This commit introduces null-safety checks for spring-core at build-time
in order to validate the consistency of Spring null-safety annotations
and generate errors when inconsistencies are detected during a build
(similar to what is done with Checkstyle).
In order to make that possible, this commit also introduces a new
org.springframework.lang.Contract annotation inspired from
org.jetbrains.annotations.Contract, which allows to specify semantics
of methods like Assert#notNull in order to prevent artificial
additional null checks in Spring Framework code base.
This commit only checks org.springframework.core package, follow-up
commits will also extend the analysis to other modules, after related
null-safety refinements.
See gh-32475
The me.champeau.mrjar Gradle plugin causes non-existent libraries to be
added to the Eclipse classpath, such as:
<classpathentry kind="lib" path="/workspaces/spring-framework/spring-core/build/classes/kotlin/java21">
<attributes>
<attribute name="gradle_used_by_scope" value="java21"/>
<attribute name="test" value="true"/>
</attributes>
</classpathentry>
This results in build errors in Eclipse IDE like the following.
Project 'spring-core' is missing required library:
'/workspaces/spring-framework/spring-core/build/classes/kotlin/java21'
This commit filters those and removes them.
This commit improves how the build deals with javadoc invalid references
in two ways.
Link/see references that are temporarily invalid during javadoc
generation of individual modules are better masked by using the option
`Xdoclint:syntax` instead of `Xdoclint:none` (warnings were still
visible in some cases, e.g. when individually building the javadoc for
a specific module).
Global javadoc-building task `api` now combines `syntax` and `reference`
`Xdoclint` groups, allowing to raise truly invalid references even when
all the modules have been aggregated.
This commit also fixes the 20+ errors which appeared following the later
change in doclet configuration.
Closes gh-30428
This commit removes the `-PmainToolchain` option from our build, since
it was not broadly used. Instead, the language level is now configured
in the `JavaConventions` for JDK 17.
The `-PtestToolchain` option is still available for testing Spring
Framework with other JDKs (i.e., compiling and running tests with a JDK
that's not the baseline).
See gh-30339
After setting up a JDK 20 CI build pipeline, numerous tests involving
date/time parsing and formatting began to fail [0]. (The failing JPA
tests are not specific to JDK 20.)
For example, we encounter visually confusing assertion failures such as
the following.
org.opentest4j.AssertionFailedError:
expected: "12:00 PM"
but was: "12:00 PM"
The expected string contains a normal space (which has always been the
case prior to JDK 20); whereas, the actual string now contains a narrow
non-breaking space.
The cause of this is mentioned in the JDK 20 Release Notes [1] as "NBSP
prefixed to a, instead of a normal space". Note, however, that the
links for the first two bullet points in that section are mixed up.
"NBSP prefixed to a, instead of a normal space" should point to [2].
Furthermore, the new whitespace character is not a non-breaking space
(NBSP) but rather a narrow non-breaking space (NNBSP). In addition, the
second bullet point should technically read "NNBSP prefixed to `a`,
instead of a normal space" -- even though `a` provides limited value to
most readers.
The downside for the Java community is that this constitutes a breaking
change for parsing and formatting date/time values that include "AM"
and "PM" units (any may potentially apply to other date/time
parsing/formatting scenarios). In Spring Framework's test suite we have
witnessed this in conjunction with Spring's @DateTimeFormat and
DateTimeFormatterFactory infrastructure as well as with Google's
Gson-to-JSON support.
A colleague who works at Oracle graciously informed me that one can use
"legacy locale data" by supplying `-Djava.locale.providers=COMPAT` as a
JVM argument, noting however that this option limits some newer
functionalities (but without enumerating which new functionalities one
might be missing when using this option).
In any case, this commit adds that JVM argument to our Gradle toolchain
builds so that our test suite passes on JDK 20, and we will continue to
investigate further options for our builds and for our users.
Note, however, that one must manually configure the
`-Djava.locale.providers=COMPAT` JVM argument when running affected
tests within an IDE.
See gh-30185
[0] https://ge.spring.io/s/kmiq2bz2afafs/tests/overview?outcome=failed
[1] https://jdk.java.net/20/release-notes#JDK-8284840
[2] https://unicode-org.atlassian.net/browse/CLDR-14032
This commit creates a new "framework-docs" module, which is dedicated to
documentation generation (reference and API docs).
This commit refactors the build configuration and moves the asciidoc
files to a separate location, but does not change the name nor the
nature of published artifacts.
Closes gh-29417
The plugin is configured to detect flaky tests and retry them 3 times
when running on the CI, but still reports failures. This will provide a
standard way to detect flaky tests as failures and successful attempts
are shown in the tests report.
Commit 9a180d1811 introduced Kotlin-based
JMH benchmarks which prevent spring-beans from being imported into the
Eclipse IDE.
This commit ensures that src/jmh/kotlin is imported in Eclipse with the
"test" attribute set.
See gh-24104
Prior to this commit, the ApiDiff Gradle task would be configured for
all submodules of the Spring Framework project and would assume that
they all existed for the baseline version considered for the API diff.
This would cause issues when:
* the sub-project is not published as it's not an official "spring-*"
module
* the "spring-*" module is new and did not exist for the baseline
version
This commit ensures that only "spring-*" modules are considered for this
task and that we trigger an early resolution of the baseline version -
if the version doesn't exist, a warn message is logged and we assume
that this is a new module, to be compared with an empty configuration.
This commit also renames the "spring-core-graalvm" project to
"graalvm-feature", since this sub-project is not an official module
published to Maven Central, but rather an internal dependency.
Fixes gh-28818
This commit leverages a subset of @philwebb initial experimentation
to compute at build time the value of specific boolean static fields
in native images. This enhancement is implemented for now as a
GraalVM feature.
The goal here is to keep an optimized footprint via build time code
removal without leveraging build-time class initialization which is known
for the blocking compatibility issues it introduces due to its viral nature.
For now, the static fields initialized at build time with native are:
- NativeDetector#imageCode
- Fields with a name ending by "Present" in "org.springframework" package
typically used for classpath check with ClassUtils#isPresent
Closes gh-28624
Kotlin currently only supports JDK17 language level, which is our
minimum requirement so far.
This commit simplifies the toolchain setup until this is needed again.
See 303e363adf
The custom stylesheet used with JDK 8 no longer works with JDK 17.
Thus in order to make the generated Javadoc usable, we are disabling
the custom stylesheet for the time being.
See gh-27496
Due to an SSL/TLS issue with hamcrest.org, the Dokka task fails with the
following.
> Failed to download package-list from https://hamcrest.org/JavaHamcrest/javadoc/2.1/package-list,
> this might suggest that remote resource is not available, module is
> empty or dokka output got corrupted
See: https://github.com/hamcrest/JavaHamcrest/issues/280
As a workaround, this commit switches to javadoc.io to reference the
Hamcrest Javadoc APIs.
Sébastien Deleuze
Stéphane Nicoll
Werner Fouché
Sam Brannen
Brian Clozel
Simon Baslé
Juergen Hoeller
Sam Brannen
Jay Bryant
Marc Wrobel
Sébastien Deleuze
xixingya