generator.py

# Copyright 2018 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import jinja2
import yaml
import itertools
import re
import os
import pathlib
import typing
from typing import Any, DefaultDict, Dict, Mapping, Optional, Tuple
from hashlib import sha256
from collections import OrderedDict, defaultdict
from gapic.samplegen_utils.utils import (
    coerce_response_name,
    is_valid_sample_cfg,
    render_format_string,
)
from gapic.samplegen_utils.types import DuplicateSample
from gapic.samplegen_utils import snippet_index, snippet_metadata_pb2
from gapic.samplegen import manifest, samplegen
from gapic.generator import formatter
from gapic.schema import api
from gapic import utils
from gapic.utils import Options
from google.protobuf.compiler.plugin_pb2 import CodeGeneratorResponse


class Generator:
    """A protoc code generator for client libraries.

    This class provides an interface for getting a
    :class:`~.plugin_pb2.CodeGeneratorResponse` for an :class:`~api.API`
    schema object (which it does through rendering templates).

    Args:
        opts (~.options.Options): An options instance.
        templates (str): Optional. Path to the templates to be
            rendered. If this is not provided, the templates included with
            this application are used.
    """

    def __init__(self, opts: Options) -> None:
        # Create the jinja environment with which to render templates.
        self._env = jinja2.Environment(
            loader=jinja2.FileSystemLoader(searchpath=opts.templates),
            undefined=jinja2.StrictUndefined,
            extensions=["jinja2.ext.do"],
            trim_blocks=True,
            lstrip_blocks=True,
        )

        # Add filters which templates require.
        self._env.filters["rst"] = utils.rst
        self._env.filters["make_private"] = utils.make_private
        self._env.filters["snake_case"] = utils.to_snake_case
        self._env.filters["camel_case"] = utils.to_camel_case
        self._env.filters["sort_lines"] = utils.sort_lines
        self._env.filters["wrap"] = utils.wrap
        self._env.filters["coerce_response_name"] = coerce_response_name
        self._env.filters["render_format_string"] = render_format_string

        # Add tests to determine type of expressions stored in strings
        self._env.tests["str_field_pb"] = utils.is_str_field_pb
        self._env.tests["msg_field_pb"] = utils.is_msg_field_pb

        self._sample_configs = opts.sample_configs

    def get_response(self, api_schema: api.API, opts: Options) -> CodeGeneratorResponse:
        """Return a :class:`~.CodeGeneratorResponse` for this library.

        This is a complete response to be written to (usually) stdout, and
        thus read by ``protoc``.

        Args:
            api_schema (~api.API): An API schema object.
            opts (~.options.Options): An options instance.

        Returns:
            ~.CodeGeneratorResponse: A response describing appropriate
            files and contents. See ``plugin.proto``.
        """
        output_files: Dict[str, CodeGeneratorResponse.File] = OrderedDict()
        sample_templates, client_templates = utils.partition(
            lambda fname: os.path.basename(fname) == samplegen.DEFAULT_TEMPLATE_NAME,
            self._env.loader.list_templates(),  # type: ignore
        )

        # We generate code snippets *before* the library code so snippets
        # can be inserted into method docstrings.
        snippet_idx = snippet_index.SnippetIndex(api_schema)
        if sample_templates:
            sample_output, snippet_idx = self._generate_samples_and_manifest(
                api_schema,
                snippet_idx,
                self._env.get_template(sample_templates[0]),
                opts=opts,
            )
            output_files.update(sample_output)

        # Iterate over each template and add the appropriate output files
        # based on that template.
        # Sample templates work differently: there's (usually) only one,
        # and instead of iterating over it/them, we iterate over samples
        # and plug those into the template.
        for template_name in client_templates:
            # Quick check: Skip "private" templates.
            filename = template_name.split("/")[-1]
            if filename.startswith("_") and filename != "__init__.py.j2":
                continue

            # Append to the output files dictionary.
            output_files.update(
                self._render_template(
                    template_name,
                    api_schema=api_schema,
                    opts=opts,
                    snippet_index=snippet_idx,
                )
            )

        # Return the CodeGeneratorResponse output.
        res = CodeGeneratorResponse(file=[i for i in output_files.values()])  # type: ignore
        res.supported_features |= CodeGeneratorResponse.Feature.FEATURE_PROTO3_OPTIONAL  # type: ignore
        return res

    def _generate_samples_and_manifest(
        self,
        api_schema: api.API,
        index: snippet_index.SnippetIndex,
        sample_template: jinja2.Template,
        *,
        opts: Options,
    ) -> Tuple[Dict, snippet_index.SnippetIndex]:
        """Generate samples and samplegen manifest for the API.

        Arguments:
            api_schema (api.API): The schema for the API to which the samples belong.
            sample_template (jinja2.Template): The template to use to generate samples.
            opts (Options): Additional generator options.

        Returns:
            Tuple[Dict[str, CodeGeneratorResponse.File], snippet_index.SnippetIndex] : A dict mapping filepath to rendered file.
        """
        # The two-layer data structure lets us do two things:
        # * detect duplicate samples, which is an error
        # * detect distinct samples with the same ID, which are disambiguated
        id_to_hash_to_spec: DefaultDict[str, Dict[str, Any]] = defaultdict(dict)

        # Autogenerated sample specs
        autogen_specs: typing.List[typing.Dict[str, Any]] = []
        if opts.autogen_snippets:
            autogen_specs = list(samplegen.generate_sample_specs(api_schema, opts=opts))

        # Also process any handwritten sample specs
        handwritten_specs = samplegen.parse_handwritten_specs(self._sample_configs)

        sample_specs = autogen_specs + list(handwritten_specs)

        for spec in sample_specs:
            # Every sample requires an ID. This may be provided
            # by a samplegen config author.
            # If no ID is provided, fall back to the region tag.
            #
            # Ideally the sample author should pick a descriptive, unique ID,
            # but this may be impractical and can be error-prone.
            spec_hash = sha256(str(spec).encode("utf8")).hexdigest()[:8]
            sample_id = spec.get("id") or spec.get("region_tag") or spec_hash
            spec["id"] = sample_id

            hash_to_spec = id_to_hash_to_spec[sample_id]

            if spec_hash in hash_to_spec:
                raise DuplicateSample(f"Duplicate samplegen spec found: {spec}")

            hash_to_spec[spec_hash] = spec

        out_dir = "samples/generated_samples"
        fpath_to_spec_and_rendered = {}
        for hash_to_spec in id_to_hash_to_spec.values():
            for spec_hash, spec in hash_to_spec.items():
                id_is_unique = len(hash_to_spec) == 1
                # The ID is used to generate the file name. It must be globally unique.
                if not id_is_unique:
                    spec["id"] += f"_{spec_hash}"

                sample, snippet_metadata = samplegen.generate_sample(
                    spec,
                    api_schema,
                    sample_template,
                )

                fpath = utils.to_snake_case(spec["id"]) + ".py"
                fpath_to_spec_and_rendered[os.path.join(out_dir, fpath)] = (
                    spec,
                    sample,
                )

                snippet_metadata.file = fpath
                snippet_metadata.title = fpath

                index.add_snippet(snippet_index.Snippet(sample, snippet_metadata))

        output_files = {
            fname: CodeGeneratorResponse.File(
                content=formatter.fix_whitespace(sample), name=fname
            )
            for fname, (_, sample) in fpath_to_spec_and_rendered.items()
        }

        if index.metadata_index.snippets:
            # NOTE(busunkim): Not all fields are yet populated in the snippet metadata.
            # Expected filename: snippet_metadata.{proto_package}.json
            # For example: snippet_metadata_google.cloud.aiplatform.v1.json
            snippet_metadata_path = str(
                pathlib.Path(out_dir)
                / f"snippet_metadata_{api_schema.naming.proto_package}.json"
            ).lower()
            output_files[snippet_metadata_path] = CodeGeneratorResponse.File(
                content=formatter.fix_whitespace(index.get_metadata_json()),
                name=snippet_metadata_path,
            )

        return output_files, index

    def _render_template(
        self,
        template_name: str,
        *,
        api_schema: api.API,
        opts: Options,
        snippet_index: snippet_index.SnippetIndex,
    ) -> Dict[str, CodeGeneratorResponse.File]:
        """Render the requested templates.

        Args:
            template_name (str): The template to be rendered.
                It is expected that these come from
                :class:`jinja2.FileSystemLoader`, and they should be
                able to be sent to the :meth:`jinja2.Environment.get_template`
                method.
            api_schema (~.api.API): An API schema object.

        Returns:
            Sequence[~.CodeGeneratorResponse.File]: A sequence of File
                objects for inclusion in the final response.
        """
        answer: Dict[str, CodeGeneratorResponse.File] = OrderedDict()
        skip_subpackages = False

        # Very, very special case. This flag exists to gate this one file.
        if not opts.metadata and template_name.endswith("gapic_metadata.json.j2"):
            return answer

        # Disables generation of an unversioned Python package for this client
        # library. This means that the module names will need to be versioned in
        # import statements. For example `import google.cloud.library_v2` instead
        # of `import google.cloud.library`.
        if (
            template_name.startswith("%namespace/%name/")
            and api_schema.all_library_settings[
                api_schema.naming.proto_package
            ].python_settings.experimental_features.unversioned_package_disabled
        ):
            return answer

        # Quick check: Rendering per service and per proto would be a
        # combinatorial explosion and is almost certainly not what anyone
        # ever wants. Error colorfully on it.
        if "%service" in template_name and "%proto" in template_name:
            raise ValueError(
                "Template files may live under a %proto or "
                "%service directory, but not both."
            )

        # If this template should be rendered for subpackages, process it
        # for all subpackages and set the strict flag (restricting what
        # services and protos we pull from for the remainder of the method).
        if "%sub" in template_name:
            for subpackage in api_schema.subpackages.values():
                answer.update(
                    self._render_template(
                        template_name,
                        api_schema=subpackage,
                        opts=opts,
                        snippet_index=snippet_index,
                    )
                )
                skip_subpackages = True

        # If this template should be rendered once per proto, iterate over
        # all protos to be rendered
        if "%proto" in template_name:
            for proto in api_schema.protos.values():
                if (
                    skip_subpackages
                    and proto.meta.address.subpackage != api_schema.subpackage_view
                ):
                    continue

                answer.update(
                    self._get_file(
                        template_name,
                        api_schema=api_schema,
                        proto=proto,
                        opts=opts,
                        snippet_index=snippet_index,
                    )
                )

            return answer

        # If this template should be rendered once per service, iterate
        # over all services to be rendered.
        if "%service" in template_name:
            for service in api_schema.services.values():
                if (
                    (
                        skip_subpackages
                        and service.meta.address.subpackage
                        != api_schema.subpackage_view
                    )
                    or (
                        "transport" in template_name
                        and not self._is_desired_transport(template_name, opts)
                    )
                    or
                    # TODO(https://github.com/googleapis/gapic-generator-python/issues/2121): Remove this condition when async rest is GA.
                    (
                        "async_client" in template_name
                        and "grpc" not in opts.transport
                        and not api_schema.all_library_settings[
                            api_schema.naming.proto_package
                        ].python_settings.experimental_features.rest_async_io_enabled
                    )
                    or (
                        "rest_asyncio" in template_name
                        and not api_schema.all_library_settings[
                            api_schema.naming.proto_package
                        ].python_settings.experimental_features.rest_async_io_enabled
                    )
                    or ("rest_base" in template_name and "rest" not in opts.transport)
                ):
                    continue

                answer.update(
                    self._get_file(
                        template_name,
                        api_schema=api_schema,
                        service=service,
                        opts=opts,
                        snippet_index=snippet_index,
                    )
                )
            return answer

        # This file is not iterating over anything else; return back
        # the one applicable file.
        answer.update(
            self._get_file(
                template_name,
                api_schema=api_schema,
                opts=opts,
                snippet_index=snippet_index,
            )
        )
        return answer

    def _is_desired_transport(self, template_name: str, opts: Options) -> bool:
        """Returns true if template name contains a desired transport"""
        desired_transports = ["__init__", "base", "README"] + opts.transport
        return any(transport in template_name for transport in desired_transports)

    def _get_file(
        self,
        template_name: str,
        *,
        opts: Options,
        api_schema: api.API,
        **context,
    ):
        """Render a template to a protobuf plugin File object."""
        # Determine the target filename.
        fn = self._get_filename(
            template_name,
            api_schema=api_schema,
            context=context,
        )

        # Render the file contents.
        cgr_file = CodeGeneratorResponse.File(
            content=formatter.fix_whitespace(
                self._env.get_template(template_name).render(
                    api=api_schema, opts=opts, **context
                ),
            ),
            name=fn,
        )

        # Quick check: Do not render empty files.
        if utils.empty(cgr_file.content) and not fn.endswith(
            ("py.typed", "__init__.py")
        ):
            return {}

        # Return the filename and content in a length-1 dictionary
        # (because we track output files overall in a dictionary).
        return {fn: cgr_file}

    def _get_filename(
        self,
        template_name: str,
        *,
        api_schema: api.API,
        context: Optional[dict] = None,
    ) -> str:
        """Return the appropriate output filename for this template.

        This entails running the template name through a series of
        replacements to replace the "filename variables" (``%name``,
        ``%service``, etc.).

        Additionally, any of these variables may be substituted with an
        empty value, and we should do the right thing in this case.
        (The exception to this is ``%service``, which is guaranteed to be
        set if it is needed.)

        Args:
            template_name (str): The filename of the template, from the
                filesystem, relative to ``templates/``.
            api_schema (~.api.API): An API schema object.
            context (Mapping): Additional context being sent to the template.

        Returns:
            str: The appropriate output filename.
        """
        filename = template_name[: -len(".j2")]

        # Replace the %namespace variable.
        filename = filename.replace(
            "%namespace",
            os.path.sep.join(i.lower() for i in api_schema.naming.namespace),
        ).lstrip(os.path.sep)

        # Replace the %name, %version, and %sub variables.
        filename = filename.replace(
            "%name_%version", api_schema.naming.versioned_module_name
        )
        filename = filename.replace("%version", api_schema.naming.version)
        filename = filename.replace("%name", api_schema.naming.module_name)
        filename = filename.replace("%sub", "/".join(api_schema.subpackage_view))

        # Replace the %service variable if applicable.
        if context and "service" in context:
            filename = filename.replace(
                "%service",
                context["service"].module_name,
            )

        # Replace the %proto variable if appliable.
        # In the cases of protos, we also honor subpackages.
        if context and "proto" in context:
            filename = filename.replace(
                "%proto",
                context["proto"].module_name,
            )

        # Paths may have empty path segments if components are empty
        # (e.g. no %version); handle this.
        filename = re.sub(r"/+", "/", filename)

        # Done, return the filename.
        return filename


__all__ = ("Generator",)