshot_unreachable.py
 tests/test_plan_autocontinue.py
 tests/test_providers_unreachable.py
+tests/test_read_index.py
 tests/test_runaway_repetition.py
 tests/test_skills.py
 tests/test_stop.py

{drydock_cli-3.0.55 → drydock_cli-3.0.56}/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
 # PyPI distribution name is drydock-cli (the established install name, continued
 # from the retired v2 fork); the import package + CLI command stay `drydock`.
 name = "drydock-cli"
-version = "3.0.55"
+version = "3.0.56"
 description = "Drydock — a local, provider-agnostic terminal coding agent for local LLMs"
 readme = "README.md"
 requires-python = ">=3.11"

drydock_cli-3.0.56/tests/test_read_index.py

@@ -0,0 +1,52 @@
+"""Semantic chunking: Read returns a STRUCTURE INDEX for a huge file (no
+explicit window) instead of dumping it, so it can't blow the context window."""
+from __future__ import annotations
+
+from drydock.tools import tool_read
+
+
+def _big_py(tmp_path, n=2000):
+    lines = []
+    for i in range(n):
+        if i % 400 == 0:
+            lines.append(f"def func_{i}(x):")
+        elif i % 400 == 1:
+            lines.append("    return x")
+        else:
+            lines.append(f"    # comment {i}")
+    (tmp_path / "big.py").write_text("\n".join(lines))
+    return {"cwd": str(tmp_path)}
+
+
+def test_huge_file_no_window_returns_index(tmp_path):
+    cfg = _big_py(tmp_path)
+    out = tool_read({"file_path": "big.py"}, cfg)
+    assert "STRUCTURE INDEX" in out
+    assert "def func_0" in out and "def func_400" in out
+    assert "# comment" not in out  # comments are not anchors
+
+
+def test_index_is_small_not_a_dump(tmp_path):
+    cfg = _big_py(tmp_path)
+    out = tool_read({"file_path": "big.py"}, cfg)
+    assert len(out.splitlines()) < 50  # an index, not 2000 lines
+
+
+def test_explicit_window_still_reads_slice(tmp_path):
+    cfg = _big_py(tmp_path)
+    out = tool_read({"file_path": "big.py", "offset": 0, "limit": 3}, cfg)
+    assert "STRUCTURE INDEX" not in out
+    assert out.startswith("1\tdef func_0")
+
+
+def test_small_file_reads_fully(tmp_path):
+    (tmp_path / "s.py").write_text("def a():\n    return 1\n")
+    out = tool_read({"file_path": "s.py"}, {"cwd": str(tmp_path)})
+    assert "STRUCTURE INDEX" not in out and "def a()" in out
+
+
+def test_markdown_headers_indexed(tmp_path):
+    body = "\n".join(["# Title"] + [f"para {i}" for i in range(2000)] + ["## Section B"])
+    (tmp_path / "doc.md").write_text(body)
+    out = tool_read({"file_path": "doc.md"}, {"cwd": str(tmp_path)})
+    assert "STRUCTURE INDEX" in out and "# Title" in out and "## Section B" in out