zulip/zerver/openapi/openapi.py

483 lines
18 KiB
Python
Raw Normal View History

# Zulip's OpenAPI-based API documentation system is documented at
# https://zulip.readthedocs.io/en/latest/documentation/api.html
#
# This file contains helper functions to interact with the OpenAPI
# definitions and validate that Zulip's implementation matches what is
# described in our documentation.
import os
import re
from typing import Any, Dict, List, Optional, Set
from jsonschema.exceptions import ValidationError as JsonSchemaValidationError
from openapi_core import create_spec
from openapi_core.testing import MockRequest
from openapi_core.validation.request.validators import RequestValidator
from openapi_schema_validator import OAS30Validator
OPENAPI_SPEC_PATH = os.path.abspath(
os.path.join(os.path.dirname(__file__), "../openapi/zulip.yaml")
)
# A list of endpoint-methods such that the endpoint
# has documentation but not with this particular method.
EXCLUDE_UNDOCUMENTED_ENDPOINTS = {
("/realm/emoji/{emoji_name}", "delete"),
("/users", "patch"),
}
# Consists of endpoints with some documentation remaining.
# These are skipped but return true as the validator cannot exclude objects
EXCLUDE_DOCUMENTED_ENDPOINTS = {
("/settings/notifications", "patch"),
}
# Most of our code expects allOf to be preprocessed away because that is what
# yamole did. Its algorithm for doing so is not standards compliant, but we
# replicate it here.
def naively_merge(a: Dict[str, object], b: Dict[str, object]) -> Dict[str, object]:
ret: Dict[str, object] = a.copy()
for key, b_value in b.items():
if key == "example" or key not in ret:
ret[key] = b_value
continue
a_value = ret[key]
if isinstance(b_value, list):
assert isinstance(a_value, list)
ret[key] = a_value + b_value
elif isinstance(b_value, dict):
assert isinstance(a_value, dict)
ret[key] = naively_merge(a_value, b_value)
return ret
def naively_merge_allOf(obj: object) -> object:
if isinstance(obj, dict):
return naively_merge_allOf_dict(obj)
elif isinstance(obj, list):
return list(map(naively_merge_allOf, obj))
else:
return obj
def naively_merge_allOf_dict(obj: Dict[str, object]) -> Dict[str, object]:
if "allOf" in obj:
ret = obj.copy()
subschemas = ret.pop("allOf")
ret = naively_merge_allOf_dict(ret)
assert isinstance(subschemas, list)
for subschema in subschemas:
assert isinstance(subschema, dict)
ret = naively_merge(ret, naively_merge_allOf_dict(subschema))
return ret
return {key: naively_merge_allOf(value) for key, value in obj.items()}
class OpenAPISpec:
def __init__(self, openapi_path: str) -> None:
self.openapi_path = openapi_path
self.mtime: Optional[float] = None
self._openapi: Dict[str, Any] = {}
self._endpoints_dict: Dict[str, str] = {}
self._request_validator: Optional[RequestValidator] = None
def check_reload(self) -> None:
# Because importing yaml takes significant time, and we only
# use python-yaml for our API docs, importing it lazily here
# is a significant optimization to `manage.py` startup.
#
# There is a bit of a race here...we may have two processes
# accessing this module level object and both trying to
# populate self.data at the same time. Hopefully this will
# only cause some extra processing at startup and not data
# corruption.
import yaml
from jsonref import JsonRef
with open(self.openapi_path) as f:
mtime = os.fstat(f.fileno()).st_mtime
# Using == rather than >= to cover the corner case of users placing an
# earlier version than the current one
if self.mtime == mtime:
return
openapi = yaml.load(f, Loader=yaml.CSafeLoader)
spec = create_spec(openapi)
self._request_validator = RequestValidator(spec)
self._openapi = naively_merge_allOf_dict(JsonRef.replace_refs(openapi))
self.create_endpoints_dict()
self.mtime = mtime
def create_endpoints_dict(self) -> None:
# Algorithm description:
# We have 2 types of endpoints
# 1.with path arguments 2. without path arguments
# In validate_against_openapi_schema we directly check
# if we have a without path endpoint, since it does not
# require regex. Hence they are not part of the regex dict
# and now we are left with only:
# endpoint with path arguments.
# Now for this case, the regex has been created carefully,
# numeric arguments are matched with [0-9] only and
# emails are matched with their regex. This is why there are zero
# collisions. Hence if this regex matches
# an incorrect endpoint then there is some backend problem.
# For example if we have users/{name}/presence then it will
# conflict with users/me/presence even in the backend.
# Care should be taken though that if we have special strings
# such as email they must be substituted with proper regex.
email_regex = r"([a-zA-Z0-9_\-\.]+)@([a-zA-Z0-9_\-\.]+)\.([a-zA-Z]{2,5})"
self._endpoints_dict = {}
for endpoint in self._openapi["paths"]:
if "{" not in endpoint:
continue
path_regex = "^" + endpoint + "$"
# Numeric arguments have id at their end
# so find such arguments and replace them with numeric
# regex
path_regex = re.sub(r"{[^}]*id}", r"[0-9]*", path_regex)
# Email arguments end with email
path_regex = re.sub(r"{[^}]*email}", email_regex, path_regex)
# All other types of arguments are supposed to be
# all-encompassing string.
path_regex = re.sub(r"{[^}]*}", r"[^\/]*", path_regex)
path_regex = path_regex.replace(r"/", r"\/")
self._endpoints_dict[path_regex] = endpoint
def openapi(self) -> Dict[str, Any]:
"""Reload the OpenAPI file if it has been modified after the last time
it was read, and then return the parsed data.
"""
self.check_reload()
assert len(self._openapi) > 0
return self._openapi
def endpoints_dict(self) -> Dict[str, str]:
"""Reload the OpenAPI file if it has been modified after the last time
it was read, and then return the parsed data.
"""
self.check_reload()
assert len(self._endpoints_dict) > 0
return self._endpoints_dict
def request_validator(self) -> RequestValidator:
"""Reload the OpenAPI file if it has been modified after the last time
it was read, and then return the openapi_core validator object. Similar
to preceding functions. Used for proper access to OpenAPI objects.
"""
self.check_reload()
assert self._request_validator is not None
return self._request_validator
class SchemaError(Exception):
pass
openapi_spec = OpenAPISpec(OPENAPI_SPEC_PATH)
def get_schema(endpoint: str, method: str, status_code: str) -> Dict[str, Any]:
if len(status_code) == 3 and (
"oneOf"
in openapi_spec.openapi()["paths"][endpoint][method.lower()]["responses"][status_code][
"content"
]["application/json"]["schema"]
):
# Currently at places where multiple schemas are defined they only
# differ in example so either can be used.
status_code += "_0"
if len(status_code) == 3:
schema = openapi_spec.openapi()["paths"][endpoint][method.lower()]["responses"][
status_code
]["content"]["application/json"]["schema"]
return schema
else:
subschema_index = int(status_code[4])
status_code = status_code[0:3]
schema = openapi_spec.openapi()["paths"][endpoint][method.lower()]["responses"][
status_code
]["content"]["application/json"]["schema"]["oneOf"][subschema_index]
return schema
def get_openapi_fixture(endpoint: str, method: str, status_code: str = "200") -> Dict[str, Any]:
"""Fetch a fixture from the full spec object."""
return get_schema(endpoint, method, status_code)["example"]
def get_openapi_description(endpoint: str, method: str) -> str:
"""Fetch a description from the full spec object."""
return openapi_spec.openapi()["paths"][endpoint][method.lower()]["description"]
def get_openapi_summary(endpoint: str, method: str) -> str:
"""Fetch a summary from the full spec object."""
return openapi_spec.openapi()["paths"][endpoint][method.lower()]["summary"]
def get_openapi_paths() -> Set[str]:
return set(openapi_spec.openapi()["paths"].keys())
def get_openapi_parameters(
endpoint: str, method: str, include_url_parameters: bool = True
) -> List[Dict[str, Any]]:
operation = openapi_spec.openapi()["paths"][endpoint][method.lower()]
# We do a `.get()` for this last bit to distinguish documented
# endpoints with no parameters (empty list) from undocumented
# endpoints (KeyError exception).
parameters = operation.get("parameters", [])
# Also, we skip parameters defined in the URL.
if not include_url_parameters:
parameters = [parameter for parameter in parameters if parameter["in"] != "path"]
return parameters
def get_openapi_return_values(endpoint: str, method: str) -> List[Dict[str, Any]]:
operation = openapi_spec.openapi()["paths"][endpoint][method.lower()]
schema = operation["responses"]["200"]["content"]["application/json"]["schema"]
# In cases where we have used oneOf, the schemas only differ in examples
# So we can choose any.
if "oneOf" in schema:
schema = schema["oneOf"][0]
return schema["properties"]
def find_openapi_endpoint(path: str) -> Optional[str]:
for path_regex, endpoint in openapi_spec.endpoints_dict().items():
matches = re.match(path_regex, path)
if matches:
return endpoint
return None
def get_event_type(event: Dict[str, Any]) -> str:
return event["type"] + ":" + event.get("op", "")
def fix_events(content: Dict[str, Any]) -> None:
"""Remove undocumented events from events array. This is a makeshift
function so that further documentation of `/events` can happen with
only zulip.yaml changes and minimal other changes. It should be removed
as soon as `/events` documentation is complete.
"""
# 'user' is deprecated so remove its occurrences from the events array
for event in content["events"]:
event.pop("user", None)
def validate_against_openapi_schema(
content: Dict[str, Any],
path: str,
method: str,
status_code: str,
display_brief_error: bool = False,
) -> bool:
"""Compare a "content" dict with the defined schema for a specific method
in an endpoint. Return true if validated and false if skipped.
"""
# This first set of checks are primarily training wheels that we
# hope to eliminate over time as we improve our API documentation.
# No 500 responses have been documented, so skip them
if status_code.startswith("5"):
return False
if path not in openapi_spec.openapi()["paths"].keys():
endpoint = find_openapi_endpoint(path)
# If it doesn't match it hasn't been documented yet.
if endpoint is None:
return False
else:
endpoint = path
# Excluded endpoint/methods
if (endpoint, method) in EXCLUDE_UNDOCUMENTED_ENDPOINTS:
return False
# Return true for endpoints with only response documentation remaining
if (endpoint, method) in EXCLUDE_DOCUMENTED_ENDPOINTS:
return True
# Check if the response matches its code
if status_code.startswith("2") and (content.get("result", "success").lower() != "success"):
raise SchemaError("Response is not 200 but is validating against 200 schema")
# Code is not declared but appears in various 400 responses. If
# common, it can be added to 400 response schema
if status_code.startswith("4"):
# This return statement should ideally be not here. But since
# we have not defined 400 responses for various paths this has
# been added as all 400 have the same schema. When all 400
# response have been defined this should be removed.
return True
# The actual work of validating that the response matches the
# schema is done via the third-party OAS30Validator.
schema = get_schema(endpoint, method, status_code)
if endpoint == "/events" and method == "get":
# This a temporary function for checking only documented events
# as all events haven't been documented yet.
# TODO: Remove this after all events have been documented.
fix_events(content)
validator = OAS30Validator(schema)
try:
validator.validate(content)
except JsonSchemaValidationError as error:
if not display_brief_error:
raise error
# display_brief_error is designed to avoid printing 1000 lines
# of output when the schema to validate is extremely large
# (E.g. the several dozen format variants for individual
# events returned by GET /events) and instead just display the
# specific variant we expect to match the response.
brief_error_display_schema = {"nullable": False, "oneOf": list()}
brief_error_display_schema_oneOf = []
brief_error_validator_value = []
for validator_value in error.validator_value:
if validator_value["example"]["type"] == error.instance["type"]:
brief_error_validator_value.append(validator_value)
for i_schema in error.schema["oneOf"]:
if i_schema["example"]["type"] == error.instance["type"]:
brief_error_display_schema_oneOf.append(i_schema)
brief_error_display_schema["oneOf"] = brief_error_display_schema_oneOf
# Field list from https://python-jsonschema.readthedocs.io/en/stable/errors/
raise JsonSchemaValidationError(
message=error.message,
validator=error.validator,
path=error.path,
instance=error.instance,
schema_path=error.schema_path,
schema=brief_error_display_schema,
validator_value=brief_error_validator_value,
cause=error.cause,
)
return True
def validate_schema(schema: Dict[str, Any]) -> None:
"""Check if opaque objects are present in the OpenAPI spec; this is an
important part of our policy for ensuring every detail of Zulip's
API responses is correct.
This is done by checking for the presence of the
`additionalProperties` attribute for all objects (dictionaries).
"""
if "oneOf" in schema:
for subschema in schema["oneOf"]:
validate_schema(subschema)
elif schema["type"] == "array":
validate_schema(schema["items"])
elif schema["type"] == "object":
if "additionalProperties" not in schema:
raise SchemaError(
"additionalProperties needs to be defined for objects to make"
+ "sure they have no additional properties left to be documented."
)
for property_schema in schema.get("properties", {}).values():
validate_schema(property_schema)
if schema["additionalProperties"]:
validate_schema(schema["additionalProperties"])
def to_python_type(py_type: str) -> type:
"""Transform an OpenAPI-like type to a Python one.
https://swagger.io/docs/specification/data-models/data-types
"""
TYPES = {
"string": str,
"number": float,
"integer": int,
"boolean": bool,
"array": list,
"object": dict,
}
return TYPES[py_type]
def likely_deprecated_parameter(parameter_description: str) -> bool:
if "**Changes**: Deprecated" in parameter_description:
return True
return "**Deprecated**" in parameter_description
def check_deprecated_consistency(argument: Dict[str, Any], description: str) -> None:
# Test to make sure deprecated parameters are marked so.
if likely_deprecated_parameter(description):
assert argument["deprecated"]
if "deprecated" in argument:
assert likely_deprecated_parameter(description)
# Skip those JSON endpoints whose query parameters are different from
# their `/api/v1` counterpart. This is a legacy code issue that we
# plan to fix by changing the implementation.
SKIP_JSON = {
("/fetch_api_key", "post"),
}
def validate_request(
url: str,
method: str,
data: Dict[str, Any],
http_headers: Dict[str, Any],
json_url: bool,
status_code: str,
intentionally_undocumented: bool = False,
) -> None:
# Some JSON endpoints have different parameters compared to
# their `/api/v1` counterparts.
if json_url and (url, method) in SKIP_JSON:
return
# TODO: Add support for file upload endpoints that lack the /json/
# or /api/v1/ prefix.
if url == "/user_uploads" or url.startswith("/realm/emoji/"):
return
# Now using the openapi_core APIs, validate the request schema
# against the OpenAPI documentation.
mock_request = MockRequest(
"http://localhost:9991/", method, "/api/v1" + url, headers=http_headers, args=data
)
result = openapi_spec.request_validator().validate(mock_request)
if len(result.errors) != 0:
# Requests that do not validate against the OpenAPI spec must either:
# * Have returned a 400 (bad request) error
# * Have returned a 200 (success) with this request marked as intentionally
# undocumented behavior.
if status_code.startswith("4"):
return
if status_code.startswith("2") and intentionally_undocumented:
return
# If no errors are raised, then validation is successful
if len(result.errors) == 0:
return
# Show a block error message explaining the options for fixing it.
msg = f"""
Error! The OpenAPI schema for {method} {url} is not consistent
with the parameters passed in this HTTP request. Consider:
* Updating the OpenAPI schema defined in zerver/openapi/zulip.yaml
* Adjusting the test to pass valid parameters. If the test
fails due to intentionally_undocumented features, you need to pass
`intentionally_undocumented=True` to self.client_{method.lower()} or
self.api_{method.lower()} to document your intent.
See https://zulip.readthedocs.io/en/latest/documentation/api.html for help.
The errors logged by the OpenAPI validator are below:\n"""
for error in result.errors:
msg += f"* {str(error)}\n"
raise SchemaError(msg)