Advanced Usage#

This section includes guides for advanced usage patterns.

Custom Location Handlers#

To add your own custom location handler, write a function that receives a request, and a Schema, then decorate that function with Parser.location_loader.

from webargs import fields
from webargs.flaskparser import parser


@parser.location_loader("data")
def load_data(request, schema):
    return request.data


# Now 'data' can be specified as a location
@parser.use_args({"per_page": fields.Int()}, location="data")
def posts(args):
    return "displaying {} posts".format(args["per_page"])

Note

The schema is passed so that it can be used to wrap multidict types and unpack List fields correctly. If you are writing a loader for a multidict type, consider looking at MultiDictProxy for an example of how to do this.

“meta” Locations#

You can define your own locations which mix data from several existing locations.

The json_or_form location does this – first trying to load data as JSON and then falling back to a form body – and its implementation is quite simple:

def load_json_or_form(self, req, schema):
    """Load data from a request, accepting either JSON or form-encoded
    data.

    The data will first be loaded as JSON, and, if that fails, it will be
    loaded as a form post.
    """
    data = self.load_json(req, schema)
    if data is not missing:
        return data
    return self.load_form(req, schema)

You can imagine your own locations with custom behaviors like this. For example, to mix query parameters and form body data, you might write the following:

from webargs import fields
from webargs.multidictproxy import MultiDictProxy
from webargs.flaskparser import parser


@parser.location_loader("query_and_form")
def load_data(request, schema):
    # relies on the Flask (werkzeug) MultiDict type's implementation of
    # these methods, but when you're extending webargs, you may know things
    # about your framework of choice
    newdata = request.args.copy()
    newdata.update(request.form)
    return MultiDictProxy(newdata, schema)


# Now 'query_and_form' means you can send these values in either location,
# and they will be *mixed* together into a new dict to pass to your schema
@parser.use_args({"favorite_food": fields.String()}, location="query_and_form")
def set_favorite_food(args):
    ...  # do stuff
    return "your favorite food is now set to {}".format(args["favorite_food"])

marshmallow Integration#

When you need more flexibility in defining input schemas, you can pass a marshmallow Schema instead of a dictionary to Parser.parse, Parser.use_args, and Parser.use_kwargs.

from marshmallow import Schema, fields
from webargs.flaskparser import use_args


class UserSchema(Schema):
    id = fields.Int(dump_only=True)  # read-only (won't be parsed by webargs)
    username = fields.Str(required=True)
    password = fields.Str(load_only=True)  # write-only
    first_name = fields.Str(load_default="")
    last_name = fields.Str(load_default="")
    date_registered = fields.DateTime(dump_only=True)


@use_args(UserSchema())
def profile_view(args):
    username = args["username"]
    # ...


@use_kwargs(UserSchema())
def profile_update(username, password, first_name, last_name):
    update_profile(username, password, first_name, last_name)
    # ...


# You can add additional parameters
@use_kwargs({"posts_per_page": fields.Int(load_default=10)}, location="query")
@use_args(UserSchema())
def profile_posts(args, posts_per_page):
    username = args["username"]
    # ...

Setting `unknown`#

webargs supports several ways of setting and passing the unknown parameter for handling unknown fields.

You can pass unknown=... as a parameter to any of Parser.parse, Parser.use_args, and Parser.use_kwargs.

Note

The unknown value is passed to the schema’s load() call. It therefore only applies to the top layer when nesting is used. To control unknown at multiple layers of a nested schema, you must use other mechanisms, like the unknown argument to fields.Nested.

Default `unknown`#

By default, webargs will pass unknown=marshmallow.EXCLUDE except when the location is json, form, json_or_form, or path. In those cases, it uses unknown=marshmallow.RAISE instead.

You can change these defaults by overriding DEFAULT_UNKNOWN_BY_LOCATION. This is a mapping of locations to values to pass.

For example,

from flask import Flask
from marshmallow import EXCLUDE, fields
from webargs.flaskparser import FlaskParser

app = Flask(__name__)


class Parser(FlaskParser):
    DEFAULT_UNKNOWN_BY_LOCATION = {"query": EXCLUDE}


parser = Parser()


# location is "query", which is listed in DEFAULT_UNKNOWN_BY_LOCATION,
# so EXCLUDE will be used
@app.route("/", methods=["GET"])
@parser.use_args({"foo": fields.Int()}, location="query")
def get(args):
    return f"foo x 2 = {args['foo'] * 2}"


# location is "json", which is not in DEFAULT_UNKNOWN_BY_LOCATION,
# so no value will be passed for `unknown`
@app.route("/", methods=["POST"])
@parser.use_args({"foo": fields.Int(), "bar": fields.Int()}, location="json")
def post(args):
    return f"foo x bar = {args['foo'] * args['bar']}"

You can also define a default at parser instantiation, which will take precedence over these defaults, as in

from marshmallow import INCLUDE

parser = Parser(unknown=INCLUDE)

# because `unknown` is set on the parser, `DEFAULT_UNKNOWN_BY_LOCATION` has
# effect and `INCLUDE` will always be used
@app.route("/", methods=["POST"])
@parser.use_args({"foo": fields.Int(), "bar": fields.Int()}, location="json")
def post(args):
    unexpected_args = [k for k in args.keys() if k not in ("foo", "bar")]
    return f"foo x bar = {args['foo'] * args['bar']}; unexpected args={unexpected_args}"

Using Schema-Specfied `unknown`#

If you wish to use the value of unknown specified by a schema, simply pass unknown=None. This will disable webargs’ automatic passing of values for unknown. For example,

from flask import Flask
from marshmallow import Schema, fields, EXCLUDE, missing
from webargs.flaskparser import use_args


class RectangleSchema(Schema):
    length = fields.Float()
    width = fields.Float()

    class Meta:
        unknown = EXCLUDE


app = Flask(__name__)

# because unknown=None was passed, no value is passed during schema loading
# as a result, the schema's behavior (EXCLUDE) is used
@app.route("/", methods=["POST"])
@use_args(RectangleSchema(), location="json", unknown=None)
def get(args):
    return f"area = {args['length'] * args['width']}"

You can also set unknown=None when instantiating a parser to make this behavior the default for a parser.

When to avoid `use_kwargs`#

Any Schema passed to use_kwargs MUST deserialize to a dictionary of data. If your schema has a post_load method that returns a non-dictionary, you should use use_args instead.

from marshmallow import Schema, fields, post_load
from webargs.flaskparser import use_args


class Rectangle:
    def __init__(self, length, width):
        self.length = length
        self.width = width


class RectangleSchema(Schema):
    length = fields.Float()
    width = fields.Float()

    @post_load
    def make_object(self, data, **kwargs):
        return Rectangle(**data)


@use_args(RectangleSchema)
def post(rect: Rectangle):
    return f"Area: {rect.length * rect.width}"

Packages such as marshmallow-sqlalchemy and marshmallow-dataclass generate schemas that deserialize to non-dictionary objects. Therefore, use_args should be used with those schemas.

Schema Factories#

If you need to parametrize a schema based on a given request, you can use a “Schema factory”: a callable that receives the current request and returns a marshmallow.Schema instance.

Consider the following use cases:

Filtering via a query parameter by passing only to the Schema.
Handle partial updates for PATCH requests using marshmallow’s partial loading API.

from flask import Flask
from marshmallow import Schema, fields
from webargs.flaskparser import use_args

app = Flask(__name__)


class UserSchema(Schema):
    id = fields.Int(dump_only=True)
    username = fields.Str(required=True)
    password = fields.Str(load_only=True)
    first_name = fields.Str(load_default="")
    last_name = fields.Str(load_default="")
    date_registered = fields.DateTime(dump_only=True)


def make_user_schema(request):
    # Filter based on 'fields' query parameter
    fields = request.args.get("fields", None)
    only = fields.split(",") if fields else None
    # Respect partial updates for PATCH requests
    partial = request.method == "PATCH"
    # Add current request to the schema's context
    return UserSchema(only=only, partial=partial, context={"request": request})


# Pass the factory to .parse, .use_args, or .use_kwargs
@app.route("/profile/", methods=["GET", "POST", "PATCH"])
@use_args(make_user_schema)
def profile_view(args):
    username = args.get("username")
    # ...

Reducing Boilerplate#

We can reduce boilerplate and improve [re]usability with a simple helper function:

from webargs.flaskparser import use_args


def use_args_with(schema_cls, schema_kwargs=None, **kwargs):
    schema_kwargs = schema_kwargs or {}

    def factory(request):
        # Filter based on 'fields' query parameter
        only = request.args.get("fields", None)
        # Respect partial updates for PATCH requests
        partial = request.method == "PATCH"
        return schema_cls(
            only=only, partial=partial, context={"request": request}, **schema_kwargs
        )

    return use_args(factory, **kwargs)

Now we can attach input schemas to our view functions like so:

@use_args_with(UserSchema)
def profile_view(args):
    # ...
    get_profile(**args)

Custom Fields#

See the “Custom Fields” section of the marshmallow docs for a detailed guide on defining custom fields which you can pass to webargs parsers: https://marshmallow.readthedocs.io/en/latest/custom_fields.html.

Using `Method` and `Function` Fields with webargs#

Using the Method and Function fields requires that you pass the deserialize parameter.

@use_args({"cube": fields.Function(deserialize=lambda x: int(x) ** 3)})
def math_view(args):
    cube = args["cube"]
    # ...

Custom Parsers#

To add your own parser, extend Parser and implement the load_* method(s) you need to override. For example, here is a custom Flask parser that handles nested query string arguments.

import re

from webargs.flaskparser import FlaskParser


class NestedQueryFlaskParser(FlaskParser):
    """Parses nested query args

    This parser handles nested query args. It expects nested levels
    delimited by a period and then deserializes the query args into a
    nested dict.

    For example, the URL query params `?name.first=John&name.last=Boone`
    will yield the following dict:

        {
            'name': {
                'first': 'John',
                'last': 'Boone',
            }
        }
    """

    def load_querystring(self, req, schema):
        return _structure_dict(req.args)


def _structure_dict(dict_):
    def structure_dict_pair(r, key, value):
        m = re.match(r"(\w+)\.(.*)", key)
        if m:
            if r.get(m.group(1)) is None:
                r[m.group(1)] = {}
            structure_dict_pair(r[m.group(1)], m.group(2), value)
        else:
            r[key] = value

    r = {}
    for k, v in dict_.items():
        structure_dict_pair(r, k, v)
    return r

Parser pre_load#

Similar to @pre_load decorated hooks on marshmallow Schemas, Parser classes define a method, pre_load which can be overridden to provide per-parser transformations of data. The only way to make use of pre_load is to subclass a Parser and provide an implementation.

pre_load is given the data fetched from a location, the schema which will be used, the request object, and the location name which was requested. For example, to define a FlaskParser which strips whitespace from form and query data, one could write the following:

from webargs.flaskparser import FlaskParser
import typing


def _strip_whitespace(value):
    if isinstance(value, str):
        value = value.strip()
    elif isinstance(value, typing.Mapping):
        return {k: _strip_whitespace(value[k]) for k in value}
    elif isinstance(value, (list, tuple)):
        return type(value)(map(_strip_whitespace, value))
    return value


class WhitspaceStrippingFlaskParser(FlaskParser):
    def pre_load(self, location_data, *, schema, req, location):
        if location in ("query", "form"):
            return _strip_whitespace(location_data)
        return location_data

Note that Parser.pre_load is run after location loading but before Schema.load is called. It can therefore be called on multiple types of mapping objects, including MultiDictProxy, depending on what the location loader returns.

Returning HTTP 400 Responses#

If you’d prefer validation errors to return status code 400 instead of 422, you can override DEFAULT_VALIDATION_STATUS on a Parser.

Subclass the parser for your framework to do so. For example, using Falcon:

from webargs.falconparser import FalconParser


class Parser(FalconParser):
    DEFAULT_VALIDATION_STATUS = 400


parser = Parser()
use_args = parser.use_args
use_kwargs = parser.use_kwargs

Bulk-type Arguments#

In order to parse a JSON array of objects, pass many=True to your input Schema .

For example, you might implement JSON PATCH according to RFC 6902 like so:

from webargs import fields
from webargs.flaskparser import use_args
from marshmallow import Schema, validate


class PatchSchema(Schema):
    op = fields.Str(
        required=True,
        validate=validate.OneOf(["add", "remove", "replace", "move", "copy"]),
    )
    path = fields.Str(required=True)
    value = fields.Str(required=True)


@app.route("/profile/", methods=["patch"])
@use_args(PatchSchema(many=True))
def patch_blog(args):
    """Implements JSON Patch for the user profile

    Example JSON body:

    [
        {"op": "replace", "path": "/email", "value": "mynewemail@test.org"}
    ]
    """
    # ...

Multi-Field Detection#

If a List field is used to parse data from a location like query parameters – where one or multiple values can be passed for a single parameter name – then webargs will automatically treat that field as a list and parse multiple values if present.

To implement this behavior, webargs will examine schemas for marshmallow.fields.List fields. List fields get unpacked to list values when data is loaded, and other fields do not. This also applies to fields which inherit from List.

Note

In webargs v8, Tuple will be treated this way as well, in addition to List.

What if you have a list which should be treated as a “multi-field” but which does not inherit from List? webargs offers two solutions. You can add the custom attribute is_multiple=True to your field or you can add your class to your parser’s list of KNOWN_MULTI_FIELDS.

First, let’s define a “multiplexing field” which takes a string or list of strings to serve as an example:

# a custom field class which can accept values like List(String()) or String()
class CustomMultiplexingField(fields.String):
    def _deserialize(self, value, attr, data, **kwargs):
        if isinstance(value, str):
            return super()._deserialize(value, attr, data, **kwargs)
        return [
            self._deserialize(v, attr, data, **kwargs)
            for v in value
            if isinstance(v, str)
        ]

    def _serialize(self, value, attr, **kwargs):
        if isinstance(value, str):
            return super()._serialize(value, attr, **kwargs)
        return [self._serialize(v, attr, **kwargs) for v in value if isinstance(v, str)]

If you control the definition of CustomMultiplexingField, you can just add is_multiple=True to it:

# option 1: define the field with is_multiple = True
from webargs.flaskparser import parser


class CustomMultiplexingField(fields.Field):
    is_multiple = True  # <----- this marks this as a multi-field

    ...  # as above

If you don’t control the definition of CustomMultiplexingField, for example because it comes from a library, you can add it to the list of known multifields:

# option 2: add the field to the parer's list of multi-fields
class MyParser(FlaskParser):
    KNOWN_MULTI_FIELDS = list(FlaskParser.KNOWN_MULTI_FIELDS) + [
        CustomMultiplexingField
    ]


parser = MyParser()

In either case, the end result is that you can use the multifield and it will be detected as a list when unpacking query string data:

# gracefully handles
#   ...?foo=a
#   ...?foo=a&foo=b
# and treats them as ["a"] and ["a", "b"] respectively
@parser.use_args({"foo": CustomMultiplexingField()}, location="query")
def show_foos(foo):
    ...

Mixing Locations#

Arguments for different locations can be specified by passing location to each use_args call:

# "json" is the default, used explicitly below
@app.route("/stacked", methods=["POST"])
@use_args({"page": fields.Int(), "q": fields.Str()}, location="query")
@use_args({"name": fields.Str()}, location="json")
def viewfunc(query_parsed, json_parsed):
    page = query_parsed["page"]
    name = json_parsed["name"]
    # ...

To reduce boilerplate, you could create shortcuts, like so:

import functools

query = functools.partial(use_args, location="query")
body = functools.partial(use_args, location="json")


@query({"page": fields.Int(), "q": fields.Int()})
@body({"name": fields.Str()})
def viewfunc(query_parsed, json_parsed):
    page = query_parsed["page"]
    name = json_parsed["name"]
    # ...

Argument Passing and `arg_name`#

Note

This section describes behaviors which are planned to change in webargs version 9. In version 8, behavior will be as follows. In version 9, USE_ARGS_POSITIONAL will be removed and will always be False.

By default, webargs provides two ways of passing arguments via decorators, Parser.use_args, and Parser.use_kwargs. use_args passes parsed arguments as positionals, and use_kwargs expands dict-like parsed arguments into keyword arguments.

For use_args, the result is that sometimes it is non-obvious which order arguments will be passed in. Consider the following nearly identical example snippets:

# correct ordering, top-to-bottom
@use_args({"foo": fields.Int(), "bar": fields.Str()}, location="query")
@use_args({"baz": fields.Str()}, location="json")
def viewfunc(query_args, json_args):
    ...


# incorrect ordering, bottom-to-top
@use_args({"foo": fields.Int(), "bar": fields.Str()}, location="query")
@use_args({"baz": fields.Str()}, location="json")
def viewfunc(json_args, query_args):
    ...

To resolve this ambiguity, webargs version 9 will pass arguments from use_args as keyword arguments. You can opt-in to this behavior today by setting USE_ARGS_POSITIONAL = False on a parser class. This will cause webargs to pass arguments named {location}_args for each location used. For example,

from webargs.flaskparser import FlaskParser
from flask import Flask


class KeywordOnlyParser(FlaskParser):
    USE_ARGS_POSITIONAL = False


app = Flask(__name__)
parser = KeywordOnlyParser()


@app.route("/")
@parser.use_args({"foo": fields.Int(), "bar": fields.Str()}, location="query")
@parser.use_args({"baz": fields.Str()}, location="json")
def myview(*, query_args, json_args):
    ...

You can also customize the names of passed arguments using the arg_name parameter:

@app.route("/")
@parser.use_args(
    {"foo": fields.Int(), "bar": fields.Str()}, location="query", arg_name="query"
)
@parser.use_args({"baz": fields.Str()}, location="json", arg_name="payload")
def myview(*, query, payload):
    ...

Note that arg_name is available even on parsers where USE_ARGS_POSITIONAL is not set.

Next Steps#

See the Framework Support page for framework-specific guides.
For example applications, check out the examples directory.

Advanced Usage#

Custom Location Handlers#

“meta” Locations#

marshmallow Integration#

Setting unknown#

Default unknown#

Using Schema-Specfied unknown#

When to avoid use_kwargs#

Schema Factories#

Reducing Boilerplate#

Custom Fields#

Using Method and Function Fields with webargs#

Custom Parsers#

Parser pre_load#

Returning HTTP 400 Responses#

Bulk-type Arguments#

Multi-Field Detection#

Mixing Locations#

Argument Passing and arg_name#

Next Steps#

Setting `unknown`#

Default `unknown`#

Using Schema-Specfied `unknown`#

When to avoid `use_kwargs`#

Using `Method` and `Function` Fields with webargs#

Argument Passing and `arg_name`#