Prior to this commit, when errors happened before the response was
committed, the `Content-Length` response header would be left as is.
This can be problematic since the error can be handled later in the
chain and the response body changed accordingly. For example, Spring
Boot renders error pages in those cases. If the `Content-Length` is set,
HTTP clients can get confused and only consider part of the error
response body.
This commit ensures that any `Content-Length` response header is removed
in case of errors, if the response is not already committed.
This is done at the `AbstractServerHttpResponse` level, since errors can
be handled in multiple places and the response itself is the safest
place to handle this case.
As a consequence, this commit also removes `Content-Length` checks in
`EncoderHttpMessageWriter` since we now consider that we should rely on
the response body we're about to write rather than any previously set
value.
Issue: SPR-17502
(Cherry-picked from 3203d39821)
Review and update Servlet and Undertow adapters to release any data
buffers they be holding on to at the time of error or cancellation.
Also remove onDiscard hooks from Reactor and Undertow request body.
For Reactor we expect it to be handled. For Undertow there isn't
any Reactor Core upstream for the callback to be useful.
Issue: SPR-17410
Prior to this commit, an bug introduced in SPR-16949 prevented
`Mono.empty` bodies from being written to the response.
This commit ensures that empty bodies still trigger the writing to the
response and does not hang the processing of the exchange.
Issue: SPR-17220
Cherry-picked from: 280da61d5c
Empty Maps are preferably initialized without capacity (not initializing them at all or lazily initializing with default capacity when needed).
Issue: SPR-17105
(cherry picked from commit 4a147d26fc)
Since SPR-15205, the `FormHttpMessageConverter` is adding a `charset`
directive to the `Content-Type` request header in order to help servers
understand which charset is being used to encode headers of each part.
As reported in SPR-17030 and others, some servers are not parsing
properly such header values and assume that `boundary` is the last
directive in the `Content-Type` header.
This commit reorders the charset information right before the boundary
declaration to get around those issues.
Issue: SPR-17030
(Cherry-picked from 390bb871d8)
See Javadoc on UriComponentsBuilder#uriVariables for details.
This helps to prepare for SPR-17027 where the MvcUriComponentsBuilder
already does a partial expand but was forced to build UriComonents
and then create a new UriComponentsBuilder from it to continue. This
change makes it possible to stay with the same builder instance.
Issue: SPR-17027
After the latest changes, two small fixes in the clone method to copy
the encode flag, and in the encodeUriTemplate method to account for
possible null query params.
Improvements in the URI encoding section.
Issue: SPR-17039, SPR-17027
LinkedList remains in place where a List is likely to remain empty or single-element (in order to avoid unused capacity).
Issue: SPR-17037
(cherry picked from commit 9c08a48)
Juergen Hoeller
Rossen Stoyanchev
Brian Clozel
Kazuhiro Sera