feat: add autogenerated snippets (#845)

This PR targets iteration 1 and 2 specified in the [Snippet Gen Design](go/snippet-gen-design): Full canonical coverage of simple requests, paginated, LRO, server streaming, and Bidi streaming with empty request objects.

Snippet generation is hidden behind a new option `autogen-snippets`.

After discussion with folks on different language teams on snippetgen, I decided using "golden" snippet files would be easier than following the unit testing strategy used to check the library surface. I also believe goldens will be be easier for review for other Python DPEs.

Other notes:
- I've commented out the existing metadata generation code and tests. The new metadata format is still under discussion.
- Async samples are excluded as the existing samplegen infrastructure was written pre-async. I will add the async samples in the next PR.

Co-authored-by: Dov Shlachter <dovs@google.com>

Author: busunkim96

.github/snippet-bot.yml

@@ -0,0 +1,3 @@
+# https://github.com/googleapis/repo-automation-bots/tree/master/packages/snippet-bot
+ignoreFiles:
+  - "**/*.py"

.github/workflows/tests.yaml

@@ -265,6 +265,26 @@ jobs:
         run: python -m pip install nox
       - name: Typecheck the generated output.
         run: nox -s showcase_mypy${{ matrix.variant }}
+  snippetgen:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Cancel Previous Runs
+        uses: styfle/cancel-workflow-action@0.7.0
+        with:
+          access_token: ${{ github.token }}
+      - uses: actions/checkout@v2
+      - name: Set up Python 3.8
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.8
+      - name: Install system dependencies.
+        run: |
+            sudo apt-get update
+            sudo apt-get install -y curl pandoc unzip gcc
+      - name: Install nox.
+        run: python -m pip install nox
+      - name: Check autogenerated snippets.
+        run: nox -s snippetgen
   unit:
     strategy:
       matrix:

.gitignore

@@ -65,3 +65,7 @@ pylintrc.test
 
 # pyenv
 .python-version
+
+# Test dependencies and output
+api-common-protos
+tests/snippetgen/.test_output

gapic/generator/generator.py

@@ -14,8 +14,10 @@
 
 import jinja2
 import yaml
+import itertools
 import re
 import os
+import typing
 from typing import Any, DefaultDict, Dict, Mapping
 from hashlib import sha256
 from collections import OrderedDict, defaultdict
@@ -107,12 +109,12 @@ def get_response(
                     template_name, api_schema=api_schema, opts=opts)
             )
 
-        sample_output = self._generate_samples_and_manifest(
-            api_schema,
-            self._env.get_template(sample_templates[0]),
-        ) if sample_templates else {}
-
-        output_files.update(sample_output)
+        if sample_templates:
+            sample_output = self._generate_samples_and_manifest(
+                api_schema, self._env.get_template(sample_templates[0]),
+                opts=opts,
+                )
+            output_files.update(sample_output)
 
         # Return the CodeGeneratorResponse output.
         res = CodeGeneratorResponse(
@@ -121,12 +123,13 @@ def get_response(
         return res
 
     def _generate_samples_and_manifest(
-        self, api_schema: api.API, sample_template: jinja2.Template,
-    ) -> Dict[str, CodeGeneratorResponse.File]:
+            self, api_schema: api.API, sample_template: jinja2.Template, *, opts: Options) -> Dict:
         """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:
             Dict[str, CodeGeneratorResponse.File]: A dict mapping filepath to rendered file.
@@ -137,56 +140,50 @@ def _generate_samples_and_manifest(
         id_to_hash_to_spec: DefaultDict[str,
                                         Dict[str, Any]] = defaultdict(dict)
 
-        STANDALONE_TYPE = "standalone"
-        for config_fpath in self._sample_configs:
-            with open(config_fpath) as f:
-                configs = yaml.safe_load_all(f.read())
-
-            spec_generator = (
-                spec
-                for cfg in configs
-                if is_valid_sample_cfg(cfg)
-                for spec in cfg.get("samples", [])
-                # If unspecified, assume a sample config describes a standalone.
-                # If sample_types are specified, standalone samples must be
-                # explicitly enabled.
-                if STANDALONE_TYPE in spec.get("sample_type", [STANDALONE_TYPE])
-            )
+        # 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
 
-            for spec in spec_generator:
-                # Every sample requires an ID, preferably provided by the
-                # samplegen config author.
-                # If no ID is provided, fall back to the region tag.
-                # If there's no region tag, generate a unique ID.
-                #
-                # 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"
+            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 and by sample tester
-                # to link filenames to invoked samples. It must be globally unique.
+                # 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 = samplegen.generate_sample(
                     spec, api_schema, sample_template,)
 
-                fpath = spec["id"] + ".py"
+                fpath = utils.to_snake_case(spec["id"]) + ".py"
                 fpath_to_spec_and_rendered[os.path.join(out_dir, fpath)] = (
                     spec,
                     sample,
@@ -199,20 +196,24 @@ def _generate_samples_and_manifest(
             for fname, (_, sample) in fpath_to_spec_and_rendered.items()
         }
 
-        # Only generate a manifest if we generated samples.
-        if output_files:
-            manifest_fname, manifest_doc = manifest.generate(
-                (
-                    (fname, spec)
-                    for fname, (spec, _) in fpath_to_spec_and_rendered.items()
-                ),
-                api_schema,
-            )
-
-            manifest_fname = os.path.join(out_dir, manifest_fname)
-            output_files[manifest_fname] = CodeGeneratorResponse.File(
-                content=manifest_doc.render(), name=manifest_fname
-            )
+        # TODO(busunkim): Re-enable manifest generation once metadata
+        # format has been formalized.
+        # https://docs.google.com/document/d/1ghBam8vMj3xdoe4xfXhzVcOAIwrkbTpkMLgKc9RPD9k/edit#heading=h.sakzausv6hue
+        #
+        # if output_files:
+
+        # manifest_fname, manifest_doc = manifest.generate(
+        #     (
+        #         (fname, spec)
+        #         for fname, (spec, _) in fpath_to_spec_and_rendered.items()
+        #     ),
+        #     api_schema,
+        # )
+
+        # manifest_fname = os.path.join(out_dir, manifest_fname)
+        # output_files[manifest_fname] = CodeGeneratorResponse.File(
+        #     content=manifest_doc.render(), name=manifest_fname
+        # )
 
         return output_files