Use flit_core instead of setuptools as build backend.

Fix parsing of multipart bodies. #2734

Adjust index of last newline in data start. #2761

Parsing ints from header values strips spacing first. #2734

Fix empty file streaming when testing. #2740

Clearer error message when URL rule does not start with slash. #2750

Accept q value can be a float without a decimal part. #2751

FileStorage.content_length does not fail if the form data did not provide a value. #2726

Python 3.12 compatibility. #2704

Fix handling of invalid base64 values in Authorization.from_header . #2717

The debugger escapes the exception message in the page title. #2719

When binding routing.Map , a long IDNA server_name with a port does not fail encoding. #2700

iri_to_uri shows a deprecation warning instead of an error when passing bytes. #2708

When parsing numbers in HTTP request headers such as Content-Length , only ASCII digits are accepted rather than any format that Python’s int and float accept. #2716

Authorization.from_header and WWWAuthenticate.from_header detects tokens that end with base64 padding ( = ). #2685

Remove usage of warnings.catch_warnings . #2690

Remove max_form_parts restriction from standard form data parsing and only use if for multipart content. #2694

Response will avoid converting the Location header in some cases to preserve invalid URL schemes like itms-services . #2691

Fix parsing of large multipart bodies. Remove invalid leading newline, and restore parsing speed. #2658 , #2675

The cookie Path attribute is set to / by default again, to prevent clients from falling back to RFC 6265’s default-path behavior. #2672 , #2679

Parse the cookie Expires attribute correctly in the test client. #2669

max_content_length can only be enforced on streaming requests if the server sets wsgi.input_terminated . #2668

Percent-encode plus (+) when building URLs and in test requests. #2657

Cookie values don’t quote characters defined in RFC 6265. #2659

Include pyi files for datastructures type annotations. #2660

Authorization and WWWAuthenticate objects can be compared for equality. #2665

Drop support for Python 3.7. #2648

Remove previously deprecated code. #2592

Passing bytes where strings are expected is deprecated, as well as the charset and errors parameters in many places. Anywhere that was annotated, documented, or tested to accept bytes shows a warning. Removing this artifact of the transition from Python 2 to 3 removes a significant amount of overhead in instance checks and encoding cycles. In general, always work with UTF-8, the modern HTML, URL, and HTTP standards all strongly recommend this. #2602

Deprecate the werkzeug.urls module, except for the uri_to_iri and iri_to_uri functions. Use the urllib.parse library instead. #2600

Update which characters are considered safe when using percent encoding in URLs, based on the WhatWG URL Standard. #2601

Update which characters are considered safe when using percent encoding for Unicode filenames in downloads. #2598

Deprecate the safe_conversion parameter of iri_to_uri . The Location header is converted to IRI using the same process as everywhere else. #2609

Deprecate werkzeug.wsgi.make_line_iter and make_chunk_iter . #2613

Use modern packaging metadata with pyproject.toml instead of setup.cfg . #2574

Request.get_json() will raise a 415 Unsupported Media Type error if the Content-Type header is not application/json , instead of a generic 400. #2550

A URL converter’s part_isolating defaults to False if its regex contains a / . #2582

A custom converter’s regex can have capturing groups without breaking the router. #2596

The reloader can pick up arguments to python like -X dev , and does not require heuristics to determine how to reload the command. Only available on Python >= 3.10. #2589

The Watchdog reloader ignores file opened events. Bump the minimum version of Watchdog to 2.3.0. #2603

When using a Unix socket for the development server, the path can start with a dot. #2595

Increase default work factor for PBKDF2 to 600,000 iterations. #2611

parse_options_header is 2-3 times faster. It conforms to RFC 9110 , some invalid parts that were previously accepted are now ignored. #1628

The is_filename parameter to unquote_header_value is deprecated. #2614

Deprecate the extra_chars parameter and passing bytes to quote_header_value , the allow_token parameter to dump_header , and the cls parameter and passing bytes to parse_dict_header . #2618

Improve parse_accept_header implementation. Parse according to RFC 9110 . Discard items with invalid q values. #1623

quote_header_value quotes the empty string. #2618

dump_options_header skips None values rather than using a bare key. #2618

dump_header and dump_options_header will not quote a value if the key ends with an asterisk * .

parse_dict_header will decode values with charsets. #2618

Refactor the Authorization and WWWAuthenticate header data structures. #1769 , #2619

Both classes have type , parameters , and token attributes. The token attribute supports auth schemes that use a single opaque token rather than key=value parameters, such as Bearer .

Neither class is a dict anymore, although they still implement getting, setting, and deleting auth[key] and auth.key syntax, as well as auth.get(key) and key in auth .

Both classes have a from_header class method. parse_authorization_header and parse_www_authenticate_header are deprecated.

The methods WWWAuthenticate.set_basic and set_digest are deprecated. Instead, an instance should be created and assigned to response.www_authenticate .

A list of instances can be assigned to response.www_authenticate to set multiple header values. However, accessing the property only returns the first instance.

Refactor parse_cookie and dump_cookie . #2637

parse_cookie is up to 40% faster, dump_cookie is up to 60% faster.

Passing bytes to parse_cookie and dump_cookie is deprecated. The dump_cookie charset parameter is deprecated.

dump_cookie allows domain values that do not include a dot . , and strips off a leading dot.

dump_cookie does not set path="/" unnecessarily by default.

Refactor the test client cookie implementation. #1060 , #1680

The cookie_jar attribute is deprecated. http.cookiejar is no longer used for storage.

Domain and path matching is used when sending cookies in requests. The domain and path parameters default to localhost and / .

Added a get_cookie method to inspect cookies.

Cookies have decoded_key and decoded_value attributes to match what the app sees rather than the encoded values a client would see.

The first positional server_name parameter to set_cookie and delete_cookie is deprecated. Use the domain parameter instead.

Other parameters to delete_cookie besides domain , path , and value are deprecated.

If request.max_content_length is set, it is checked immediately when accessing the stream, and while reading from the stream in general, rather than only during form parsing. #1513

The development server, which must not be used in production, will exhaust the request stream up to 10GB or 1000 reads. This allows clients to see a 413 error if max_content_length is exceeded, instead of a “connection reset” failure. #2620

The development server discards header keys that contain underscores _ , as they are ambiguous with dashes - in WSGI. #2622

secure_filename looks for more Windows reserved file names. #2623

Update type annotation for best_match to make default parameter clearer. #2625

Multipart parser handles empty fields correctly. #2632

The Map charset parameter and Request.url_charset property are deprecated. Percent encoding in URLs must always represent UTF-8 bytes. Invalid bytes are left percent encoded rather than replaced. #2602

The Request.charset , Request.encoding_errors , Response.charset , and Client.charset attributes are deprecated. Request and response data must always use UTF-8. #2602

Header values that have charset information only allow ASCII, UTF-8, and ISO-8859-1. #2614 , #2641

Update type annotation for ProfilerMiddleware stream parameter. #2642

Use postponed evaluation of annotations. #2644

The development server escapes ASCII control characters in decoded URLs before logging the request to the terminal. #2652

The FormDataParser parse_functions attribute and get_parse_func method, and the invalid application/x-url-encoded content type, are deprecated. #2653

generate_password_hash supports scrypt. Plain hash methods are deprecated, only scrypt and pbkdf2 are supported. #2654

Ensure that URL rules using path converters will redirect with strict slashes when the trailing slash is missing. #2533

Type signature for get_json specifies that return type is not optional when silent=False . #2508

parse_content_range_header returns None for a value like bytes */-1 where the length is invalid, instead of raising an AssertionError . #2531

Address remaining ResourceWarning related to the socket used by run_simple . Remove prepare_socket , which now happens when creating the server. #2421

Update pre-existing headers for multipart/form-data requests with the test client. #2549

Fix handling of header extended parameters such that they are no longer quoted. #2529

LimitedStream.read works correctly when wrapping a stream that may not return the requested size in one read call. #2558

A cookie header that starts with = is treated as an empty key and discarded, rather than stripping the leading == .

Specify a maximum number of multipart parts, default 1000, after which a RequestEntityTooLarge exception is raised on parsing. This mitigates a DoS attack where a larger number of form/file parts would result in disproportionate resource use.

Fix router to restore the 2.1 strict_slashes == False behaviour whereby leaf-requests match branch rules and vice versa. #2489

Fix router to identify invalid rules rather than hang parsing them, and to correctly parse / within converter arguments. #2489

Update subpackage imports in werkzeug.routing to use the import as syntax for explicitly re-exporting public attributes. #2493

Parsing of some invalid header characters is more robust. #2494

When starting the development server, a warning not to use it in a production deployment is always shown. #2480

LocalProxy.__wrapped__ is always set to the wrapped object when the proxy is unbound, fixing an issue in doctest that would cause it to fail. #2485

Address one ResourceWarning related to the socket used by run_simple . #2421

Fix router so that /path/ will match a rule /path if strict slashes mode is disabled for the rule. #2467

Fix router so that partial part matches are not allowed i.e. /2df does not match /<int> . #2470

Fix router static part weighting, so that simpler routes are matched before more complex ones. #2471

Restore ValidationError to be importable from werkzeug.routing . #2465

Deprecated get_script_name , get_query_string , peek_path_info , pop_path_info , and extract_path_info . #2461

Remove previously deprecated code. #2461

Add MarkupSafe as a dependency and use it to escape values when rendering HTML. #2419

Added the werkzeug.debug.preserve_context mechanism for restoring context-local data for a request when running code in the debug console. #2439

Fix compatibility with Python 3.11 by ensuring that end_lineno and end_col_offset are present on AST nodes. #2425

Add a new faster URL matching router based on a state machine. If a custom converter needs to match a / it must set the class variable part_isolating = False . #2433

Fix branch leaf path masking branch paths when strict-slashes is disabled. #1074

Names within options headers are always converted to lowercase. This matches RFC 6266 that the case is not relevant. #2442

AnyConverter validates the value passed for it when building URLs. #2388

The debugger shows enhanced error locations in tracebacks in Python 3.11. #2407

Added Sans-IO is_resource_modified and parse_cookie functions based on WSGI versions. #2408

Added Sans-IO get_content_length function. #2415

Don’t assume a mimetype for test responses. #2450

Type checking FileStorage accepts os.PathLike . #2418

The development server does not set Transfer-Encoding: chunked for 1xx, 204, 304, and HEAD responses. #2375

Response HTML for exceptions and redirects starts with <!doctype html> and <html lang=en> . #2390

Fix ability to set some cache_control attributes to False . #2379

Disable keep-alive connections in the development server, which are not supported sufficiently by Python’s http.server . #2397

ResponseCacheControl.s_maxage converts its value to an int, like max_age . #2364

Drop support for Python 3.6. #2277

Using gevent or eventlet requires greenlet>=1.0 or PyPy>=7.3.7. werkzeug.locals and contextvars will not work correctly with older versions. #2278

Remove previously deprecated code. #2276

Remove the non-standard shutdown function from the WSGI environ when running the development server. See the docs for alternatives.

Request and response mixins have all been merged into the Request and Response classes.

The user agent parser and the useragents module is removed. The user_agent module provides an interface that can be subclassed to add a parser, such as ua-parser. By default it only stores the whole string.

The test client returns TestResponse instances and can no longer be treated as a tuple. All data is available as properties on the response.

Remove locals.get_ident and related thread-local code from locals , it no longer makes sense when moving to a contextvars-based implementation.

Remove the python -m werkzeug.serving CLI.

The has_key method on some mapping datastructures; use key in data instead.

Request.disable_data_descriptor is removed, pass shallow=True instead.

Remove the no_etag parameter from Response.freeze() .

Remove the HTTPException.wrap class method.

Remove the cookie_date function. Use http_date instead.

Remove the pbkdf2_hex , pbkdf2_bin , and safe_str_cmp functions. Use equivalents in hashlib and hmac modules instead.

Remove the Href class.

Remove the HTMLBuilder class.

Remove the invalidate_cached_property function. Use del obj.attr instead.

Remove bind_arguments and validate_arguments . Use Signature.bind() and inspect.signature() instead.

Remove detect_utf_encoding , it’s built-in to json.loads .

Remove format_string , use string.Template instead.

Remove escape and unescape . Use MarkupSafe instead.

The multiple parameter of parse_options_header is deprecated. #2357

Rely on PEP 538 and PEP 540 to handle decoding file names with the correct filesystem encoding. The filesystem module is removed. #1760

Default values passed to Headers are validated the same way values added later are. #1608

Setting CacheControl int properties, such as max_age , will convert the value to an int. #2230

Always use socket.fromfd when restarting the dev server. #2287

When passing a dict of URL values to Map.build , list values do not filter out None or collapse to a single value. Passing a MultiDict does collapse single items. This undoes a previous change that made it difficult to pass a list, or None values in a list, to custom URL converters. #2249

run_simple shows instructions for dealing with “address already in use” errors, including extra instructions for macOS. #2321

Extend list of characters considered always safe in URLs based on RFC 3986 . #2319

Optimize the stat reloader to avoid watching unnecessary files in more cases. The watchdog reloader is still recommended for performance and accuracy. #2141

The development server uses Transfer-Encoding: chunked for streaming responses when it is configured for HTTP/1.1. #2090 , #1327 , #2091

The development server uses HTTP/1.1, which enables keep-alive connections and chunked streaming responses, when threaded or processes is enabled. #2323

cached_property works for classes with __slots__ if a corresponding _cache_{name} slot is added. #2332

Refactor the debugger traceback formatter to use Python’s built-in traceback module as much as possible. #1753

The TestResponse.text property is a shortcut for r.get_data(as_text=True) , for convenient testing against text instead of bytes. #2337

safe_join ensures that the path remains relative if the trusted directory is the empty string. #2349

Percent-encoded newlines ( %0a ), which are decoded by WSGI servers, are considered when routing instead of terminating the match early. #2350

The test client doesn’t set duplicate headers for CONTENT_LENGTH and CONTENT_TYPE . #2348

append_slash_redirect handles PATH_INFO with internal slashes. #1972 , #2338

The default status code for append_slash_redirect is 308 instead of 301. This preserves the request body, and matches a previous change to strict_slashes in routing. #2351

Fix ValueError: I/O operation on closed file. with the test client when following more than one redirect. #2353

Response.autocorrect_location_header is disabled by default. The Location header URL will remain relative, and exclude the scheme and domain, by default. #2352

Request.get_json() will raise a 400 BadRequest error if the Content-Type header is not application/json . This makes a very common source of confusion more visible. #2339

ProxyFix supports IPv6 addresses. #2262

Type annotation for Response.make_conditional , HTTPException.get_response , and Map.bind_to_environ accepts Request in addition to WSGIEnvironment for the first parameter. #2290

Fix type annotation for Request.user_agent_class . #2273

Accessing LocalProxy.__class__ and __doc__ on an unbound proxy returns the fallback value instead of a method object. #2188

Redirects with the test client set RAW_URI and REQUEST_URI correctly. #2151

Handle multiple tokens in Connection header when routing WebSocket requests. #2131

Set the debugger pin cookie secure flag when on https. #2150

Fix type annotation for MultiDict.update to accept iterable values #2142

Prevent double encoding of redirect URL when merge_slash=True for Rule.match . #2157

CombinedMultiDict.to_dict with flat=False considers all component dicts when building value lists. #2189

send_file only sets a detected Content-Encoding if as_attachment is disabled to avoid browsers saving decompressed .tar.gz files. #2149

Fix type annotations for TypeConversionDict.get to not return an Optional value if both default and type are not None . #2169

Fix type annotation for routing rule factories to accept Iterable[RuleFactory] instead of Iterable[Rule] for the rules parameter. #2183

Add missing type annotation for FileStorage.__getattr__ #2155

The debugger pin cookie is set with SameSite set to Strict instead of None to be compatible with modern browser security. #2156

Type annotations use IO[bytes] and IO[str] instead of BinaryIO and TextIO for wider type compatibility. #2130

Ad-hoc TLS certs are generated with SAN matching CN. #2158

Fix memory usage for locals when using Python 3.6 or pre 0.4.17 greenlet versions. #2212

Fix type annotation in CallbackDict , because it is not utilizing a bound TypeVar. #2235

Fix setting CSP header options on the response. #2237

Fix an issue with with the interactive debugger where lines would not expand on click for very long tracebacks. #2239

The interactive debugger handles displaying an exception that does not have a traceback, such as from ProcessPoolExecutor . #2217

Fix type annotation for send_file max_age callable. Don’t pass pathlib.Path to max_age . #2119

Mark top-level names as exported so type checking understands imports in user projects. #2122

Fix some types that weren’t available in Python 3.6.0. #2123

cached_property is generic over its return type, properties decorated with it report the correct type. #2113

Fix multipart parsing bug when boundary contains special regex characters. #2125

Type checking understands that calling headers.get with a string default will always return a string. #2128

If HTTPException.description is not a string, get_description will convert it to a string. #2115

Drop support for Python 2 and 3.5. #1693

Deprecate utils.format_string() , use string.Template instead. #1756

Deprecate utils.bind_arguments() and utils.validate_arguments() , use Signature.bind() and inspect.signature() instead. #1757

Deprecate utils.HTMLBuilder . #1761

Deprecate utils.escape() and utils.unescape() , use MarkupSafe instead. #1758

Deprecate the undocumented python -m werkzeug.serving CLI. #1834

Deprecate the environ["werkzeug.server.shutdown"] function that is available when running the development server. #1752

Deprecate the useragents module and the built-in user agent parser. Use a dedicated parser library instead by subclassing user_agent.UserAgent and setting Request.user_agent_class . #2078

Remove the unused, internal posixemulation module. #1759

All datetime values are timezone-aware with tzinfo=timezone.utc . This applies to anything using http.parse_date : Request.date , .if_modified_since , .if_unmodified_since ; Response.date , .expires , .last_modified , .retry_after ; parse_if_range_header , and IfRange.date . When comparing values, the other values must also be aware, or these values must be made naive. When passing parameters or setting attributes, naive values are still assumed to be in UTC. #2040

Merge all request and response wrapper mixin code into single Request and Response classes. Using the mixin classes is no longer necessary and will show a deprecation warning. Checking isinstance or issubclass against BaseRequest and BaseResponse will show a deprecation warning and check against Request or Response instead. #1963

JSON support no longer uses simplejson if it’s installed. To use another JSON module, override Request.json_module and Response.json_module . #1766

Response.get_json() no longer caches the result, and the cache parameter is removed. #1698

Response.freeze() generates an ETag header if one is not set. The no_etag parameter (which usually wasn’t visible anyway) is no longer used. #1963

Add a url_scheme argument to build() to override the bound scheme. #1721

Passing an empty list as a query string parameter to build() won’t append an unnecessary ? . Also drop any number of None items in a list. #1992

When passing a Headers object to a test client method or EnvironBuilder , multiple values for a key are joined into one comma separated value. This matches the HTTP spec on multi-value headers. #1655

Setting Response.status and status_code uses identical parsing and error checking. #1658 , #1728

MethodNotAllowed and RequestedRangeNotSatisfiable take a response kwarg, consistent with other HTTP errors. #1748

The response generated by Unauthorized produces one WWW-Authenticate header per value in www_authenticate , rather than joining them into a single value, to improve interoperability with browsers and other clients. #1755

If parse_authorization_header can’t decode the header value, it returns None instead of raising a UnicodeDecodeError . #1816

The debugger no longer uses jQuery. #1807

The test client includes the query string in REQUEST_URI and RAW_URI . #1781

Switch the parameter order of default_stream_factory to match the order used when calling it. #1085

Add send_file function to generate a response that serves a file. Adapted from Flask’s implementation. #265 , #1850

Add send_from_directory function to safely serve an untrusted path within a trusted directory. Adapted from Flask’s implementation. #1880

send_file takes download_name , which is passed even if as_attachment=False by using Content-Disposition: inline . download_name replaces Flask’s attachment_filename . #1869

send_file sets conditional=True and max_age=None by default. Cache-Control is set to no-cache if max_age is not set, otherwise public . This tells browsers to validate conditional requests instead of using a timed cache. max_age=None replaces Flask’s cache_timeout=43200 . #1882

send_file can be called with etag="string" to set a custom ETag instead of generating one. etag replaces Flask’s add_etags . #1868

send_file sets the Content-Encoding header if an encoding is returned when guessing mimetype from download_name . #3896

Update the defaults used by generate_password_hash . Increase PBKDF2 iterations to 260000 from 150000. Increase salt length to 16 from 8. Use secrets module to generate salt. #1935

The reloader doesn’t crash if sys.stdin is somehow None . #1915

Add arguments to delete_cookie to match set_cookie and the attributes modern browsers expect. #1889

utils.cookie_date is deprecated, use utils.http_date instead. The value for Set-Cookie expires is no longer “-” delimited. #2040

Use request.headers instead of request.environ to look up header attributes. #1808

The test Client request methods ( client.get , etc.) always return an instance of TestResponse . In addition to the normal behavior of Response , this class provides request with the request that produced the response, and history to track intermediate responses when follow_redirects is used. #763 , #1894

The test Client request methods takes an auth parameter to add an Authorization header. It can be an Authorization object or a (username, password) tuple for Basic auth. #1809

Calling response.close() on a response from the test Client will close the request input stream. This matches file behavior and can prevent a ResourceWarning in some cases. #1785

EnvironBuilder.from_environ decodes values encoded for WSGI, to avoid double encoding the new values. #1959

The default stat reloader will watch Python files under non-system/virtualenv sys.path entries, which should contain most user code. It will also watch all Python files under directories given in extra_files . #1945

The reloader ignores __pycache__ directories again. #1945

run_simple takes exclude_patterns a list of fnmatch patterns that will not be scanned by the reloader. #1333

Cookie names are no longer unquoted. This was against RFC 6265 and potentially allowed setting __Secure prefixed cookies. #1965

Fix some word matches for user agent platform when the word can be a substring. #1923

The development server logs ignored SSL errors. #1967

Temporary files for form data are opened in rb+ instead of wb+ mode for better compatibility with some libraries. #1961

Use SHA-1 instead of MD5 for generating ETags and the debugger pin, and in some tests. MD5 is not available in some environments, such as FIPS 140. This may invalidate some caches since the ETag will be different. #1897

Add Cross-Origin-Opener-Policy and Cross-Origin-Embedder-Policy response header properties. #2008

run_simple tries to show a valid IP address when binding to all addresses, instead of 0.0.0.0 or :: . It also warns about not running the development server in production in this case. #1964

Colors in the development server log are displayed if Colorama is installed on Windows. For all platforms, style support no longer requires Click. #1832

A range request for an empty file (or other data with length 0) will return a 200 response with the empty file instead of a 416 error. #1937

New sans-IO base classes for Request and Response have been extracted to contain all the behavior that is not WSGI or IO dependent. These are not a public API, they are part of an ongoing refactor to let ASGI frameworks use Werkzeug. #2005

Parsing multipart/form-data has been refactored to use sans-io patterns. This should also make parsing forms with large binary file uploads significantly faster. #1788 , #875

LocalProxy matches the current Python data model special methods, including all r-ops, in-place ops, and async. __class__ is proxied, so the proxy will look like the object in more cases, including isinstance . Use issubclass(type(obj), LocalProxy) to check if an object is actually a proxy. #1754

Local uses ContextVar on Python 3.7+ instead of threading.local . #1778

request.values does not include form for GET requests (even though GET bodies are undefined). This prevents bad caching proxies from caching form data instead of query strings. #2037

The development server adds the underlying socket to environ as werkzeug.socket . This is non-standard and specific to the dev server, other servers may expose this under their own key. It is useful for handling a WebSocket upgrade request. #2052

URL matching assumes websocket=True mode for WebSocket upgrade requests. #2052

Updated UserAgentParser to handle more cases. #1971

werzeug.DechunkedInput.readinto will not read beyond the size of the buffer. #2021

Fix connection reset when exceeding max content size. #2051

pbkdf2_hex , pbkdf2_bin , and safe_str_cmp are deprecated. hashlib and hmac provide equivalents. #2083

invalidate_cached_property is deprecated. Use del obj.name instead. #2084

Href is deprecated. Use werkzeug.routing instead. #2085

Request.disable_data_descriptor is deprecated. Create the request with shallow=True instead. #2085

HTTPException.wrap is deprecated. Create a subclass manually instead. #2085

Make the argument to RequestRedirect.get_response optional. #1718

Only allow a single access control allow origin value. #1723

Fix crash when trying to parse a non-existent Content Security Policy header. #1731

http_date zero fills years < 1000 to always output four digits. #1739

Fix missing local variables in interactive debugger console. #1746

Fix passing file-like objects like io.BytesIO to FileStorage.save . #1733

Drop support for Python 3.4. ( #1478 )

Remove code that issued deprecation warnings in version 0.15. ( #1477 )

Remove most top-level attributes provided by the werkzeug module in favor of direct imports. For example, instead of import werkzeug; werkzeug.url_quote , do from werkzeug.urls import url_quote . Install version 0.16 first to see deprecation warnings while upgrading. #2 , #1640

Added utils.invalidate_cached_property() to invalidate cached properties. ( #1474 )

Directive keys for the Set-Cookie response header are not ignored when parsing the Cookie request header. This allows cookies with names such as “expires” and “version”. ( #1495 )

Request cookies are parsed into a MultiDict to capture all values for cookies with the same key. cookies[key] returns the first value rather than the last. Use cookies.getlist(key) to get all values. parse_cookie also defaults to a MultiDict . #1562 , #1458

Add charset=utf-8 to an HTTP exception response’s CONTENT_TYPE header. ( #1526 )

The interactive debugger handles outer variables in nested scopes such as lambdas and comprehensions. #913 , #1037 , #1532

The user agent for Opera 60 on Mac is correctly reported as “opera” instead of “chrome”. #1556

The platform for Crosswalk on Android is correctly reported as “android” instead of “chromeos”. ( #1572 )

Issue a warning when the current server name does not match the configured server name. #760

A configured server name with the default port for a scheme will match the current server name without the port if the current scheme matches. #1584

InternalServerError has a original_exception attribute that frameworks can use to track the original cause of the error. #1590

Headers are tested for equality independent of the header key case, such that X-Foo is the same as x-foo . #1605

http.dump_cookie() accepts 'None' as a value for samesite . #1549

set_cookie() accepts a samesite argument. #1705

Support the Content Security Policy header through the Response.content_security_policy data structure. #1617

LanguageAccept will fall back to matching “en” for “en-US” or “en-US” for “en” to better support clients or translations that only match at the primary language tag. #450 , #1507

MIMEAccept uses MIME parameters for specificity when matching. #458 , #1574

If the development server is started with an SSLContext configured to verify client certificates, the certificate in PEM format will be available as environ["SSL_CLIENT_CERT"] . #1469

is_resource_modified will run for methods other than GET and HEAD , rather than always returning False . #409

SharedDataMiddleware returns 404 rather than 500 when trying to access a directory instead of a file with the package loader. The dependency on setuptools and pkg_resources is removed. #1599

Add a response.cache_control.immutable flag. Keep in mind that browser support for this Cache-Control header option is still experimental and may not be implemented. #1185

Optional request log highlighting with the development server is handled by Click instead of termcolor. #1235

Optional ad-hoc TLS support for the development server is handled by cryptography instead of pyOpenSSL. #1555

FileStorage.save() supports pathlib and PEP 519 PathLike objects. #1653

The debugger security pin is unique in containers managed by Podman. #1661

Building a URL when host_matching is enabled takes into account the current host when there are duplicate endpoints with different hosts. #488

The 429 TooManyRequests and 503 ServiceUnavailable HTTP exceptions takes a retry_after parameter to set the Retry-After header. #1657

Map and Rule have a merge_slashes option to collapse multiple slashes into one, similar to how many HTTP servers behave. This is enabled by default. #1286 , #1694

Add HTTP 103, 208, 306, 425, 506, 508, and 511 to the list of status codes. #1678

Add update , setlist , and setlistdefault methods to the Headers data structure. extend method can take MultiDict and kwargs. #1687 , #1697

The development server accepts paths that start with two slashes, rather than stripping off the first path segment. #491

Add access control (Cross Origin Request Sharing, CORS) header properties to the Request and Response wrappers. #1699

Accept values are no longer ordered alphabetically for equal quality tags. Instead the initial order is preserved. #1686

Added Map.lock_class attribute for alternative implementations. #1702

Support matching and building WebSocket rules in the routing system, for use by async frameworks. #1709

Range requests that span an entire file respond with 206 instead of 200, to be more compliant with RFC 7233 . This may help serving media to older browsers. #410 , #1704

The SharedDataMiddleware default fallback_mimetype is application/octet-stream . If a filename looks like a text mimetype, the utf-8 charset is added to it. This matches the behavior of BaseResponse and Flask’s send_file() . #1689

Fix import location in deprecation messages for subpackages. #1663

Fix an SSL error on Python 3.5 when the dev server responds with no content. #1659

Deprecate most top-level attributes provided by the werkzeug module in favor of direct imports. The deprecated imports will be removed in version 1.0.

For example, instead of import werkzeug; werkzeug.url_quote , do from werkzeug.urls import url_quote . A deprecation warning will show the correct import to use. werkzeug.exceptions and werkzeug.routing should also be imported instead of accessed, but for technical reasons can’t show a warning.

#2 , #1640

Work around a bug in pip that caused the reloader to fail on Windows when the script was an entry point. This fixes the issue with Flask’s flask run command failing with “No module named Scriptsflask”. #1614

ProxyFix trusts the X-Forwarded-Proto header by default. #1630

The deprecated num_proxies argument to ProxyFix sets x_for , x_proto , and x_host to match 0.14 behavior. This is intended to make intermediate upgrades less disruptive, but the argument will still be removed in 1.0. #1630

Fix a TypeError due to changes to ast.Module in Python 3.8. #1551

Fix a C assertion failure in debug builds of some Python 2.7 releases. #1553

BadRequestKeyError adds the KeyError message to the description if e.show_exception is set to True . This is a more secure default than the original 0.15.0 behavior and makes it easier to control without losing information. #1592

Upgrade the debugger to jQuery 3.4.1. #1581

Work around an issue in some external debuggers that caused the reloader to fail. #1607

Work around an issue where the reloader couldn’t introspect a setuptools script installed as an egg. #1600

The reloader will use sys.executable even if the script is marked executable, reverting a behavior intended for NixOS introduced in 0.15. The reloader should no longer cause OSError: [Errno 8] Exec format error . #1482 , #1580

SharedDataMiddleware safely handles paths with Windows drive names. #1589

Fix a SyntaxError on Python 2.7.5. ( #1544 )

Properly handle multi-line header folding in development server in Python 2.7. ( #1080 )

Restore the response argument to Unauthorized . ( #1527 )

Unauthorized doesn’t add the WWW-Authenticate header if www_authenticate is not given. ( #1516 )

The default URL converter correctly encodes bytes to string rather than representing them with b'' . ( #1502 )

Fix the filename format string in ProfilerMiddleware to correctly handle float values. ( #1511 )

Update LintMiddleware to work on Python 3. ( #1510 )

The debugger detects cycles in chained exceptions and does not time out in that case. ( #1536 )

When running the development server in Docker, the debugger security pin is now unique per container.

Rule code generation uses a filename that coverage will ignore. The previous value, “generated”, was causing coverage to fail. ( #1487 )

The test client removes the cookie header if there are no persisted cookies. This fixes an issue introduced in 0.15.0 where the cookies from the original request were used for redirects, causing functions such as logout to fail. ( #1491 )

The test client copies the environ before passing it to the app, to prevent in-place modifications from affecting redirect requests. ( #1498 )

The "werkzeug" logger only adds a handler if there is no handler configured for its level in the logging chain. This avoids double logging if other code configures logging first. ( #1492 )

Unauthorized takes description as the first argument, restoring previous behavior. The new www_authenticate argument is listed second. ( #1483 )

Building URLs is ~7x faster. Each Rule compiles an optimized function for building itself. ( #1281 )

MapAdapter.build() can be passed a MultiDict to represent multiple values for a key. It already did this when passing a dict with a list value. ( #724 )

path_info defaults to '/' for Map.bind() . ( #740 , #768 , #1316 )

Change RequestRedirect code from 301 to 308, preserving the verb and request body (form data) during redirect. ( #1342 )

int and float converters in URL rules will handle negative values if passed the signed=True parameter. For example, /jump/<int(signed=True):count> . ( #1355 )

Location autocorrection in Response.get_wsgi_headers() is relative to the current path rather than the root path. ( #693 , #718 , #1315 )

412 responses once again include entity headers and an error message in the body. They were originally omitted when implementing If-Match ( #1233 ), but the spec doesn’t seem to disallow it. ( #1231 , #1255 )

The Content-Length header is removed for 1xx and 204 responses. This fixes a previous change where no body would be sent, but the header would still be present. The new behavior matches RFC 7230. ( #1294 )

Unauthorized takes a www_authenticate parameter to set the WWW-Authenticate header for the response, which is technically required for a valid 401 response. ( #772 , #795 )

Add support for status code 424 FailedDependency . ( #1358 )

http.parse_cookie() ignores empty segments rather than producing a cookie with no key or value. ( #1245 , #1301 )

parse_authorization_header() (and Authorization , authorization ) treats the authorization header as UTF-8. On Python 2, basic auth username and password are unicode . ( #1325 )

parse_options_header() understands RFC 2231 parameter continuations. ( #1417 )

uri_to_iri() does not unquote ASCII characters in the unreserved class, such as space, and leaves invalid bytes quoted when decoding. iri_to_uri() does not quote reserved characters. See RFC 3987 for these character classes. ( #1433 )

get_content_type appends a charset for any mimetype that ends with +xml , not just those that start with application/ . Known text types such as application/javascript are also given charsets. ( #1439 )

Clean up werkzeug.security module, remove outdated hashlib support. ( #1282 )

In generate_password_hash() , PBKDF2 uses 150000 iterations by default, increased from 50000. ( #1377 )

ClosingIterator calls close on the wrapped iterable , not the internal iterator. This doesn’t affect objects where __iter__ returned self . For other objects, the method was not called before. ( #1259 , #1260 )

Bytes may be used as keys in Headers , they will be decoded as Latin-1 like values are. ( #1346 )

Range validates that list of range tuples passed to it would produce a valid Range header. ( #1412 )

FileStorage looks up attributes on stream._file if they don’t exist on stream , working around an issue where tempfile.SpooledTemporaryFile() didn’t implement all of io.IOBase . See https://github.com/python/cpython/pull/3249 . ( #1409 )

CombinedMultiDict.copy() returns a shallow mutable copy as a MultiDict . The copy no longer reflects changes to the combined dicts, but is more generally useful. ( #1420 )

The version of jQuery used by the debugger is updated to 3.3.1. ( #1390 )

The debugger correctly renders long markupsafe.Markup instances. ( #1393 )

The debugger can serve resources when Werkzeug is installed as a zip file. DebuggedApplication.get_resource uses pkgutil.get_data . ( #1401 )

The debugger and server log support Python 3’s chained exceptions. ( #1396 )

The interactive debugger highlights frames that come from user code to make them easy to pick out in a long stack trace. Note that if an env was created with virtualenv instead of venv, the debugger may incorrectly classify some frames. ( #1421 )

Clicking the error message at the top of the interactive debugger will jump down to the bottom of the traceback. ( #1422 )

When generating a PIN, the debugger will ignore a KeyError raised when the current UID doesn’t have an associated username, which can happen in Docker. ( #1471 )

BadRequestKeyError adds the KeyError message to the description, making it clearer what caused the 400 error. Frameworks like Flask can omit this information in production by setting e.args = () . ( #1395 )

If a nested ImportError occurs from import_string() the traceback mentions the nested import. Removes an untested code path for handling “modules not yet set up by the parent.” ( #735 )

Triggering a reload while using a tool such as PDB no longer hides input. ( #1318 )

The reloader will not prepend the Python executable to the command line if the Python file is marked executable. This allows the reloader to work on NixOS. ( #1242 )

Fix an issue where sys.path would change between reloads when running with python -m app . The reloader can detect that a module was run with “-m” and reconstructs that instead of the file path in sys.argv when reloading. ( #1416 )

The dev server can bind to a Unix socket by passing a hostname like unix://app.socket . ( #209 , #1019 )

Server uses IPPROTO_TCP constant instead of SOL_TCP for Jython compatibility. ( #1375 )

When using an adhoc SSL cert with run_simple() , the cert is shown as self-signed rather than signed by an invalid authority. ( #1430 )

The development server logs the unquoted IRI rather than the raw request line, to make it easier to work with Unicode in request paths during development. ( #1115 )

The development server recognizes ConnectionError on Python 3 to silence client disconnects, and does not silence other OSErrors that may have been raised inside the application. ( #1418 )

The environ keys REQUEST_URI and RAW_URI contain the raw path before it was percent-decoded. This is non-standard, but many WSGI servers add them. Middleware could replace PATH_INFO with this to route based on the raw value. ( #1419 )

EnvironBuilder doesn’t set CONTENT_TYPE or CONTENT_LENGTH in the environ if they aren’t set. Previously these used default values if they weren’t set. Now it’s possible to distinguish between empty and unset values. ( #1308 )

The test client raises a ValueError if a query string argument would overwrite a query string in the path. ( #1338 )

test.EnvironBuilder and test.Client take a json argument instead of manually passing data and content_type . This is serialized using the test.EnvironBuilder.json_dumps() method. ( #1404 )

test.Client redirect handling is rewritten. ( #1402 )

The redirect environ is copied from the initial request environ.

Script root and path are correctly distinguished when redirecting to a path under the root.

The HEAD method is not changed to GET.

307 and 308 codes preserve the method and body. All others ignore the body and related headers.

Headers are passed to the new request for all codes, following what browsers do.

test.EnvironBuilder sets the content type and length headers in addition to the WSGI keys when detecting them from the data.

Intermediate response bodies are iterated over even when buffered=False to ensure iterator middleware can run cleanup code safely. Only the last response is not buffered. ( #988 )

EnvironBuilder , FileStorage , and wsgi.get_input_stream() no longer share a global _empty_stream instance. This improves test isolation by preventing cases where closing the stream in one request would affect other usages. ( #1340 )

The default SecureCookie.serialization_method will change from pickle to json in 1.0. To upgrade existing tokens, override unquote() to try pickle if json fails. ( #1413 )

CGIRootFix no longer modifies PATH_INFO for very old versions of Lighttpd. LighttpdCGIRootFix was renamed to CGIRootFix in 0.9. Both are deprecated and will be removed in version 1.0. ( #1141 )

werkzeug.wrappers.json.JSONMixin has been replaced with Flask’s implementation. Check the docs for the full API. ( #1445 )

The contrib modules are deprecated and will either be moved into werkzeug core or removed completely in version 1.0. Some modules that already issued deprecation warnings have been removed. Be sure to run or test your code with python -W default::DeprecationWarning to catch any deprecated code you’re using. ( #4 )

LintMiddleware has moved to werkzeug.middleware.lint .

ProfilerMiddleware has moved to werkzeug.middleware.profiler .

ProxyFix has moved to werkzeug.middleware.proxy_fix .

JSONRequestMixin has moved to werkzeug.wrappers.json .

cache has been extracted into a separate project, cachelib . The version in Werkzeug is deprecated.

securecookie and sessions have been extracted into a separate project, secure-cookie . The version in Werkzeug is deprecated.

Everything in fixers , except ProxyFix , is deprecated.

Everything in wrappers , except JSONMixin , is deprecated.

atom is deprecated. This did not fit in with the rest of Werkzeug, and is better served by a dedicated library in the community.

jsrouting is removed. Set URLs when rendering templates or JSON responses instead.

limiter is removed. Its specific use is handled by Werkzeug directly, but stream limiting is better handled by the WSGI server in general.

testtools is removed. It did not offer significant benefit over the default test client.

iterio is deprecated.

wsgi.get_host() no longer looks at X-Forwarded-For . Use ProxyFix to handle that. ( #609 , #1303 )

ProxyFix is refactored to support more headers, multiple values, and more secure configuration.

Each header supports multiple values. The trusted number of proxies is configured separately for each header. The num_proxies argument is deprecated. ( #1314 )

Sets SERVER_NAME and SERVER_PORT based on X-Forwarded-Host . ( #1314 )

Sets SERVER_PORT and modifies HTTP_HOST based on X-Forwarded-Port . ( #1023 , #1304 )

Sets SCRIPT_NAME based on X-Forwarded-Prefix . ( #1237 )

The original WSGI environment values are stored in the werkzeug.proxy_fix.orig key, a dict. The individual keys werkzeug.proxy_fix.orig_remote_addr , werkzeug.proxy_fix.orig_wsgi_url_scheme , and werkzeug.proxy_fix.orig_http_host are deprecated.

Middleware from werkzeug.wsgi has moved to separate modules under werkzeug.middleware , along with the middleware moved from werkzeug.contrib . The old werkzeug.wsgi imports are deprecated and will be removed in version 1.0. ( #1452 )

werkzeug.wsgi.DispatcherMiddleware has moved to werkzeug.middleware.dispatcher.DispatcherMiddleware .

werkzeug.wsgi.ProxyMiddleware as moved to werkzeug.middleware.http_proxy.ProxyMiddleware .

werkzeug.wsgi.SharedDataMiddleware has moved to werkzeug.middleware.shared_data.SharedDataMiddleware .

ProxyMiddleware proxies the query string. ( #1252 )

The filenames generated by ProfilerMiddleware can be customized. ( #1283 )

The werkzeug.wrappers module has been converted to a package, and its various classes have been organized into separate modules. Any previously documented classes, understood to be the existing public API, are still importable from werkzeug.wrappers , or may be imported from their specific modules. ( #1456 )

Resolved a regression with status code handling in the integrated development server.

HTTP exceptions are now automatically caught by Request.application .

Added support for edge as browser.

Added support for platforms that lack SpooledTemporaryFile .

Add support for etag handling through if-match

Added support for the SameSite cookie attribute.

Added werkzeug.wsgi.ProxyMiddleware

Implemented has for NullCache

get_multi on cache clients now returns lists all the time.

Improved the watchdog observer shutdown for the reloader to not crash on exit on older Python versions.

Added support for filename* filename attributes according to RFC 2231

Resolved an issue where machine ID for the reloader PIN was not read accurately on windows.

Added a workaround for syntax errors in init files in the reloader.

Added support for using the reloader with console scripts on windows.

The built-in HTTP server will no longer close a connection in cases where no HTTP body is expected (204, 204, HEAD requests etc.)

The EnvironHeaders object now skips over empty content type and lengths if they are set to falsy values.

Werkzeug will no longer send the content-length header on 1xx or 204/304 responses.

Cookie values are now also permitted to include slashes and equal signs without quoting.

Relaxed the regex for the routing converter arguments.

If cookies are sent without values they are now assumed to have an empty value and the parser accepts this. Previously this could have corrupted cookies that followed the value.

The test Client and EnvironBuilder now support mimetypes like the request object does.

Added support for static weights in URL rules.

Better handle some more complex reloader scenarios where sys.path contained non directory paths.

EnvironHeaders no longer raises weird errors if non string keys are passed to it.

Deprecate support for Python 2.6 and 3.3. CI tests will not run for these versions, and support will be dropped completely in the next version. ( pallets/meta#24 )

Raise TypeError when port is not an integer. ( #1088 )

Fully deprecate werkzeug.script . Use Click instead. ( #1090 )

response.age is parsed as a timedelta . Previously, it was incorrectly treated as a datetime . The header value is an integer number of seconds, not a date string. ( #414 )

Fix a bug in TypeConversionDict where errors are not propagated when using the converter. ( #1102 )

Authorization.qop is a string instead of a set, to comply with RFC 2617. ( #984 )

An exception is raised when an encoded cookie is larger than, by default, 4093 bytes. Browsers may silently ignore cookies larger than this. BaseResponse has a new attribute max_cookie_size and dump_cookie has a new argument max_size to configure this. ( #780 , #1109 )

Fix a TypeError in werkzeug.contrib.lint.GuardedIterator.close . ( #1116 )

BaseResponse.calculate_content_length now correctly works for Unicode responses on Python 3. It first encodes using iter_encoded . ( #705 )

Secure cookie contrib works with string secret key on Python 3. ( #1205 )

Shared data middleware accepts a list instead of a dict of static locations to preserve lookup order. ( #1197 )

HTTP header values without encoding can contain single quotes. ( #1208 )

The built-in dev server supports receiving requests with chunked transfer encoding. ( #1198 )

Fix regression: Pull request #892 prevented Werkzeug from correctly logging the IP of a remote client behind a reverse proxy, even when using ProxyFix .

Fix a bug in safe_join on Windows.

Fix crash of reloader (used on debug mode) on Windows. ( OSError: [WinError 10038] ). See pull request #1081

Partially revert change to class hierarchy of Headers . See #1084 .

Spit out big deprecation warnings for werkzeug.script

Use inspect.getfullargspec internally when available as inspect.getargspec is gone in 3.6

Added support for status code 451 and 423

Improved the build error suggestions. In particular only if someone stringifies the error will the suggestions be calculated.

Added support for uWSGI’s caching backend.

Fix a bug where iterating over a FileStorage would result in an infinite loop.

Datastructures now inherit from the relevant baseclasses from the collections module in the stdlib. See #794.

Add support for recognizing NetBSD, OpenBSD, FreeBSD, DragonFlyBSD platforms in the user agent string.

Recognize SeaMonkey browser name and version correctly

Recognize Baiduspider, and bingbot user agents

If LocalProxy ’s wrapped object is a function, refer to it with __wrapped__ attribute.

The defaults of generate_password_hash have been changed to more secure ones, see pull request #753 .

Add support for encoding in options header parsing, see pull request #933 .

test.Client now properly handles Location headers with relative URLs, see pull request #879 .

When HTTPException is raised, it now prints the description, for easier debugging.

Werkzeug’s dict-like datastructures now have view -methods under Python 2, see pull request #968 .

Fix a bug in MultiPartParser when no stream_factory was provided during initialization, see pull request #973 .

Disable autocorrect and spellchecker in the debugger middleware’s Python prompt, see pull request #994 .

Don’t redirect to slash route when method doesn’t match, see pull request #907 .

Fix a bug when using SharedDataMiddleware with frozen packages, see pull request #959 .

Range header parsing function fixed for invalid values #974 .

Add support for byte Range Requests, see pull request #978 .

Use modern cryptographic defaults in the dev servers #1004 .

the post() method of the test client now accept file object through the data parameter.

Color run_simple’s terminal output based on HTTP codes #1013 .

Fix self-XSS in debugger console, see #1031 .

Fix IPython 5.x shell support, see #1033 .

Change Accept datastructure to sort by specificity first, allowing for more accurate results when using best_match for mime types (for example in requests.accept_mimetypes.best_match )

werkzeug.serving: set CONTENT_TYPE / CONTENT_LENGTH if only they’re provided by the client

werkzeug.serving: Fix crash of reloader when using python -m werkzeug.serving .

Bugfix for the bugfix in the previous release.

Check if platform can fork before importing ForkingMixIn , raise exception when creating ForkingWSGIServer on such a platform, see PR #999 .

Correct fix for the reloader issuer on certain Windows installations.

Fix more bugs in multidicts regarding empty lists. See #1000 .

Add some docstrings to some EnvironBuilder properties that were previously unintentionally missing.

Added a workaround for the reloader on windows.

Fix JSONRequestMixin for Python3. See #731

Fix broken string handling in test client when passing integers. See #852

Fix a bug in parse_options_header where an invalid content type starting with comma or semi-colon would result in an invalid return value, see issue #995 .

Fix a bug in multidicts when passing empty lists as values, see issue #979 .

Fix a security issue that allows XSS on the Werkzeug debugger. See #1001 .

Fixed a bug that occurs when running on Python 2.6 and using a broken locale. See pull request #912.

Fixed a crash when running the debugger on Google App Engine. See issue #925.

Fixed an issue with multipart parsing that could cause memory exhaustion.

Corrected an issue that caused the debugger not to use the machine GUID on POSIX systems.

Corrected a Unicode error on Python 3 for the debugger’s PIN usage.

Corrected the timestamp verification in the pin debug code. Without this fix the pin was remembered for too long.

fixed a problem with the machine GUID detection code on OS X on Python 3.

fixed a regression on Python 3 for the debugger.

werkzeug.serving: Still show the client address on bad requests.

improved the PIN based protection for the debugger to make it harder to brute force via trying cookies. Please keep in mind that the debugger is not intended for running on production environments

increased the pin timeout to a week to make it less annoying for people which should decrease the chance that users disable the pin check entirely.

werkzeug.serving: Fix broken HTTP_HOST when path starts with double slash.

werkzeug.serving: Fix crash when attempting SSL connection to HTTP server.

Fixed werkzeug.serving not working from -m flag.

Fixed incorrect weak etag handling.

Fixed an issue with copy operations not working against proxies.

Changed the logging operations of the development server to correctly log where the server is running in all situations again.

Fixed another regression with SSL wrapping similar to the fix in 0.11.2 but for a different code path.

Fix inheritable sockets on Windows on Python 3.

Fixed an issue with the forking server not starting any longer.

Fixed SSL wrapping on platforms that supported opening sockets by file descriptor.

No longer log from the watchdog reloader.

Unicode errors in hosts are now better caught or converted into bad request errors.

Fixed a regression on Python 3 in the debugger.

Added reloader_paths option to run_simple and other functions in werkzeug.serving . This allows the user to completely override the Python module watching of Werkzeug with custom paths.

Many custom cached properties of Werkzeug’s classes are now subclasses of Python’s property type (issue #616 ).

bind_to_environ now doesn’t differentiate between implicit and explicit default port numbers in HTTP_HOST (pull request #204 ).

BuildErrors are now more informative. They come with a complete sentence as error message, and also provide suggestions (pull request #691 ).

Fix a bug in the user agent parser where Safari’s build number instead of version would be extracted (pull request #703 ).

Fixed issue where RedisCache set_many was broken for twemproxy, which doesn’t support the default MULTI command (pull request #702 ).

mimetype parameters on request and response classes are now always converted to lowercase.

Changed cache so that cache never expires if timeout is 0. This also fixes an issue with redis setex (issue #550 )

Werkzeug now assumes UTF-8 as filesystem encoding on Unix if Python detected it as ASCII.

New optional has method on caches.

Fixed various bugs in parse_options_header (pull request #643 ).

If the reloader is enabled the server will now open the socket in the parent process if this is possible. This means that when the reloader kicks in the connection from client will wait instead of tearing down. This does not work on all Python versions.

Implemented PIN based authentication for the debugger. This can optionally be disabled but is discouraged. This change was necessary as it has been discovered that too many people run the debugger in production.

Devserver no longer requires SSL module to be installed.

Reloader: Correctly detect file changes made by moving temporary files over the original, which is e.g. the case with PyCharm (pull request #722 ).

Fix bool behavior of werkzeug.datastructures.ETags under Python 3 (issue #744 ).

Re-release of 0.10.3 with packaging artifacts manually removed.

Re-release of 0.10.2 without packaging artifacts.

Fixed issue where empty could break third-party libraries that relied on keyword arguments (pull request #675 )

Improved Rule.empty by providing a `get_empty_kwargs to allow setting custom kwargs without having to override entire empty method. (pull request #675 )

Fixed `extra_files` parameter for reloader to not cause startup to crash when included in server params

Using MultiDict when building URLs is now not supported again. The behavior introduced several regressions.

Fix performance problems with stat-reloader (pull request #715 ).

Fixed regression with multiple query values for URLs (pull request #667 ).

Fix issues with eventlet’s monkeypatching and the builtin server (pull request #663 ).

Changed the error handling of and improved testsuite for the caches in contrib.cache .

Fixed a bug on Python 3 when creating adhoc ssl contexts, due to sys.maxint not being defined.

Fixed a bug on Python 3, that caused make_ssl_devcert() to fail with an exception.

Added exceptions for 504 and 505.

Added support for ChromeOS detection.

Added UUID converter to the routing system.

Added message that explains how to quit the server.

Fixed a bug on Python 2, that caused len for werkzeug.datastructures.CombinedMultiDict to crash.

Added support for stdlib pbkdf2 hmac if a compatible digest is found.

Ported testsuite to use py.test .

Minor optimizations to various middlewares (pull requests #496 and #571 ).

Use stdlib ssl module instead of OpenSSL for the builtin server (issue #434 ). This means that OpenSSL contexts are not supported anymore, but instead ssl.SSLContext from the stdlib.

Allow protocol-relative URLs when building external URLs.

Fixed Atom syndication to print time zone offset for tz-aware datetime objects (pull request #254 ).

Improved reloader to track added files and to recover from broken sys.modules setups with syntax errors in packages.

cache.RedisCache now supports arbitrary **kwargs for the redis object.

werkzeug.test.Client now uses the original request method when resolving 307 redirects (pull request #556 ).

werkzeug.datastructures.MIMEAccept now properly deals with mimetype parameters (pull request #205 ).

werkzeug.datastructures.Accept now handles a quality of 0 as intolerable, as per RFC 2616 (pull request #536 ).

werkzeug.urls.url_fix now properly encodes hostnames with idna encoding (issue #559 ). It also doesn’t crash on malformed URLs anymore (issue #582 ).

werkzeug.routing.MapAdapter.match now recognizes the difference between the path / and an empty one (issue #360 ).

The interactive debugger now tries to decode non-ascii filenames (issue #469 ).

Increased default key size of generated SSL certificates to 1024 bits (issue #611 ).

Added support for specifying a Response subclass to use when calling redirect() .

werkzeug.test.EnvironBuilder now doesn’t use the request method anymore to guess the content type, and purely relies on the form , files and input_stream properties (issue #620 ).

Added Symbian to the user agent platform list.

Fixed make_conditional to respect automatically_set_content_length

Unset Content-Length when writing to response.stream (issue #451 )

wrappers.Request.method is now always uppercase, eliminating inconsistencies of the WSGI environment (issue 647 ).

routing.Rule.empty now works correctly with subclasses of Rule (pull request #645 ).

Made map updating safe in light of concurrent updates.

Allow multiple values for the same field for url building (issue #658 ).

Fix unicode problems in werkzeug.debug.tbtools .

Fix Python 3-compatibility problems in werkzeug.posixemulation .

Backport fix of fatal typo for ImmutableList (issue #492 ).

Make creation of the cache dir for FileSystemCache atomic (issue #468 ).

Use native strings for memcached keys to work with Python 3 client (issue #539 ).

Fix charset detection for werkzeug.debug.tbtools.Frame objects (issues #547 and #532 ).

Fix AttributeError masking in werkzeug.utils.import_string (issue #182 ).

Explicitly shut down server (issue #519 ).

Fix timeouts greater than 2592000 being misinterpreted as UNIX timestamps in werkzeug.contrib.cache.MemcachedCache (issue #533 ).

Fix bug where werkzeug.exceptions.abort would raise an arbitrary subclass of the expected class (issue #422 ).

Fix broken jsrouting (due to removal of werkzeug.templates )

werkzeug.urls.url_fix now doesn’t crash on malformed URLs anymore, but returns them unmodified. This is a cheap workaround for #582 , the proper fix is included in version 0.10.

The repr of werkzeug.wrappers.Request doesn’t crash on non-ASCII-values anymore (pull request #466 ).

Fix bug in cache.RedisCache when combined with redis.StrictRedis object (pull request #583 ).

The qop parameter for WWW-Authenticate headers is now always quoted, as required by RFC 2617 (issue #633 ).

Fix bug in werkzeug.contrib.cache.SimpleCache with Python 3 where add/set may throw an exception when pruning old entries from the cache (pull request #651 ).

Added a safe conversion for IRI to URI conversion and use that internally to work around issues with spec violations for protocols such as itms-service .

Fixed uri_to_iri() not re-encoding hashes in query string parameters.

Forward charset argument from request objects to the environ builder.

Fixed error handling for missing boundaries in multipart data.

Fixed session creation on systems without os.urandom() .

Fixed pluses in dictionary keys not being properly URL encoded.

Fixed a problem with deepcopy not working for multi dicts.

Fixed a double quoting issue on redirects.

Fixed a problem with unicode keys appearing in headers on 2.x.

Fixed a bug with unicode strings in the test builder.

Fixed a unicode bug on Python 3 in the WSGI profiler.

Fixed an issue with the safe string compare function on Python 2.7.7 and Python 3.4.

Fixed an issue with Python 3.3 and an edge case in cookie parsing.

Fixed decoding errors not handled properly through the WSGI decoding dance.

Fixed URI to IRI conversion incorrectly decoding percent signs.

Restored behavior of the data descriptor of the request class to pre 0.9 behavior. This now also means that .data and .get_data() have different behavior. New code should use .get_data() always.

In addition to that there is now a flag for the .get_data() method that controls what should happen with form data parsing and the form parser will honor cached data. This makes dealing with custom form data more consistent.

Added unsafe parameter to url_quote() .

Fixed an issue with url_quote_plus() not quoting ‘+’ correctly.

Ported remaining parts of RedisCache to Python 3.3.

Ported remaining parts of MemcachedCache to Python 3.3

Fixed a deprecation warning in the contrib atom module.

Fixed a regression with setting of content types through the headers dictionary instead with the content type parameter.

Use correct name for stdlib secure string comparison function.

Fixed a wrong reference in the docstring of release_local() .

Fixed an AttributeError that sometimes occurred when accessing the werkzeug.wrappers.BaseResponse.is_streamed attribute.

Fixed an issue with integers no longer being accepted in certain parts of the routing system or URL quoting functions.

Fixed an issue with url_quote not producing the right escape codes for single digit codepoints.

Fixed an issue with SharedDataMiddleware not reading the path correctly and breaking on etag generation in some cases.

Properly handle Expect: 100-continue in the development server to resolve issues with curl.

Automatically exhaust the input stream on request close. This should fix issues where not touching request files results in a timeout.

Fixed exhausting of streams not doing anything if a non-limited stream was passed into the multipart parser.

Raised the buffer sizes for the multipart parser.

Added support for tell() on the limited stream.

ETags now is nonzero if it contains at least one etag of any kind, including weak ones.

Added a workaround for a bug in the stdlib for SSL servers.

Improved SSL interface of the devserver so that it can generate certificates easily and load them from files.

Refactored test client to invoke the open method on the class for redirects. This makes subclassing more powerful.

werkzeug.wsgi.make_chunk_iter() and werkzeug.wsgi.make_line_iter() now support processing of iterators and streams.

URL generation by the routing system now no longer quotes

URL fixing now no longer quotes certain reserved characters.

The werkzeug.security.generate_password_hash() and check functions now support any of the hashlib algorithms.

wsgi.get_current_url is now ascii safe for browsers sending non-ascii data in query strings.

improved parsing behavior for werkzeug.http.parse_options_header()

added more operators to local proxies.

added a hook to override the default converter in the routing system.

The description field of HTTP exceptions is now always escaped. Use markup objects to disable that.

Added number of proxy argument to the proxy fix to make it more secure out of the box on common proxy setups. It will by default no longer trust the x-forwarded-for header as much as it did before.

Added support for fragment handling in URI/IRI functions.

Added custom class support for werkzeug.http.parse_dict_header() .

Renamed LighttpdCGIRootFix to CGIRootFix .

Always treat + as safe when fixing URLs as people love misusing them.

Added support to profiling into directories in the contrib profiler.

The escape function now by default escapes quotes.

Changed repr of exceptions to be less magical.

Simplified exception interface to no longer require environments to be passed to receive the response object.

Added sentinel argument to IterIO objects.

Added pbkdf2 support for the security module.

Added a plain request type that disables all form parsing to only leave the stream behind.

Removed support for deprecated fix_headers .

Removed support for deprecated header_list .

Removed support for deprecated parameter for iter_encoded .

Removed support for deprecated non-silent usage of the limited stream object.

Removed support for previous dummy writable parameter on the cached property.

Added support for explicitly closing request objects to close associated resources.

Conditional request handling or access to the data property on responses no longer ignores direct passthrough mode.

Removed werkzeug.templates and werkzeug.contrib.kickstart.

Changed host lookup logic for forwarded hosts to allow lists of hosts in which case only the first one is picked up.

Added wsgi.get_query_string , wsgi.get_path_info and wsgi.get_script_name and made the wsgi.pop_path_info and wsgi.peek_path_info functions perform unicode decoding. This was necessary to avoid having to expose the WSGI encoding dance on Python 3.

Added content_encoding and content_md5 to the request object’s common request descriptor mixin.

added options and trace to the test client.

Overhauled the utilization of the input stream to be easier to use and better to extend. The detection of content payload on the input side is now more compliant with HTTP by detecting off the content type header instead of the request method. This also now means that the stream property on the request class is always available instead of just when the parsing fails.

Added support for using werkzeug.wrappers.BaseResponse in a with statement.

Changed get_app_iter to fetch the response early so that it does not fail when wrapping a response iterable. This makes filtering easier.

Introduced get_data and set_data methods for responses.

Introduced get_data for requests.

Soft deprecated the data descriptors for request and response objects.

Added as_bytes operations to some of the headers to simplify working with things like cookies.

Made the debugger paste tracebacks into github’s gist service as private pastes.

Added a favicon to the debugger which fixes problem with state changes being triggered through a request to /favicon.ico in Google Chrome. This should fix some problems with Flask and other frameworks that use context local objects on a stack with context preservation on errors.

Fixed an issue with scrolling up in the debugger.

Fixed an issue with debuggers running on a different URL than the URL root.

Fixed a problem with proxies not forwarding some rarely used special methods properly.

Added a workaround to prevent the XSS protection from Chrome breaking the debugger.

Skip redis tests if redis is not running.

Fixed a typo in the multipart parser that caused content-type to not be picked up properly.

Fixed another issue with werkzeug.wsgi.make_line_iter() where lines longer than the buffer size were not handled properly.

Restore stdout after debug console finished executing so that the debugger can be used on GAE better.

Fixed a bug with the redis cache for int subclasses (affects bool caching).

Fixed an XSS problem with redirect targets coming from untrusted sources.

Redis cache backend now supports password authentication.

Fixed a problem with request handling of the builtin server not responding to socket errors properly.

The routing request redirect exception’s code attribute is now used properly.

Fixed a bug with shutdowns on Windows.

Fixed a few unicode issues with non-ascii characters being hardcoded in URL rules.

Fixed two property docstrings being assigned to fdel instead of __doc__ .

Fixed an issue where CRLF line endings could be split into two by the line iter function, causing problems with multipart file uploads.

Fixed an issue with the memcache not working properly.

Fixed an issue for Python 2.7.1 and higher that broke copying of multidicts with copy.copy() .

Changed hashing methodology of immutable ordered multi dicts for a potential problem with alternative Python implementations.

Removed data structure specific KeyErrors for a general purpose BadRequestKeyError .

Documented werkzeug.wrappers.BaseRequest._load_form_data() .

The routing system now also accepts strings instead of dictionaries for the query_args parameter since we’re only passing them through for redirects.

Werkzeug now automatically sets the content length immediately when the data attribute is set for efficiency and simplicity reasons.

The routing system will now normalize server names to lowercase.

The routing system will no longer raise ValueErrors in case the configuration for the server name was incorrect. This should make deployment much easier because you can ignore that factor now.

Fixed a bug with parsing HTTP digest headers. It rejected headers with missing nc and nonce params.

Proxy fix now also updates wsgi.url_scheme based on X-Forwarded-Proto.

Added support for key prefixes to the redis cache.

Added the ability to suppress some auto corrections in the wrappers that are now controlled via autocorrect_location_header and automatically_set_content_length on the response objects.

Werkzeug now uses a new method to check that the length of incoming data is complete and will raise IO errors by itself if the server fails to do so.

make_line_iter() now requires a limit that is not higher than the length the stream can provide.

Refactored form parsing into a form parser class that makes it possible to hook into individual parts of the parsing process for debugging and extending.

For conditional responses the content length is no longer set when it is already there and added if missing.

Immutable datastructures are hashable now.

Headers datastructure no longer allows newlines in values to avoid header injection attacks.

Made it possible through subclassing to select a different remote addr in the proxy fix.

Added stream based URL decoding. This reduces memory usage on large transmitted form data that is URL decoded since Werkzeug will no longer load all the unparsed data into memory.

Memcache client now no longer uses the buggy cmemcache module and supports pylibmc. GAE is not tried automatically and the dedicated class is no longer necessary.

Redis cache now properly serializes data.

Removed support for Python 2.4

Fixed a CSRF problem with the debugger.

The debugger is now generating private pastes on lodgeit.

If URL maps are now bound to environments the query arguments are properly decoded from it for redirects.

Fixed a problem with newer versions of IPython.

Disabled pyinotify based reloader which does not work reliably.

Add support for python-libmemcached to the Werkzeug cache abstraction layer.

Improved url_decode() and url_encode() performance.

Fixed an issue where the SharedDataMiddleware could cause an internal server error on weird paths when loading via pkg_resources.

Fixed an URL generation bug that caused URLs to be invalid if a generated component contains a colon.

werkzeug.import_string() now works with partially set up packages properly.

Disabled automatic socket switching for IPv6 on the development server due to problems it caused.

Werkzeug no longer overrides the Date header when creating a conditional HTTP response.

The routing system provides a method to retrieve the matching methods for a given path.

The routing system now accepts a parameter to change the encoding error behaviour.

The local manager can now accept custom ident functions in the constructor that are forwarded to the wrapped local objects.

url_unquote_plus now accepts unicode strings again.

Fixed an issue with the filesystem session support’s prune function and concurrent usage.

Fixed a problem with external URL generation discarding the port.

Added support for pylibmc to the Werkzeug cache abstraction layer.

Fixed an issue with the new multipart parser that happened when a linebreak happened to be on the chunk limit.

Cookies are now set properly if ports are in use. A runtime error is raised if one tries to set a cookie for a domain without a dot.

Fixed an issue with Template.from_file not working for file descriptors.

Reloader can now use inotify to track reloads. This requires the pyinotify library to be installed.

Werkzeug debugger can now submit to custom lodgeit installations.

redirect function’s status code assertion now allows 201 to be used as redirection code. While it’s not a real redirect, it shares enough with redirects for the function to still be useful.

Fixed securecookie for pypy.

Fixed ValueErrors being raised on calls to best_match on MIMEAccept objects when invalid user data was supplied.

Deprecated werkzeug.contrib.kickstart and werkzeug.contrib.testtools

URL routing now can be passed the URL arguments to keep them for redirects. In the future matching on URL arguments might also be possible.

Header encoding changed from utf-8 to latin1 to support a port to Python 3. Bytestrings passed to the object stay untouched which makes it possible to have utf-8 cookies. This is a part where the Python 3 version will later change in that it will always operate on latin1 values.

Fixed a bug in the form parser that caused the last character to be dropped off if certain values in multipart data are used.

Multipart parser now looks at the part-individual content type header to override the global charset.

Introduced mimetype and mimetype_params attribute for the file storage object.

Changed FileStorage filename fallback logic to skip special filenames that Python uses for marking special files like stdin.

Introduced more HTTP exception classes.

call_on_close now can be used as a decorator.

Support for redis as cache backend.

Added BaseRequest.scheme .

Support for the RFC 5789 PATCH method.

New custom routing parser and better ordering.

Removed support for is_behind_proxy . Use a WSGI middleware instead that rewrites the REMOTE_ADDR according to your setup. Also see the werkzeug.contrib.fixers.ProxyFix for a drop-in replacement.

Added cookie forging support to the test client.

Added support for host based matching in the routing system.

Switched from the default ‘ignore’ to the better ‘replace’ unicode error handling mode.

The builtin server now adds a function named ‘werkzeug.server.shutdown’ into the WSGI env to initiate a shutdown. This currently only works in Python 2.6 and later.

Headers are now assumed to be latin1 for better compatibility with Python 3 once we have support.

Added werkzeug.security.safe_join() .

Added accept_json property analogous to accept_html on the werkzeug.datastructures.MIMEAccept .

werkzeug.utils.import_string() now fails with much better error messages that pinpoint to the problem.

Added support for parsing of the If-Range header ( werkzeug.http.parse_if_range_header() and werkzeug.datastructures.IfRange ).

Added support for parsing of the Range header ( werkzeug.http.parse_range_header() and werkzeug.datastructures.Range ).

Added support for parsing of the Content-Range header of responses and provided an accessor object for it ( werkzeug.http.parse_content_range_header() and werkzeug.datastructures.ContentRange ).

renamed the attribute implicit_seqence_conversion attribute of the request object to implicit_sequence_conversion .

heavily improved local objects. Should pick up standalone greenlet builds now and support proxies to free callables as well. There is also a stacked local now that makes it possible to invoke the same application from within itself by pushing current request/response on top of the stack.

routing build method will also build non-default method rules properly if no method is provided.

added proper IPv6 support for the builtin server.

windows specific filesystem session store fixes. (should now be more stable under high concurrency)

fixed a NameError in the session system.

fixed a bug with empty arguments in the werkzeug.script system.

fixed a bug where log lines will be duplicated if an application uses logging.basicConfig() (#499)

added secure password hashing and checking functions.

HEAD is now implicitly added as method in the routing system if GET is present. Not doing that was considered a bug because often code assumed that this is the case and in web servers that do not normalize HEAD to GET this could break HEAD requests.

the script support can start SSL servers now.

removed pending deprecations

sys.path is now printed from the testapp.

fixed an RFC 2068 incompatibility with cookie value quoting.

the FileStorage now gives access to the multipart headers.

cached_property.writeable has been deprecated.

MapAdapter.match() now accepts a return_rule keyword argument that returns the matched Rule instead of just the endpoint

routing.Map.bind_to_environ() raises a more correct error message now if the map was bound to an invalid WSGI environment.

added support for SSL to the builtin development server.

Response objects are no longer modified in place when they are evaluated as WSGI applications. For backwards compatibility the fix_headers function is still called in case it was overridden. You should however change your application to use get_wsgi_headers if you need header modifications before responses are sent as the backwards compatibility support will go away in future versions.

append_slash_redirect() no longer requires the QUERY_STRING to be in the WSGI environment.

added DynamicCharsetResponseMixin

added DynamicCharsetRequestMixin

added BaseRequest.url_charset

request and response objects have a default __repr__ now.

builtin data structures can be pickled now.

the form data parser will now look at the filename instead the content type to figure out if it should treat the upload as regular form data or file upload. This fixes a bug with Google Chrome.

improved performance of make_line_iter and the multipart parser for binary uploads.

fixed is_streamed

fixed a path quoting bug in EnvironBuilder that caused PATH_INFO and SCRIPT_NAME to end up in the environ unquoted.

werkzeug.BaseResponse.freeze() now sets the content length.

for unknown HTTP methods the request stream is now always limited instead of being empty. This makes it easier to implement DAV and other protocols on top of Werkzeug.

added werkzeug.MIMEAccept.best_match()

multi-value test-client posts from a standard dictionary are now supported. Previously you had to use a multi dict.

rule templates properly work with submounts, subdomains and other rule factories now.

deprecated non-silent usage of the werkzeug.LimitedStream .

added support for IRI handling to many parts of Werkzeug.

development server properly logs to the werkzeug logger now.

added werkzeug.extract_path_info()

fixed a querystring quoting bug in url_fix()

added fallback_mimetype to werkzeug.SharedDataMiddleware .

deprecated BaseResponse.iter_encoded() ’s charset parameter.

added BaseResponse.make_sequence() , BaseResponse.is_sequence and BaseResponse._ensure_sequence() .

added better __repr__ of werkzeug.Map

import_string accepts unicode strings as well now.

development server doesn’t break on double slashes after the host name.

better __repr__ and __str__ of werkzeug.exceptions.HTTPException

test client works correctly with multiple cookies now.

the werkzeug.routing.Map now has a class attribute with the default converter mapping. This helps subclasses to override the converters without passing them to the constructor.

implemented OrderedMultiDict

improved the session support for more efficient session storing on the filesystem. Also added support for listing of sessions currently stored in the filesystem session store.

werkzeug no longer utilizes the Python time module for parsing which means that dates in a broader range can be parsed.

the wrappers have no class attributes that make it possible to swap out the dict and list types it uses.

werkzeug debugger should work on the appengine dev server now.

the URL builder supports dropping of unexpected arguments now. Previously they were always appended to the URL as query string.

profiler now writes to the correct stream.

fixed boolean check of FileStorage

url routing system properly supports unicode URL rules now.

file upload streams no longer have to provide a truncate() method.

implemented BaseRequest._form_parsing_failed() .

fixed #394

ImmutableDict.copy() , ImmutableMultiDict.copy() and ImmutableTypeConversionDict.copy() return mutable shallow copies.

fixed a bug with the make_runserver script action.

MultiDict.items() and MutiDict.iteritems() now accept an argument to return a pair for each value of each key.

the multipart parser works better with hand-crafted multipart requests now that have extra newlines added. This fixes a bug with setuptools uploads not handled properly (#390)

fixed some minor bugs in the atom feed generator.

fixed a bug with client cookie header parsing being case sensitive.

fixed a not-working deprecation warning.

fixed package loading for SharedDataMiddleware .

fixed a bug in the secure cookie that made server-side expiration on servers with a local time that was not set to UTC impossible.

fixed console of the interactive debugger.

requires Python 2.4 now

fixed a bug in IterIO

added MIMEAccept and CharsetAccept that work like the regular Accept but have extra special normalization for mimetypes and charsets and extra convenience methods.

switched the serving system from wsgiref to something homebrew.

the Client now supports cookies.

added the fixers module with various fixes for webserver bugs and hosting setup side-effects.

added werkzeug.contrib.wrappers

added is_hop_by_hop_header()

added is_entity_header()

added remove_hop_by_hop_headers()

added pop_path_info()

added peek_path_info()

added wrap_file() and FileWrapper

moved LimitedStream from the contrib package into the regular werkzeug one and changed the default behavior to raise exceptions rather than stopping without warning. The old class will stick in the module until 0.6.

implemented experimental multipart parser that replaces the old CGI hack.

added dump_options_header() and parse_options_header()

added quote_header_value() and unquote_header_value()

url_encode() and url_decode() now accept a separator argument to switch between & and ; as pair separator. The magic switch is no longer in place.

all form data parsing functions as well as the BaseRequest object have parameters (or attributes) to limit the number of incoming bytes (either totally or per field).

added LanguageAccept

request objects are now enforced to be read only for all collections.

added many new collection classes, refactored collections in general.

test support was refactored, semi-undocumented werkzeug.test.File was replaced by werkzeug.FileStorage .

EnvironBuilder was added and unifies the previous distinct create_environ() , Client and BaseRequest.from_values() . They all work the same now which is less confusing.

officially documented imports from the internal modules as undefined behavior. These modules were never exposed as public interfaces.

removed FileStorage.__len__ which previously made the object falsy for browsers not sending the content length which all browsers

SharedDataMiddleware uses wrap_file now and has a configurable cache timeout.

added CommonRequestDescriptorsMixin

added CommonResponseDescriptorsMixin.mimetype_params

added werkzeug.contrib.lint

added passthrough_errors to run_simple .

added secure_filename

added make_line_iter()

MultiDict copies now instead of revealing internal lists to the caller for getlist and iteration functions that return lists.

added follow_redirect to the open() of Client .

added support for extra_files in make_runserver()

werkzeug.contrib.cache.Memcached accepts now objects that implement the memcache.Client interface as alternative to a list of strings with server addresses. There is also now a GAEMemcachedCache that connects to the Google appengine cache.

explicitly convert secret keys to bytestrings now because Python 2.6 no longer does that.

url_encode and all interfaces that call it, support ordering of options now which however is disabled by default.

the development server no longer resolves the addresses of clients.

Fixed a typo in werkzeug.test that broke File .

Map.bind_to_environ uses the Host header now if available.

Fixed BaseCache.get_dict (#345)

werkzeug.test.Client can now run the application buffered in which case the application is properly closed automatically.

Fixed Headers.set (#354). Caused header duplication before.

Fixed Headers.pop (#349). default parameter was not properly handled.

Fixed UnboundLocalError in create_environ (#351)

Headers is more compatible with wsgiref now.

Template.render accepts multidicts now.

dropped support for Python 2.3

Client supports an empty data argument now.

fixed a bug in Response.application that made it impossible to use it as method decorator.

the session system should work on appengine now

the secure cookie works properly in load balanced environments with different cpu architectures now.

CacheControl.no_cache and CacheControl.private behavior changed to reflect the possibilities of the HTTP RFC. Setting these attributes to None or True now sets the value to “the empty value”. More details in the documentation.

fixed werkzeug.contrib.atom.AtomFeed.__call__ . (#338)

BaseResponse.make_conditional now always returns self . Previously it didn’t for post requests and such.

fixed a bug in boolean attribute handling of html and xhtml .

added graceful error handling to the debugger pastebin feature.

added a more list like interface to Headers (slicing and indexing works now)

fixed a bug with the __setitem__ method of Headers that didn’t properly remove all keys on replacing.

added remove_entity_headers which removes all entity headers from a list of headers (or a Headers object)

the responses now automatically call remove_entity_headers if the status code is 304.

fixed a bug with Href query parameter handling. Previously the last item of a call to Href was not handled properly if it was a dict.

headers now support a pop operation to better work with environ properties.

fixed a security problem with werkzeug.contrib.SecureCookie .

added support for redirecting in url routing.

added Authorization and AuthorizationMixin

added WWWAuthenticate and WWWAuthenticateMixin

added parse_list_header

added parse_dict_header

added parse_authorization_header

added parse_www_authenticate_header

added _get_current_object method to LocalProxy objects

added parse_form_data

MultiDict , CombinedMultiDict , Headers , and EnvironHeaders raise special key errors now that are subclasses of BadRequest so if you don’t catch them they give meaningful HTTP responses.

added support for alternative encoding error handling and the new HTTPUnicodeError which (if not caught) behaves like a BadRequest .

added BadRequest.wrap .

added ETag support to the SharedDataMiddleware and added an option to disable caching.

fixed is_xhr on the request objects.

fixed error handling of the url adapter’s dispatch method. (#318)

fixed bug with SharedDataMiddleware .

fixed Accept.values .

EnvironHeaders contain content-type and content-length now

url_encode treats lists and tuples in dicts passed to it as multiple values for the same key so that one doesn’t have to pass a MultiDict to the function.

added validate_arguments

added BaseRequest.application

improved Python 2.3 support

run_simple accepts use_debugger and use_evalex parameters now, like the make_runserver factory function from the script module.

the environ_property is now read-only by default

it’s now possible to initialize requests as “shallow” requests which causes runtime errors if the request object tries to consume the input stream.

Added AnyConverter to the routing system.

Added werkzeug.contrib.securecookie

Exceptions have a get_response() method that return a response object

fixed the path ordering bug (#293), thanks Thomas Johansson

BaseReporterStream is now part of the werkzeug contrib module. From Werkzeug 0.3 onwards you will have to import it from there.

added DispatcherMiddleware .

RequestRedirect is now a subclass of HTTPException and uses a 301 status code instead of 302.

url_encode and url_decode can optionally treat keys as unicode strings now, too.

werkzeug.script has a different caller format for boolean arguments now.

renamed lazy_property to cached_property .

added import_string .

added is_* properties to request objects.

added empty() method to routing rules.

added werkzeug.contrib.profiler .

added extends to Headers .

added dump_cookie and parse_cookie .

added as_tuple to the Client .

added werkzeug.contrib.testtools .

added werkzeug.unescape

added BaseResponse.freeze

added werkzeug.contrib.atom

the HTTPExceptions accept an argument description now which overrides the default description.

the MapAdapter has a default for path info now. If you use bind_to_environ you don’t have to pass the path later.

the wsgiref subclass werkzeug uses for the dev server does not use direct sys.stderr logging any more but a logger called “werkzeug”.

implemented Href .

implemented find_modules

refactored request and response objects into base objects, mixins and full featured subclasses that implement all mixins.

added simple user agent parser

werkzeug’s routing raises MethodNotAllowed now if it matches a rule but for a different method.

many fixes and small improvements

Initial release