If we read 0 bytes, e.g. chunked encoding markup read but not the
actual data within it, don't stop reading since the server may or
may not consider it necessary to call onDataAvailable again.
Instead, we keep on reading, and although isReady likely returns
false on the next iteration, it eliminates ambiguity and ensures
the server will call onDataAvailable when more data arrives.
Closes gh-28241
With a Java 8 baseline in place for quite some time now, it no longer
makes sense to refer to features such as annotations as "Java 5
annotations".
This commit also removes old `Tiger*Tests` classes, thereby avoiding
duplicate execution of various tests.
Prior to this commit, `AsyncRestTemplate` would log errors (including
simple 404s) with WARN level. Such errors are quite common and should
not clutter logs.
This commit aligns the logging strategy with RestTemplate, using the
DEBUG level for such cases.
Fixes gh-28049
SocketUtils was introduced in Spring Framework 4.0, primarily to assist
in writing integration tests which start an external server on an
available random port. However, these utilities make no guarantee about
the subsequent availability of a given port and are therefore
unreliable. Instead of using SocketUtils to find an available local
port for a server, it is recommended that users rely on a server's
ability to start on a random port that it selects or is assigned by the
operating system. To interact with that server, the user should query
the server for the port it is currently using.
SocketUtils is now deprecated in 5.3.16 and will be removed in 6.0.
Closes gh-28052
In a small minority of cases, the multipart boundary can spread across
three incoming buffers.
Prior to this commit, MultipartParser.BodyState only supported two
buffers. If the boundary is spread across three buffers, the first
buffer of the three is sent as a whole, even though it contains the
first bytes of the boundary.
This commit fixes this bug, by enqueuing all prior buffers in a queue,
and emitting the ones that cannot contain boundary bytes.
Closes gh-27939
The failing test is for Apache HttpComponents where we cannot apply the
Content-Length together with other headers, and must pass it instead
explicitly to ReactiveEntityProducer when writing at which point it gets
set in the native headers.
Closes gh-27768
NettyHeadersAdapter::putAll uses HttpHeaders::add, which does not
overwrite existing headers. putAll should use HttpHeaders::set instead,
just like NettyHeadersAdapter::put does.
Closes gh-27887
Where unfeasible, this commit adds inline comments to explain why
try-with-resources must not be used in certain scenarios. The purpose
of the comments is to avoid accidental conversion to try-with-resources
at a later date.
Closes gh-27823
This commit ensures that only HTML references of length <
MAX_REFERENCE_SIZE are considered as potential references. This check is
possible because reference longer than 10 digits are out of bounds for
Integers.
Closes gh-1249
This commit fixes a bug in the DefaultPartHttpMessageReader, in the
check for exceeding the maximum header size. Before this commit, the
entire buffer size was considered, thus triggering an exception even
though the max header limit was not exceeded. After this commit, we only
consider the size up until the end-of-header mark (CRLFCRLF).
Furthermore, this commit increases the default maximum header size to
10k, the same default as Commons File upload.
Closes gh-27612
Prior to this commit, UndertowHeadersAdapter's remove() method violated
the java.util.Map contract by always returning null.
This commit fixes this by returning the previous list stored under the
specified key, and otherwise returning null if no previous value was
present.
Closes gh-27592
The fix made for gh-27488 resulted in a change of the default order
of codecs. This commit reverts these changes, so that the previous
order is restored.
Closes gh-27573
Brian Clozel
rstoyanchev
Sam Brannen
Arjen Poutsma
shirohoo
Juergen Hoeller
Max Demydenko
sokomishalov
Marten Deinum
d4ksn
happyWilliam0
Arjen Poutsma
no-brand
Smile