Prior to this commit, the `ReactorClientHttpRequestFactory` and the
`ReactorClientHttpRequest` would use the `Executor` from the current
event loop for performing write operations.
Depending on I/O demand, this work could be blocked and would result in
blocked Netty event loop executors and the HTTP client hanging.
This commit ensures that the client uses a separate Executor for such
operations. If the application does not provide one on the request
factory, a `Schedulers#boundedElastic` instance will be used.
Fixes gh-34707
Prior to this commit, gh-35213 allowed wildcard path elments at the
start of path patterns. This came with an additional constraint that
rejected such patterns if the pattern segment following the wildcard one
was not a literal:
* `/**/{name}` was rejected
* `/**/something/{name}` was accepted
The motivation here was to make the performance impact of wildard
patterns as small as possible at runtime.
This commit relaxes this constraint because `/**/*.js` patterns are very
popular in the security space for request matchers.
Closes gh-35686
This commit improves the reference document to better reflect the
different between `*` or `{name}` on one side, and `**` or `{*path}` on
the other.
The former patterns only consider a single path segment and its content,
while the latter variants consider zero or more path segments. This
explains why `/test/{*path}` can match `/test`.
Closes gh-35727
Prior to this commit, the `MappingMediaTypeFileExtensionResolver` would
resolve file extensions for a given media type by using a direct lookup
using the given media type provided by the request.
If the request contains a quality parameter like
"application/json;q=0.9", this would not resolve configured file
extensions for this media type.
While other media type parameters can be meaningful, the quality
parameter should not be used for lookups. This commit ensures that the
quality parameter is dropped before performing lookups.
Fixes gh-35754
Prior to this commit, the `HttpComponentsClientHttpRequestFactory` would
set the connection timeout on the request configuration. This has been
deprecated by the client itself and this value should be set while
creating the client on the connection manager itself.
This commit deprecates this method, as there is no way for the factory
to set this value anymore.
Closes gh-35748
Prior to this commit, a regexp path segment ending with a double wilcard
(like "/path**") would be incorrectly parsed as a double wildcard
segment ("/**").
This commit fixes the incorrect parsing.
See gh-35679
Prior to this commit, the `PathPattern` and `PathPatternParser` would
allow multiple-segments matching and capturing with the following:
* "/files/**" (matching 0-N segments until the end)
* "/files/{*path}" (matching 0-N segments until the end and capturing
the value as the "path" variable)
This would be only allowed as the last path element in the pattern and
the parser would reject other combinations.
This commit expands the support and allows multiple segments matching at
the beginning of the path:
* "/**/index.html" (matching 0-N segments from the start)
* "/{*path}/index.html" (matching 0-N segments until the end and capturing
the value as the "path" variable)
This does come with additional restrictions:
1. "/files/**/file.txt" and "/files/{*path}/file.txt" are invalid,
as multiple segment matching is not allowed in the middle of the
pattern.
2. "/{*path}/files/**" is not allowed, as a single "{*path}" or "/**"
element is allowed in a pattern
3. "/{*path}/{folder}/file.txt" "/**/{folder:[a-z]+}/file.txt" are
invalid because only a literal pattern is allowed right after
multiple segments path elements.
Closes gh-35679
Prior to this commit, the `IntrospectingClientHttpResponse` would try
and read the HTTP response stream in order to check for the presence of
a non-empty message body.
Developers reported that in some cases, an `EOFException` is thrown
instead of returning -1 from the `read()` method. This commit ensures
that this case is taken into account and that we report the response as
an empty body in these cases.
Closes gh-35361
Prior to this commit, gh-32097 added native support for Jetty for both
client and server integrations. The `JettyDataBufferFactory` was
promoted as a first class citizen, extracted from a private class in the
client support. To accomodate with server-side requirements, an extra
`buffer.retain()` call was performed.
While this is useful for server-side support, this introduced a bug in
the data buffer factory, as wrapping an existing chunk means that this
chunk is already retained.
This commit fixes the buffer factory implementation and moved existing
tests from mocks to actual pooled buffer implementations from Jetty.
The extra `buffer.retain()` is now done from the server support, right
before wrapping the buffer.
Fixes gh-35319
Prior to this commit, the MergedAnnotations support (specifically
AnnotationsScanner) and AnnotatedMethod did not find annotations on
overridden methods in type hierarchies with unresolved generics.
The reason for this is that ResolvableType.resolve() returns null for
such an unresolved type, which prevents the search algorithms from
considering such methods as override candidates.
For example, given the following type hierarchy, the compiler does not
generate a method corresponding to processOneAndTwo(Long, String) for
GenericInterfaceImpl. Nonetheless, one would expect an invocation of
processOneAndTwo(Long, String) to be @Transactional since it is
effectively an invocation of processOneAndTwo(Long, C) in
GenericAbstractSuperclass, which overrides/implements
processOneAndTwo(A, B) in GenericInterface, which is annotated with
@Transactional.
However, the MergedAnnotations infrastructure currently does not
determine that processOneAndTwo(Long, C) is @Transactional since it is
not able to determine that processOneAndTwo(Long, C) overrides
processOneAndTwo(A, B) because of the unresolved generic C.
interface GenericInterface<A, B> {
@Transactional
void processOneAndTwo(A value1, B value2);
}
abstract class GenericAbstractSuperclass<C> implements GenericInterface<Long, C> {
@Override
public void processOneAndTwo(Long value1, C value2) {
}
}
static GenericInterfaceImpl extends GenericAbstractSuperclass<String> {
}
To address such issues, this commit changes the logic in
AnnotationsScanner.hasSameGenericTypeParameters() and
AnnotatedMethod.isOverrideFor() so that they use
ResolvableType.toClass() instead of ResolvableType.resolve(). The
former returns Object.class for an unresolved generic which in turn
allows the search algorithms to properly detect method overrides in
such type hierarchies.
Closes gh-35342
Handle CancellationException in order to throw an HttpTimeoutException
when the timeout handler caused the cancellation.
See gh-34721
Signed-off-by: giampaolo <giampaorr@gmail.com>
fix: use timeoutHandler with a flag isTimeout
Closes gh-33973
Signed-off-by: giampaolo <giampaorr@gmail.com>
Refine the StringUtils#uriDecode method in the following ways:
- Use a StringBuilder instead of ByteArrayOutputStream, and only decode
%-encoded sequences.
- Use HexFormat.fromHexDigits to decode hex sequences.
- Decode to a byte array that is only allocated if encoded sequences are
encountered.
This commit adds another optimization mainly for the use case where
there is no encoded sequence, and updates the Javadoc of both
StringUtils#uriDecode and UriUtils#decode to match the implementation.
Signed-off-by: Patrick Strawderman <pstrawderman@netflix.com>
Co-Authored-by: Sebastien Deleuze <sebastien.deleuze@broadcom.com>
Closes gh-35253
This commit allows RestClient to handle StreamingHttpOutputMessage
properly by checking the type of the request and invoking setBody()
when appropriate. This improves interoperability with components that
expect streamed output.
A new integration test has been added to verify the functionality.
See gh-35102
Signed-off-by: Daniil Razorenov <daniltmb@gmail.com>
This commit fixes a regression introduced in gh-34971 where GET and
DELETE requests would not allow request bodies anymore for
`JdkClientHttpRequest`.
We are now using `builder.GET()` and `builder.DELETE()` methods only if
the provided body is null.
Fixes gh-35068
Prior to this commit, the `DefaultServerWebExchange` would attempt to
decode request bodies as form-data or multipart of the request
content-type was compatible with the expected media types.
If requests are sent with an invalid wildcard content-type such as "*/*"
or "multipart/*", we should not attempt to decode here.
Fixes gh-34660
Prior to this commit, the WebFlux server support would try reading
form-data from the request by:
* first, checking that request content-type is compatible with a
form-data content-type
* then by selecting a message reader that is compatible with the given
request content-type
This approach is flawed because if the content-type provided by the
request is too broad, another message reader could be selected that's
not meant to be used for reading form-data. Typically, a JSON message
reader could be selected and would fail when reading the request.
This problem was previously hidden because message readers would not
support `MultiValueMap` as a target type. Now that some readers support
this type, this can lead to deserialization errors.
This commit now ensures that in all cases, we attempt to read form-data
with a message reader that supports the
"application/x-www-form-urlencoded" media type.
Fixes gh-34660
Prior to this commit, the "filename" parameter value for the
"Content-Disposition" header would contain non-printable characters,
causing parsing issues for HTTP clients.
This commit ensures that all non-printable characters are encoded.
Fixes gh-35034
Dmitry Sulman
Brian Clozel
rstoyanchev
Sam Brannen
Juergen Hoeller
Anton Xu
Igor Murzich
秦利斌
giampaolo
Patrick Strawderman
Sébastien Deleuze
Daniil Razorenov
Spring Builds
Johnny Lim