Previously, when the liquibase.url, .username, and .password
properties were used to configure a DataSource specifically for
Liquibase that DataSource would never be explicitly closed. As it is
created by DataSourceBuilder with no explicitly configured type it
will use whichever connection pool is available and, therefore, will
create and keep open the pool's minimum number of connections. This
is an unnecessary use of resources both in the application and in the
database.
This commit updates LiquibaseAutoConfiguration so that if it uses
DataSourceBuilder to create a DataSource then it will also close that
DataSource once the database has been migrated.
Closes gh-9218
Move `spring.rabbitmq.listener.*` to `spring.rabbitmq.listener.simple.*`
in preparation for Spring AMQP 2.0 that supports different container
types.
Closes gh-9108
See gh-9055
This commit ensures that a primary JSR 303 and Spring Validator will be
exposed if the auto-configuration kicks in. As `LocalValidatorFactoryBean`
exposes 3 contracts (JSR-303 `Validator` and `ValidatorFactory` as well as
the `Spring` validator one), this makes sure that those types can be
injected by type.
`LocalValidatorFactoryBean` exposes 3 contracts and we're only checking
for the absence of a `javax.validation.Validator` to auto-configure a
`LocalValidatorFactoryBean`. If no standard JSR validator exists but a
Spring's `Validator` exists and is primary, we shouldn't flag the
auto-configured one as `@Primary`. Previous iterations on this feature
have made sure that we'll auto-configure at most one
`javax.validation.Validator` so not flagging it `@Primary` is no problem.
This commit also restores and adds tests that validates
`ValidationAutoConfiguration` will configure a JSR validator even if a
Spring Validator is present.
This effectively fixes gh-8495 in a different way.
Closes gh-8979
Closes gh-8976
`BasicErrorControllerIntegrationTest` is handling the context in each test
and yet it starts a useless test via `@SpringBootTest`. This commit
removes the useless runner and associated annotations.
Replace the MVC validator post processor with an `@Import` so
that a "cannot enhance @Configuration bean definition"
warning does not occur.
Fixes gh-8951
See gh-8495
Update `PropertiesLauncher` so that classes can be loaded outside of
`BOOT-INF/classes`. You can use a subdirectory, or the root directory
of an external jar (but not the parent archive to avoid issues
with agents and awkward delegation models).
Fixes gh-8480
Closes gh-8486
Relax `SpringBootTestContextBootstrapper` rules so that a test can
specify an `ApplicationContextInitializer` and still have
`@SpringBootConfiguration` detected.
Prior to this commit detection would not occur because it's possible
that an initializer _could_ register configuration. This scenario is
actually quite unlikely to occur, certainly less likely than wanting to
use an initializer in combination with auto-detection.
Fixes gh-8483
Update `ValidationAutoConfiguration` and `WebMvcAutoConfiguration` to
ensure as much as possible that only a single Validator bean of each
type is registered.
Validation auto-configuration now does the following:
- If no validator is found: Registers a `LocalValidatorFactoryBean`
(providing both Spring and JSR validation)
- If the user defines a Spring & JSR validator: Backs off
- If the user defines only a JSR validator: Adapts it to a Spring
validator (without exposing another JSR implementation)
WebMvcAutoConfiguration auto-configuration has been updated to make
MVC validation follow common Spring Boot patterns:
- If not validator beans are found (due to the user excluding
ValidationAutoConfiguration) a new `mvcValidator` bean will be
registered.
- If a single validator bean is found it will be used for MVC
validation.
- If multiple validator beans are defined it will either use the one
named `mvcValidator` or it will register a new `mvcValidator` bean
Any automatically registered `mvcValidator` bean will not implement
the JSR validator interface.
Finally, it is no longer possible to provide an MVC validator via a
`WebMvcConfigurer`.
Fixes gh-8495
Allow `PathExtensionContentNegotiationStrategy` to be bypassed by
actuator endpoints. Prior to this commit calling `/loggers/com.aaa.cab`
would return a HTTP 406 response due to `.cab` being a known extension.
Fixes gh-8765
Update `ResourceServerTokenServicesConfiguration` to fail fast if the
`getKeyFromServer()` call fails. Since the key is part of the singleton
`JwtAccessTokenConverter` bean there is not real way to refresh without
restarting the application.
A hard failure seems preferable to an inconsistent state.
Closes gh-8924
This commit makes sure to honour the `spring.aop.proxy-target-class`
property if set by the user. Previously, the
`PersistenceExceptionTranslationPostProcessor` was always configured to
use cglib, regardless of the value of that property.
Closes gh-8887
This commit makes sure to honour the `spring.aop.proxy-target-class`
property if set by the user. Previously, the
`MethodValidationPostProcessor` was always configured to use cglib,
regardless of the value of that property.
Closes gh-8869
Previously, as a result of the changes made in de50cfa21e, an
application would fail to start on Google AppEngine as it prevents
the creation of new threads.
This commit updates OnClassCondition so that it falls back to
performing the work on the main thread when its unable to shift it
to a separate thread.
Closes gh-8584
When used as a meta-annotation the value() attribute of
@ConditionalOnClass will fail silently resulting in the @Conditional
nature of the annotation being ignored.
See gh-8185
Previously, WelcomePageHandlerMapping would forward to index.html.
This assumed that the static path pattern was always /**. If it had
been customised to, for example, /foo/**, then the forward would still
be to index.html and a 404 would result as the page is actually
available at /foo/index.html.
At first glance, it would appear that the forward should be made to
foo/index.html. However, as it's a forward rather than a redirect,
any relative URLs in the index.html page would then be resolved using
/ whereas they should be resolved using /foo/. This could be addressed
by using a redirect rather than a forward, but we don't want to do
that as it's more invasive and would require a roundtrip back to the
client. Instead, this commit simply stops performing the forward when
the static path pattern is not /**.
Closes gh-8788
Phillip Webb
Andy Wilkinson
Stephane Nicoll
Spring Buildmaster
Dave Syer
Walter Seymore
Phillip Verheyden
Dr. Stefan Hüttemann