Use LinkedHashMaps/Sets wherever exposed to users, and code tests defensively in terms of expected Map/Set ordering. Otherwise, there'll be runtime order differences between JDK 7 and JDK 8 due to internal HashMap/Set implementation differences.
Issue: SPR-9639
* 3.2.x: (28 commits)
Hide 'doc' changes from jdiff reports
Document @Bean 'lite' mode vs @Configuration
Final preparations for 3.2.2
Remove Tiles 3 configuration method
Polishing
Extracted buildRequestAttributes template method from FrameworkServlet
Added "beforeExistingAdvisors" flag to AbstractAdvisingBeanPostProcessor
Minor refinements along the way of researching static CGLIB callbacks
Compare Kind references before checking log levels
Polish Javadoc in RequestAttributes
Fix copy-n-paste errors in NativeWebRequest
Fix issue with restoring included attributes
Add additional test for daylight savings glitch
Document context hierarchy support in the TCF
Fix test for daylight savings glitch
Make the methodParameter field of HandlerMethod final
Disable AsyncTests in spring-test-mvc
Reformat the testing chapter
Document context hierarchy support in the TCF
Document context hierarchy support in the TCF
...
Previously the methodParameter array field was initialized lazily since
it requires reflection. However, in practice the field is always used
and there is not much benefit from the lazy initialization.
In Spring Framework 3.2, the methodParameter field was copied when a
new HandlerMethod instance (with the resolved bean) is created for
performance reasons. That introduced a synchronization issue since
the lazy initialization was not synchronized.
Issue: SPR-10365
Check for an empty fragment in UriComponentsBuilder.fromUriString(...)
to prevent the invocation of fragment(...).
Previously, UriComponentsBuilder.fromUriString(...) threw an exception
in the case of an empty fragment being provided (e.g. /example#).
Issue: SPR-10363
Deprecate HttpStatus.MOVED_TEMPORARILY in favor of HttpStatus.FOUND
since HttpStatus.valueOf(302) will always return FOUND.
The name of the 302 status code was changed between HTTP 1.0 and
HTTP 1.1 but the underlying meaning remains the same.
Issue: SPR-10300
Before this change the presence of path params (e.g. "/foo;q=1/bar")
expected the request mapping to contain a URI variable in the place of
semicolon content (e.g. either "/{foo}/bar" or "/{foo};{fooParams}").
The change ensures path params are ignored in @RequestMapping patterns
so that "/foo/bar" matches to "/foo;q=1/bar" as well as
"/foo;q=1;p=2/bar".
Along with this change, the RequestMappingHandlerMapping no longer
defaults to having semicolon content removed from the URL, which means
@MatrixVariable is supported by default without the need for any
further configuration.
Issue: SPR-10234
Add setOutputStreaming on SimpleClientHttpRequestFactory to allow the
disabling of 'output streaming' mode on the underlying connection so
that authentication and redirection can be handled automatically.
Issue: SPR-9617
Prior to this commit several HTTP classes made use of FileCopyUtils
when reading from or writing to streams. This has the unfortunate
side effect of closing streams that should really be left open.
The problem is particularly noticeable when dealing with a
FormHttpMessageConverter that is writing a multi-part response.
Relevant HTTP classes have now been refactored to make use of a new
StreamUtils class that works in a similar way FileCopyUtils but does
not close streams.
The NonClosingOutputStream class from SimpleStreamingClientHttpRequest
has also been refactored to a StreamUtils method.
Issue: SPR-10095
Update HierarchicalUriComponents.toUri() to only prepend a missing '/'
when the scheme, user info, host or port are specified. This makes
the toUri() method behave in the same way as .toUriString() and allows
relative URIs to be created.
Issue: SPR-10231
ConfigurableWebEnvironment was introduced in 3.2.0.M1 with SPR-9439 in
order to break a cyclic dependency. At the same time, certain signatures
such as AbstractRefreshableWebApplicationContext#getEnviroment and
GenericWebApplicationContext#getEnvironment were updated to take
advantage of covariant return types and return this newer, more narrow
type and providing cast-free calls to ConfigurableWebEnvironment methods
where necessary. Similar changes were made to HttpServletBean in
3.2.0.M2 with SPR-9763.
Narrowing #getEnvironment signatures in this fashion required enforcing
at the #setEnvironment level that any Environment instance provided
(explicitly or via the EnvironmentAware callback) must be an instance of
ConfigurableWebEnvironment. This is a reasonable assertion in typical
web application scenarios, but as SPR-10138 demonstrates, there are
valid use cases in which one may want or need to inject a non-web
ConfigurableEnvironment variant, e.g. during automated unit/integration
testing.
On review, it was never strictly necessary to narrow #getEnvironment
signatures, although doing so did provided convenience and type safety.
In order to maintain as flexible and backward-compatible an arrangement
as possible, this commit relaxes these #getEnvironment signatures back
to their original, pre-3.2 state. Namely, they now return
ConfigurableEnvironment as opposed to ConfigurableWebEnvironment, and in
accordance, all instanceof assertions have been removed or relaxed to
ensure that injected Environment instances are of type
ConfigurableEnvironment.
These changes have been verified against David Winterfeldt's Spring by
Example spring-rest-services project, as described at SPR-10138.
Issue: SPR-10138, SPR-9763, SPR-9439
