Files
capnpc-py/src/capnp_stubgen/plugin.py
T

247 lines
9.0 KiB
Python
Raw Normal View History

2026-07-09 15:06:39 +08:00
"""Cap'n Proto compiler plugin entry point.
This is the ``capnpc-py`` binary. When invoked by ``capnp compile -opy``,
it reads a ``CodeGeneratorRequest`` from stdin and writes ``.pyi`` type
stub files to the current working directory.
Uses ``.to_dict()`` on the request to avoid C++ API quirks with
DynamicStruct-wrapped schema nodes (same approach as pycapnp's own
``_gen.py``).
"""
from __future__ import annotations
import os
import sys
2026-07-09 15:54:59 +08:00
from typing import Any
2026-07-09 15:06:39 +08:00
from .emitter import Emitter
2026-07-09 15:21:04 +08:00
from .gen_const import generate_const
2026-07-09 15:06:39 +08:00
from .gen_enum import generate_enum
2026-07-09 15:39:10 +08:00
from .gen_interface import fixup_interface_methods, generate_interface
2026-07-09 15:06:39 +08:00
from .gen_struct import generate_struct
2026-07-09 15:54:59 +08:00
from .models import TypeInfo, TypeRegistry
2026-07-09 15:06:39 +08:00
from .schema_walker import build_type_registry
2026-07-09 15:54:59 +08:00
from .utils import NodeDict, filename_to_module_name, type_id_to_int
2026-07-09 15:06:39 +08:00
def main() -> int:
"""Entry point. Returns 0 on success, 1 on error."""
try:
return _main_impl()
except Exception as exc:
print(f"capnpc-py: error: {exc}", file=sys.stderr)
return 1
def _main_impl() -> int:
# ── 1. Read CodeGeneratorRequest from stdin ────────────────────────
request = _read_request()
# Convert to dict for processing (avoids C++ DynamicStruct quirks)
request_dict = request.to_dict()
2026-07-09 15:54:59 +08:00
nodes: list[NodeDict] = request_dict["nodes"]
requested_files: list[NodeDict] = request_dict["requestedFiles"]
2026-07-09 15:06:39 +08:00
# ── 2. Build type registry ─────────────────────────────────────────
registry = build_type_registry(nodes, requested_files)
2026-07-09 15:39:10 +08:00
# Attach method param/result structs to their interfaces
fixup_interface_methods(registry)
# Record file ID → module name mappings
# Register both requested files and their imports
for req_file in requested_files:
file_id = type_id_to_int(req_file["id"])
module_name = filename_to_module_name(req_file["filename"])
registry.set_file_module_name(file_id, module_name)
# Also register imported files
for imp in req_file.get("imports", []):
imp_id = type_id_to_int(imp["id"])
imp_name = filename_to_module_name(imp["name"])
registry.set_file_module_name(imp_id, imp_name)
2026-07-09 15:06:39 +08:00
# ── 3. Generate .pyi for each requested file ───────────────────────
ok = True
for req_file in requested_files:
filename: str = req_file["filename"]
file_id = type_id_to_int(req_file["id"])
try:
content = _generate_file(file_id, filename, registry)
module_name = filename_to_module_name(filename)
output_path = f"{module_name}.pyi"
with open(output_path, "w") as f:
2026-07-09 15:54:59 +08:00
_ = f.write(content)
2026-07-09 15:06:39 +08:00
except Exception as exc:
print(
f"capnpc-py: error generating stubs for {filename}: {exc}",
file=sys.stderr,
)
ok = False
return 0 if ok else 1
# ── request reading ────────────────────────────────────────────────────────
def _read_request() -> Any:
"""Read ``CodeGeneratorRequest`` from stdin.
Uses pycapnp's ``SchemaParser`` to dynamically load ``schema.capnp``
(shipped with pycapnp), then casts the stdin message.
"""
2026-07-09 15:54:59 +08:00
import capnp # pyright: ignore[reportMissingTypeStubs]
2026-07-09 15:06:39 +08:00
capnp_dir = os.path.dirname(capnp.__file__)
schema_path = os.path.join(capnp_dir, "schema.capnp")
import_base = os.path.dirname(capnp_dir)
2026-07-09 15:54:59 +08:00
parser = capnp.SchemaParser() # pyright: ignore[reportAttributeAccessIssue]
2026-07-09 15:06:39 +08:00
schema_module = parser.load(schema_path, imports=[import_base])
return schema_module.CodeGeneratorRequest.read(sys.stdin.buffer)
# ── file generation ────────────────────────────────────────────────────────
def _generate_file(
file_id: int,
filename: str,
registry: TypeRegistry,
) -> str:
"""Generate the complete .pyi content for one schema file."""
emitter = Emitter(source_filename=filename)
2026-07-09 15:39:10 +08:00
# Collect cross-file imports
_resolve_cross_imports(file_id, filename, registry, emitter)
2026-07-09 15:06:39 +08:00
top_level = registry.get_top_level_types(file_id)
_ORDER = {"enum": 0, "struct": 1, "const": 2, "interface": 3}
2026-07-09 15:54:59 +08:00
def _sort_key(info: TypeInfo) -> tuple[int, str]:
2026-07-09 15:06:39 +08:00
return (_ORDER.get(info.kind, 99), info.name)
top_level.sort(key=_sort_key)
for type_info in top_level:
if type_info.kind == "enum":
generate_enum(emitter, type_info)
elif type_info.kind == "struct":
generate_struct(emitter, type_info, registry)
2026-07-09 15:39:10 +08:00
elif type_info.kind == "interface":
generate_interface(emitter, type_info, registry)
2026-07-09 15:21:04 +08:00
elif type_info.kind == "const":
generate_const(emitter, type_info, registry)
2026-07-09 15:06:39 +08:00
return emitter.render()
2026-07-09 15:39:10 +08:00
def _resolve_cross_imports(
file_id: int,
filename: str,
registry: TypeRegistry,
emitter: Emitter,
) -> None:
"""Emit Python import lines for types referenced from other files.
Walks all types in *file_id*, collects field type references that
belong to a different file, and emits relative import lines.
"""
external_types: dict[str, set[str]] = {} # module_name → {TypeName, TypeNameBuilder, ...}
# Collect all types belonging to this file
own_types: set[int] = set()
2026-07-09 15:54:59 +08:00
for info in registry.all_types():
2026-07-09 15:39:10 +08:00
if info.file_id == file_id and info.kind in ("struct", "enum", "interface"):
own_types.add(info.type_id)
# Walk types and find external references
2026-07-09 15:54:59 +08:00
for info in registry.all_types():
2026-07-09 15:39:10 +08:00
if info.file_id != file_id:
continue
refs = _collect_type_references(info, registry)
for ref_id in refs:
if ref_id not in own_types:
ref_info = registry.get(ref_id)
if ref_info is not None and ref_info.file_id is not None:
ref_file_id = ref_info.file_id
# Get the module name for the referenced file
module_name = _file_id_to_module_name(ref_file_id, registry)
if module_name:
if module_name not in external_types:
external_types[module_name] = set()
external_types[module_name].add(ref_info.scoped_name)
# Emit import lines
for module_name, type_names in sorted(external_types.items()):
# Only emit relative imports (not self-imports)
own_module = filename_to_module_name(filename)
if module_name == own_module:
continue
# Collect all import names for this module
all_imports: list[str] = []
for tname in sorted(type_names):
ref_info = None
2026-07-09 15:54:59 +08:00
for info in registry.all_types():
2026-07-09 15:39:10 +08:00
if info.scoped_name == tname and info.file_id != file_id:
ref_info = info
break
all_imports.append(tname)
emitter.add_import(
f"from .{module_name} import {', '.join(all_imports)}"
)
def _collect_type_references(info: TypeInfo, registry: TypeRegistry) -> set[int]:
"""Collect all type IDs referenced by a TypeInfo's fields."""
refs: set[int] = set()
node = info.node
if info.kind == "struct":
for field in node.get("struct", {}).get("fields", []):
_collect_field_references(field, refs, registry)
elif info.kind == "interface":
for method in node.get("interface", {}).get("methods", []):
refs.add(type_id_to_int(method["paramStructType"]))
refs.add(type_id_to_int(method["resultStructType"]))
return refs
2026-07-09 15:54:59 +08:00
def _collect_field_references(field: NodeDict, refs: set[int], _registry: TypeRegistry) -> None:
2026-07-09 15:39:10 +08:00
"""Recursively collect type IDs from a field's type."""
if "slot" in field:
_collect_type_references_from_dict(field["slot"].get("type", {}), refs)
elif "group" in field:
refs.add(type_id_to_int(field["group"]["typeId"]))
2026-07-09 15:54:59 +08:00
def _collect_type_references_from_dict(type_dict: NodeDict, refs: set[int]) -> None:
2026-07-09 15:39:10 +08:00
"""Recursively collect type IDs from a type dict."""
if "struct" in type_dict:
refs.add(type_id_to_int(type_dict["struct"]["typeId"]))
elif "enum" in type_dict:
refs.add(type_id_to_int(type_dict["enum"]["typeId"]))
elif "interface" in type_dict:
refs.add(type_id_to_int(type_dict["interface"]["typeId"]))
elif "list" in type_dict:
_collect_type_references_from_dict(
type_dict["list"].get("elementType", {}), refs
)
def _file_id_to_module_name(file_id: int, registry: TypeRegistry) -> str | None:
"""Convert a file node ID to its Python module name."""
return registry.get_file_module_name(file_id)