code-puppy-core-plugins 0.0.72__tar.gz → 0.0.74__tar.gz

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: code-puppy-core-plugins
-Version: 0.0.72
+Version: 0.0.74
 Summary: Official plugin bundle for Code Puppy
 Project-URL: Repository, https://github.com/mpfaffenberger/code_puppy_core_plugins
 Project-URL: Homepage, https://github.com/mpfaffenberger/code_puppy_core_plugins
@@ -18,10 +18,20 @@ Requires-Python: <3.15,>=3.11
 Requires-Dist: httpx2>=2.7
 Requires-Dist: httpx<1,>=0.24.1
 Requires-Dist: pydantic-ai-slim[anthropic,mcp,openai,typesafe]<3,>=2.51.0
+Requires-Dist: tree-sitter-bash>=0.25.1
+Requires-Dist: tree-sitter-c-sharp>=0.23.5
+Requires-Dist: tree-sitter-c>=0.24.2
+Requires-Dist: tree-sitter-cpp>=0.23.4
 Requires-Dist: tree-sitter-go>=0.25.0
 Requires-Dist: tree-sitter-java>=0.23.5
 Requires-Dist: tree-sitter-javascript>=0.25.0
+Requires-Dist: tree-sitter-kotlin>=1.1.0
+Requires-Dist: tree-sitter-lua>=0.5.0
+Requires-Dist: tree-sitter-php>=0.24.1
+Requires-Dist: tree-sitter-ruby>=0.23.1
 Requires-Dist: tree-sitter-rust>=0.24.2
+Requires-Dist: tree-sitter-scala>=0.26.2
+Requires-Dist: tree-sitter-swift>=0.7.3
 Requires-Dist: tree-sitter-typescript>=0.23.2
 Requires-Dist: tree-sitter<0.26,>=0.25.2
 Description-Content-Type: text/markdown

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/code_puppy_core_plugins/jev_grep/chunks.py

@@ -2,10 +2,11 @@
 
 Source files are split along their syntax tree (see ``structure.py``):
 functions and methods stay whole, long declarations split into the blocks
-inside them. Python uses the stdlib ``ast``; JavaScript/TypeScript, Go, Rust
-and Java use tree-sitter (``treesitter.py``). Everything else, and any file
-that does not parse, falls back to overlapping line windows. Line coverage
-is exact: every non-blank line lands in at least one snippet.
+inside them. Python uses the stdlib ``ast``; fourteen more languages use
+tree-sitter (``treesitter.py``), where regions that do not parse fall back to
+overlapping line windows. Other file types, and Python that does not parse,
+are cut into line windows entirely. Line coverage is exact: every non-blank
+line lands in at least one snippet.
 """
 
 from __future__ import annotations
@@ -96,7 +97,8 @@ def _chunks_from_ranges(
             continue
         part = lines[start - 1 : end]
         whole = (
-            end - start < WHOLE_DECLARATION_LINES
+            not r.window
+            and end - start < WHOLE_DECLARATION_LINES
             and len("\n".join(part)) <= MAX_CHUNK_CHARS
         )
         size, overlap = (len(part), 0) if whole else (WINDOW, OVERLAP)

code_puppy_core_plugins-0.0.74/code_puppy_core_plugins/jev_grep/completer.py

@@ -0,0 +1,40 @@
+"""Tab-completion for smart_grep's ``/set`` keys.
+
+Core's ``/set`` completer only knows its built-in keys plus whatever is
+already saved in puppy.cfg, so ``smart_grep`` was invisible until typed blind.
+This fills that gap and steps aside once a key is saved: core lists it from
+then on, and yielding it here too would show it twice.
+"""
+
+from __future__ import annotations
+
+from typing import Iterable
+
+from termflow.tui.completion import Completer, Completion, Document
+
+from code_puppy.config import get_config_keys
+
+from .config import SETTING_KEYS
+
+TRIGGER = "/set "
+
+
+class SmartGrepSetCompleter(Completer):
+    """Completes unsaved smart_grep keys after ``/set ``."""
+
+    def get_completions(
+        self, document: Document, complete_event: object
+    ) -> Iterable[Completion]:
+        text = document.text_before_cursor.lstrip()
+        if not text.startswith(TRIGGER):
+            return
+        partial = text[len(TRIGGER) :].lstrip()
+        candidates = [key for key in SETTING_KEYS if key.startswith(partial)]
+        if not candidates:
+            return  # don't touch puppy.cfg for every unrelated `/set` keystroke
+        saved = set(get_config_keys())
+        for key in candidates:
+            if key in saved:
+                continue
+            # Same "key = " shape core's SetCompleter emits for unset keys.
+            yield Completion(f"{key} = ", start_position=-len(partial))

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/code_puppy_core_plugins/jev_grep/config.py

@@ -13,6 +13,11 @@ import os
 from code_puppy.config import get_api_key, get_truthy_bool_value, get_value
 
 ENABLED_KEY = "smart_grep"
+MODEL_KEY = "smart_grep_model"
+THRESHOLD_KEY = "smart_grep_threshold"
+# What `/set` autocompletes. The API key is deliberately absent: completions
+# echo the current value, and secrets don't belong in a popup.
+SETTING_KEYS = (ENABLED_KEY, MODEL_KEY, THRESHOLD_KEY)
 # TypeSafe's official name first, then the Jev-branded alias. Config lookup
 # is case-insensitive, so `/set jev_api_key ...` matches JEV_API_KEY.
 API_KEY_NAMES = ("TYPESAFE_API_KEY", "JEV_API_KEY")
@@ -44,12 +49,12 @@ def is_available() -> bool:
 
 def get_jev_model_name() -> str:
     """Pin a version (e.g. ``jev-1.13.0``) once a threshold is tuned against it."""
-    return get_value("smart_grep_model") or DEFAULT_MODEL
+    return get_value(MODEL_KEY) or DEFAULT_MODEL
 
 
 def get_threshold() -> float:
     try:
-        value = float(get_value("smart_grep_threshold") or DEFAULT_THRESHOLD)
+        value = float(get_value(THRESHOLD_KEY) or DEFAULT_THRESHOLD)
     except ValueError:
         return DEFAULT_THRESHOLD
     return value if 0 <= value <= 1 else DEFAULT_THRESHOLD

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/code_puppy_core_plugins/jev_grep/register_callbacks.py

@@ -8,6 +8,7 @@ key is configured. ``/set`` reloads the agent, so toggling applies at once.
 
 from code_puppy.callbacks import register_callback
 
+from .completer import SmartGrepSetCompleter
 from .config import is_available
 from .tool import TOOL_NAME, register_smart_grep
 
@@ -52,3 +53,4 @@ and the query to TypeSafe, so use local grep/reads for local-only work.
 register_callback("load_prompt", _discovery_instructions)
 register_callback("register_tools", _register_tools)
 register_callback("register_agent_tools", _advertise_when_configured)
+register_callback("register_completion_provider", SmartGrepSetCompleter)

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/code_puppy_core_plugins/jev_grep/structure.py

@@ -10,7 +10,10 @@ One algorithm for every parser. A language plugs in by implementing
   them, so each range is one behaviour. Splits cover every line of the
   original range, so coverage is unchanged;
 * runs of short neighbours (imports, one-liners) merge into one range,
-  since each range costs a relevance judgment.
+  since each range costs a relevance judgment;
+* a node the parser could not read cleanly is not split or trusted: it
+  becomes a ``window`` range, cut into overlapping line windows like an
+  unparsed file. One bad macro costs its own region, not the whole file.
 """
 
 from __future__ import annotations
@@ -29,6 +32,7 @@ class Range:
     start: int
     end: int
     symbol: str | None
+    window: bool = False  # parse error inside: use line windows, not one blob
 
 
 class Syntax(Protocol):
@@ -46,6 +50,9 @@ class Syntax(Protocol):
     def statements(self, node: Any) -> Sequence[Any] | None:
         """The statements of ``node``'s body, if it has one."""
 
+    def broken(self, node: Any) -> bool:
+        """True when the parser could not read ``node`` cleanly."""
+
 
 def partition(node: Any, start: int, end: int, symbol, syntax: Syntax, depth=0):
     """Split a long body into its blocks, covering every line of [start, end]."""
@@ -106,6 +113,9 @@ def structure_ranges(nodes: Sequence[Any], syntax: Syntax) -> list[Range]:
             if ranges and ranges[-1].end < end:
                 ranges[-1].end = end
             return
+        if syntax.broken(node):
+            ranges.append(Range(start, end, symbol, window=True))
+            return
         ranges.extend(
             partition(node, start, end, symbol, syntax) or [Range(start, end, symbol)]
         )
@@ -139,7 +149,7 @@ def merge_short_neighbours(ranges: list[Range]) -> list[Range]:
             names = dict.fromkeys(s for s in (last.symbol, r.symbol) if s)
             last.end, last.symbol = r.end, ", ".join(names) or None
         else:
-            merged.append(Range(r.start, r.end, r.symbol))
+            merged.append(Range(r.start, r.end, r.symbol, r.window))
     return merged
 
 
@@ -162,6 +172,9 @@ class PythonSyntax:
             return None
         return [s for s in body if hasattr(s, "end_lineno")]
 
+    def broken(self, node: ast.AST) -> bool:
+        return False  # ast.parse either succeeds for the whole file or raises
+
 
 def python_ranges(text: str) -> list[Range]:
     """Raises SyntaxError/ValueError/RecursionError on unparseable source."""

code_puppy_core_plugins-0.0.74/code_puppy_core_plugins/jev_grep/treesitter.py

@@ -0,0 +1,409 @@
+"""tree-sitter :class:`~.structure.Syntax` for JavaScript/TypeScript, Go, Rust,
+Java, C, C++, C#, Ruby, PHP, Kotlin, Swift, Scala, Bash and Lua.
+
+Grammars are pinned, offline wheels (one package per language); nothing is
+downloaded at runtime. They are optional at runtime: if tree-sitter or a
+grammar cannot be imported (e.g. an unsupported platform) or the installed
+tree-sitter is a known-bad version, :func:`treesitter_ranges` returns
+``None`` and the caller falls back to line windows. A file with parse errors
+(macros, newer syntax than the grammar knows) keeps its structure where the
+tree is clean; only the broken regions fall back to line windows.
+"""
+
+from __future__ import annotations
+
+import importlib
+import os
+from collections.abc import Callable, Iterable, Sequence
+from functools import cache
+from importlib.metadata import PackageNotFoundError, version
+from typing import Any
+
+from .structure import Range, structure_ranges
+
+
+def _grammar(module: str, attr: str = "language") -> Callable[[], Any]:
+    """A lazy loader for one grammar (the wheel is imported on first use)."""
+
+    def load() -> Any:
+        return getattr(importlib.import_module(module), attr)()
+
+    load.__qualname__ = f"load<{module}.{attr}>"
+    return load
+
+
+_JAVASCRIPT = ("javascript", _grammar("tree_sitter_javascript"))
+_TYPESCRIPT = ("typescript", _grammar("tree_sitter_typescript", "language_typescript"))
+_C = ("c", _grammar("tree_sitter_c"))
+_CPP = ("cpp", _grammar("tree_sitter_cpp"))
+
+# extension -> (parser name, grammar loader)
+LANGUAGES: dict[str, tuple[str, Callable[[], Any]]] = {
+    **dict.fromkeys((".js", ".jsx", ".mjs", ".cjs"), _JAVASCRIPT),
+    **dict.fromkeys((".ts", ".mts", ".cts"), _TYPESCRIPT),
+    ".tsx": ("tsx", _grammar("tree_sitter_typescript", "language_tsx")),
+    ".go": ("go", _grammar("tree_sitter_go")),
+    ".rs": ("rust", _grammar("tree_sitter_rust")),
+    ".java": ("java", _grammar("tree_sitter_java")),
+    ".c": _C,
+    # .h is C or C++: parse as C++ (a near-superset), retry as C (_RETRY).
+    **dict.fromkeys(
+        (
+            ".h",
+            ".cc",
+            ".cpp",
+            ".cxx",
+            ".c++",
+            ".hh",
+            ".hpp",
+            ".hxx",
+            ".h++",
+            ".ipp",
+            ".tpp",
+            ".inl",
+        ),
+        _CPP,
+    ),
+    ".cs": ("csharp", _grammar("tree_sitter_c_sharp")),
+    **dict.fromkeys(
+        (".rb", ".rake", ".gemspec"), ("ruby", _grammar("tree_sitter_ruby"))
+    ),
+    ".php": ("php", _grammar("tree_sitter_php", "language_php")),
+    **dict.fromkeys((".kt", ".kts"), ("kotlin", _grammar("tree_sitter_kotlin"))),
+    ".swift": ("swift", _grammar("tree_sitter_swift")),
+    **dict.fromkeys((".scala", ".sc"), ("scala", _grammar("tree_sitter_scala"))),
+    **dict.fromkeys((".sh", ".bash"), ("bash", _grammar("tree_sitter_bash"))),
+    ".lua": ("lua", _grammar("tree_sitter_lua")),
+}
+# Tried in order when the primary grammar leaves parse errors.
+_RETRY: dict[str, tuple[tuple[str, Callable[[], Any]], ...]] = {".h": (_C,)}
+
+# Declarations split into a header plus one range per member.
+_CONTAINERS = frozenset(
+    {
+        "class_declaration",  # Java, C#, PHP, Kotlin, Swift (class/struct/extension)
+        "abstract_class_declaration",
+        "class",  # JS class expressions, Ruby
+        "enum_declaration",
+        "record_declaration",
+        "struct_declaration",  # C#
+        "impl_item",  # Rust
+        "trait_item",
+        "mod_item",
+        "class_specifier",  # C++ (C structs are data and stay whole)
+        "module",  # Ruby
+        "trait_declaration",  # PHP
+        "object_declaration",  # Kotlin
+        "companion_object",
+        "class_definition",  # Scala
+        "object_definition",
+        "trait_definition",
+    }
+)
+# Bodies: of functions (statements) and of containers (members).
+_BLOCKS = frozenset(
+    {
+        "statement_block",
+        "block",
+        "class_body",
+        "declaration_list",
+        "enum_body",
+        "switch_body",
+        "compound_statement",  # C, C++, PHP, Bash
+        "body_statement",  # Ruby
+        "function_body",  # Kotlin, Swift
+        "template_body",  # Scala
+        "enum_class_body",  # Kotlin
+    }
+)
+# A body that holds its statements in one wrapper node (Go, Swift, Kotlin).
+_STATEMENT_WRAPPERS = frozenset(
+    {"statement_list", "statements", "block", "body_statement"}
+)
+# Unwrapped wherever they appear: they scope code rather than being code.
+# A namespace or include guard often wraps a whole file; kept, it would make
+# the file one range.
+_TRANSPARENT = frozenset(
+    {
+        "preproc_ifdef",
+        "preproc_if",
+        "preproc_else",
+        "preproc_elif",
+        "preproc_elifdef",
+        "namespace_definition",  # C++, PHP (braced)
+        "namespace_declaration",  # C#
+        "file_scoped_namespace_declaration",  # C# `namespace X;`
+        "linkage_specification",  # extern "C" { ... }
+        "declaration_list",  # the body of the three above
+    }
+)
+_TRANSPARENT_FIELDS = ("name", "condition", "value")  # the wrapper's own label
+_WRAPPERS = frozenset({"export_statement", "template_declaration"})
+# Never named: an imported name would earn BM25's symbol boost for a line
+# that is almost never the answer (Python's ast leaves imports unnamed too).
+_UNNAMED = frozenset(
+    {
+        "import_declaration",
+        "package_declaration",
+        "use_declaration",
+        "preproc_include",
+        "using_directive",  # C#
+        "namespace_use_declaration",  # PHP
+        "import",  # Kotlin
+        "package_header",
+        "package_clause",  # Scala
+        "command",  # Bash: a call site, not a declaration
+    }
+)
+_DECLARATORS = frozenset({"lexical_declaration", "variable_declaration"})
+# Where a C/C++ declarator chain ends in a name.
+_DECLARATOR_NAMES = frozenset(
+    {
+        "identifier",
+        "field_identifier",
+        "type_identifier",
+        "qualified_identifier",
+        "destructor_name",
+        "operator_name",
+    }
+)
+_QUALIFIED = frozenset({"qualified_identifier", "method_index_expression"})
+_SKIPPED = frozenset({"access_specifier"})  # C++ `public:` labels
+_BODY_SEARCH_DEPTH = 4
+_DECLARATOR_DEPTH = 8
+_MAX_ABSORBED_TAIL = 3  # lines: a closing brace or #endif, not a comment block
+
+# py-tree-sitter 0.26.0 segfaults walking ordinary trees (reproduced on
+# CPython 3.13 and 3.14; 0.25.2 is clean on the same input). A native crash
+# would kill the whole session, so refuse it even if something pins it.
+_BROKEN_TREE_SITTER = ((0, 26),)
+
+
+def _tree_sitter_is_safe() -> bool:
+    try:
+        major, minor = (int(p) for p in version("tree-sitter").split(".")[:2])
+    except (PackageNotFoundError, ValueError):
+        return False
+    return (major, minor) not in _BROKEN_TREE_SITTER
+
+
+def _code_children(node: Any) -> list[Any]:
+    """Named children minus comments and labels. A skipped leading comment is
+    absorbed into the declaration after it (as Python's ``ast``, which has no
+    comment nodes, already behaves)."""
+    return [
+        c
+        for c in node.named_children
+        if "comment" not in c.type and c.type not in _SKIPPED
+    ]
+
+
+def _flatten(nodes: Iterable[Any]) -> list[Any]:
+    """Replace namespaces, include guards and ``extern "C"`` blocks with the
+    declarations inside them. Their own lines are absorbed by the neighbouring
+    ranges, so line coverage is unchanged."""
+    out: list[Any] = []
+    for node in nodes:
+        if node.type not in _TRANSPARENT:
+            out.append(node)
+            continue
+        labels = {
+            (c.start_byte, c.end_byte)
+            for field in _TRANSPARENT_FIELDS
+            if (c := node.child_by_field_name(field)) is not None
+        }
+        inner = [
+            c for c in _code_children(node) if (c.start_byte, c.end_byte) not in labels
+        ]
+        out.extend(_flatten(inner))
+    return out
+
+
+class TreeSitterSyntax:
+    """:class:`~.structure.Syntax` over any tree-sitter grammar.
+
+    Node text is sliced from the ``source`` bytes that were parsed, by byte
+    offset, so names come from exactly the input we hold.
+    """
+
+    def __init__(self, source: bytes) -> None:
+        self._source = source
+
+    def _text(self, node: Any) -> str:
+        return self._source[node.start_byte : node.end_byte].decode(
+            "utf-8", errors="replace"
+        )
+
+    @staticmethod
+    def _inner(node: Any) -> Any:
+        """The declaration an ``export`` or ``template<...>`` wraps (the wrapper
+        keeps the span)."""
+        if node.type == "export_statement":
+            inner = node.child_by_field_name("declaration")
+            if inner is not None:
+                return inner
+        elif node.type == "template_declaration":
+            children = _code_children(node)
+            if children and children[-1].type != "template_parameter_list":
+                return children[-1]
+        return node
+
+    def span(self, node: Any) -> tuple[int, int]:
+        start = node.start_point.row + 1
+        end = node.end_point.row + (1 if node.end_point.column else 0)
+        return start, max(start, end)
+
+    def _name(self, node: Any) -> str:
+        text = self._text(node)
+        if node.type in _QUALIFIED:  # C++ Server::port, Lua Account:deposit
+            text = text.replace("::", ".").replace(":", ".")
+        return text
+
+    def symbol(self, node: Any) -> str | None:
+        node = self._inner(node)
+        if node.type in _UNNAMED:
+            return None
+        if node.type == "impl_item":  # `impl Cache` / `impl Trait for Cache`
+            target = node.child_by_field_name("type")
+            return self._text(target) if target is not None else None
+        name = node.child_by_field_name("name")
+        if name is None:
+            declared = self._declarator_name(node)  # C/C++: int *make_buf(...)
+            if declared:
+                return declared
+            if node.type in _DECLARATORS or node.type.endswith("_declaration"):
+                return self._declared_names(node)  # const a, b / Go type / Java field
+            return None
+        receiver = node.child_by_field_name("receiver")  # Go: (s *Server) Serve
+        if receiver is not None:
+            owner = self._receiver_type(receiver)
+            if owner:
+                return f"{owner}.{self._text(name)}"
+        return self._name(name)
+
+    def _declarator_name(self, node: Any) -> str | None:
+        """Follow a C/C++ declarator chain (pointer -> function -> name)."""
+        current = node.child_by_field_name("declarator")
+        for _ in range(_DECLARATOR_DEPTH):
+            if current is None:
+                return None
+            if current.type in _DECLARATOR_NAMES:
+                return self._name(current)
+            children = current.named_children
+            current = current.child_by_field_name("declarator") or (
+                children[0] if children else None
+            )
+        return None
+
+    def _declared_names(self, node: Any) -> str | None:
+        """Names one level down: ``const a, b``, Go ``type_spec``, Java fields."""
+        candidates = [node.child_by_field_name("declarator"), *node.named_children]
+        names = dict.fromkeys(
+            self._text(name)
+            for child in candidates
+            if child is not None and (name := child.child_by_field_name("name"))
+        )
+        return ", ".join(names) or None
+
+    def _receiver_type(self, receiver: Any) -> str | None:
+        for param in receiver.named_children:
+            kind = param.child_by_field_name("type")
+            if kind is not None:
+                return self._text(kind).lstrip("*&").split("[", 1)[0] or None
+        return None
+
+    def members(self, node: Any) -> Sequence[Any] | None:
+        node = self._inner(node)
+        if node.type not in _CONTAINERS:
+            return None
+        body = node.child_by_field_name("body") or next(
+            (c for c in node.named_children if c.type in _BLOCKS), None
+        )
+        members = _flatten(_code_children(body)) if body is not None else []
+        # Only one-line members (fields, enum constants, prototypes, interface
+        # signatures): keep the declaration whole. A range per line would cost
+        # a judgment each and hold no behaviour.
+        if all(self.span(m)[1] - self.span(m)[0] < 1 for m in members):
+            return None
+        return members
+
+    def broken(self, node: Any) -> bool:
+        return node.type == "ERROR" or node.has_error
+
+    def statements(self, node: Any) -> Sequence[Any] | None:
+        block = _body_block(self._inner(node))
+        if block is None:
+            return None
+        children = _code_children(block)
+        for _ in range(_BODY_SEARCH_DEPTH):  # function_body -> block -> statements
+            if len(children) != 1 or children[0].type not in _STATEMENT_WRAPPERS:
+                break
+            children = _code_children(children[0])
+        return children or None
+
+
+def _body_block(node: Any) -> Any | None:
+    """The nearest block under ``node``: its own body, or one inside a wrapper
+    (``const f = () => {...}``, ``app.get("/x", (req, res) => {...})``)."""
+    if node.type in _BLOCKS:
+        return node
+    body = node.child_by_field_name("body")
+    if body is not None and body.type in _BLOCKS:
+        return body
+    frontier = list(node.named_children)
+    for _ in range(_BODY_SEARCH_DEPTH):
+        for child in frontier:
+            if child.type in _BLOCKS:
+                return child
+        frontier = [grand for child in frontier for grand in child.named_children]
+    return None
+
+
+@cache
+def _parser(loader: Callable[[], Any]) -> Any | None:
+    if not _tree_sitter_is_safe():
+        return None
+    try:
+        from tree_sitter import Language, Parser
+
+        return Parser(Language(loader()))
+    except Exception:  # noqa: BLE001 - unavailable grammar => line windows
+        return None
+
+
+def treesitter_ranges(text: str, path: str) -> tuple[list[Range], str] | None:
+    """(ranges, parser name), or None when this file should use line windows."""
+    ext = os.path.splitext(path)[1].lower()
+    entry = LANGUAGES.get(ext)
+    if entry is None:
+        return None
+    source = text.encode("utf-8")
+    syntax = TreeSitterSyntax(source)
+    # First clean parse wins; otherwise the tree with the fewest broken lines.
+    best: tuple[int, str, Any, list[Any]] | None = None
+    for name, loader in (entry, *_RETRY.get(ext, ())):
+        parser = _parser(loader)
+        if parser is None:
+            continue
+        root = parser.parse(source).root_node
+        nodes = _flatten(_code_children(root))
+        broken = sum(
+            syntax.span(n)[1] - syntax.span(n)[0] + 1 for n in nodes if syntax.broken(n)
+        )
+        if best is None or broken < best[0]:
+            best = (broken, name, root, nodes)
+        if not broken:
+            break
+    if best is None:
+        return None
+    broken, name, root, nodes = best
+    ranges = structure_ranges(nodes, syntax)
+    if not ranges:
+        return None
+    # Close the last range over a short tail (a namespace's closing brace,
+    # an include guard's #endif) instead of leaving it to become a
+    # judgment-costing snippet of its own. Longer tails stay separate.
+    tail = syntax.span(root)[1]
+    if 0 < tail - ranges[-1].end <= _MAX_ABSORBED_TAIL:
+        ranges[-1].end = tail
+    return ranges, f"{name}-partial" if broken else name

{code_puppy_core_plugins-0.0.72 → code_puppy_core_plugins-0.0.74}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "code-puppy-core-plugins"
-version = "0.0.72"
+version = "0.0.74"
 description = "Official plugin bundle for Code Puppy"
 readme = "README.md"
 requires-python = ">=3.11,<3.15"
@@ -15,10 +15,20 @@ dependencies = [
     # smart_grep syntax-aware chunking. <0.26: py-tree-sitter 0.26.0
     # segfaults walking ordinary trees (see jev_grep/treesitter.py).
     "tree-sitter>=0.25.2,<0.26",
+    "tree-sitter-bash>=0.25.1",
+    "tree-sitter-c>=0.24.2",
+    "tree-sitter-c-sharp>=0.23.5",
+    "tree-sitter-cpp>=0.23.4",
     "tree-sitter-go>=0.25.0",
     "tree-sitter-java>=0.23.5",
     "tree-sitter-javascript>=0.25.0",
+    "tree-sitter-kotlin>=1.1.0",
+    "tree-sitter-lua>=0.5.0",
+    "tree-sitter-php>=0.24.1",
+    "tree-sitter-ruby>=0.23.1",
     "tree-sitter-rust>=0.24.2",
+    "tree-sitter-scala>=0.26.2",
+    "tree-sitter-swift>=0.7.3",
     "tree-sitter-typescript>=0.23.2",
 ]
 license = {text = "MIT"}