Optionally skip normalize schema during marshal

[illusional]

Hey! I think we're duplicating normalize_schema work when parsing data from requests and sending the response. We normalize all the relevant schemas on url handler creation:

flask-rebar/flask_rebar/rebar.py

Lines 535 to 546 in 3229874

	headers_schema = normalize_schema(headers_schema)
	request_body_schema = normalize_schema(request_body_schema)
	query_string_schema = normalize_schema(query_string_schema)
	if response_body_schema:
	# Ensure we wrap in appropriate default (200) dict if we were passed a single Schema or class:
	if not isinstance(response_body_schema, Mapping):
	response_body_schema = {200: response_body_schema}
	# use normalize_schema to convert any class reference(s) to instantiated schema(s):
	response_body_schema = {
	code: normalize_schema(schema)
	for (code, schema) in response_body_schema.items()
	}

it looks like calls through the URL handlers are:

• calling marshal:

  flask-rebar/flask_rebar/utils/request_utils.py

  Lines 99 to 108 in 3229874

  	def marshal(data: Any, schema: Schema) -> Dict[str, Any]:
  	"""
  	Dumps an object with the given marshmallow.Schema.
  	:raises: marshmallow.ValidationError if the given data fails validation
  	of the schema.
  	"""
  	schema = normalize_schema(schema)
  	return compat.dump(schema=schema, data=data)

• and calling _get_data_or_400:

  flask-rebar/flask_rebar/utils/request_utils.py

  Lines 223 to 228 in 3229874

  	def _get_data_or_400(
  	schema: Schema, data: Any, message: messages.ErrorMessage
  	) -> Dict[str, Any]:
  	schema = normalize_schema(schema)
  	try:
  	return compat.load(schema=schema, data=data)

There is only one internal usage of marshal in our codebase so in my opinion should be safe to remove, but understand that others might be relying on it. Some other options include:

• creating a marshal_without_normalize
• (actioned) adding a should_normalize_schema: bool = True flag to marshal to preserve that existing functionality.

I think removing normalize_schema from the _get_data_or_400 is safer, but happy to consider other options too!

Edit: Running some future testing, it looks like normalizing already normalized schema is fairly quick.