Update to Resolver alpha-5 and apply some cleanups.
Notable changes:
* update resolver to 2.0.0-alpha-5
* detach model-builder from maven-artifact (this is important)
* introduce model builder own VersionParser iface (implemented in resolver-provider)
* API VersionParser implementation reuses VersionParser from resolver-provider to implement the service
* various other cleanups, removal of old plexus, etc
---
https://issues.apache.org/jira/browse/MNG-7974
* [MNG-7963] Clean up the build
Changes:
* remove hamcrest 1.3 from scope
* set proc:none to prevent warnings (Sisu APT is on classpath),
we do not use annotation processing, we produce Sisu index
explicitly.
* remove some redundant elements from POMs
Remove all ordering from Maven and delegate it to Resolver.
Supersedes https://github.com/apache/maven/pull/1212
This PR makes Maven completely rely on Resolver to order artifacts on classpath (before this PR it was several spots where explicit ordering was applied without ability to affect it). This PR makes now Maven completely rely on Resolver, never "sort" artifacts on it's own, but use resolver result, that in turn depends on session config.
Example invocation:
```
$ mvn -X package -Daether.system.dependencyVisitor=levelOrder
```
Would make Maven use "levelOrder".
Example of default (preOrder, as before) and new levelOlder (new in 2.0): notice how tree is same, but resulting artifact list (and hence, order on classpath) is different:
https://gist.github.com/cstamas/c88ca550006e53edad5b16973b54c349
The new resolver configuration is documented here:
https://maven.apache.org/resolver-archives/resolver-2.0.0-alpha-3/configuration.html
(property No 62)
---
https://issues.apache.org/jira/browse/MNG-7953
* Add an InternalSession interface to avoid casting to the implementation in various places
* Upgrade to Guice 6.0.0 and add support for jakarta.inject annotations
* Maven 4 Plugin API
Integrate better and obtain better control over Resolver. These changes did stem from "JPMS module experiment" and are considered improvement but does not implement any functionality related to JPMS module support.
Changes:
* Maven4 should stop "disconnected coexistence" of two type systems (ArtifactHandlers and Resolver ArtifactTypeRegistry), it should unify them.
* Maven4 Core should provide generic and extensible means to introduce new artifact types (fully in extension, and extension should get extended data via "roundtrip" in core/resolver)
---
https://issues.apache.org/jira/browse/MNG-7924
Upgrade to Resolver 2.0.0.
Changes:
* many UT code used `new DefaultRepositorySystem()` ctor that is gone (was present due SL only), replaced with mocks
* dropped MavenResolverModule Guice module (as AetherGuice module is gone as well)
* updated Resolver version to 2.0.0-alpha-1
* added jdk transport (that prevails apache on Java 11)
* rename of "native" into "apache", deprecate "native" name
* introduce "jdk" transport
---
https://issues.apache.org/jira/browse/MNG-7909
Multiple optimizations :
- renderLevel() method use static constants instead of rebuilding the strings on each call
- replace + operator usage with more PrintStream.print() calls to reduce temporary strings creation
- reduce usage of MessageBuilder.a() method usage with more PrintStream.print() calls to reduce temporary strings creation
- replace the builder() method with a static import
- replace String.format with a simple string concat to reduce memory allocation (garbage)
- change static constants to class member in MavenSimpleLogger
- add a setLength(int) method in MessageBuilder to enable reuse
The purpose of the BOM packaging is to more cleanly identity POM that will be consumed as BOM from POM that will be used as parents. The BOMs can be turned into a consumer POM, whereas parents can not loose any information.
(that are not applicable, but look ugly).
These tests are NOT built, but are taken from real projects
several years ago, and used in POM related UTs.
Hence, for example update of velocity is completely okay,
even if in "real life" it would not compile.
This is most probably "just" about broken Throwable implementations.
Some override getCause but does not perform checks to what it was inited
as "this" means "not yet inited" actually.
---
https://issues.apache.org/jira/browse/MNG-7846
Changes:
* always parse it at session start
* hence, will WARN if needed there as well, and will warn once
* do not re-parse and possibly warn always, reuse parsed enum
---
https://issues.apache.org/jira/browse/MNG-7823
Given maven-shared-utils is being deprecated (see MSHARED-1269), we need to remove its usage in maven. Maven core only uses the logging/message related classes from maven-shared-utils, so this PR enhance the initial v4 api with the needed interfaces and move the jansi specific code into maven-embedder.
* Move maven-compat to jsr330
* Remove more ties to RepositorySystem
* Move more stuff to maven-compat
* Move PluginManager and LocalRepositoryNotAccessibleException to maven-compat
* Fix tests
* Break circular dependency loop in components
* Move unused deprecated classes
Checks the paths before relativizing them.
Normalize and relative before adding to result
Rename local vars
Apply to ExecutionEventLogger
---
https://issues.apache.org/jira/browse/MNG-7795
Co-authored-by: Andreas Dangel <adangel@apache.org>
Refactored and now dropped the unused pluginDescriptor/dependencies validation. Now we have "plugin dependency" (using POM) components to perform dependency validation.
Also, two "inlined" validation from plugin dependencies resolver factored out to components, along with 3 existing moved to newly added (they were wrongly validating descriptor dependencies while intent was to validate "real" dependencies) validators.
Ported from
6c562a46ee
---
https://issues.apache.org/jira/browse/MNG-7789
Reword the validation warning and add new check for real plugin dependencies in wrong scopes (do not rely on build-time derived descriptor, but on real data instead).
---
https://issues.apache.org/jira/browse/MNG-7786
* [MNG-7743] Make the build work on JDK 20
* the behaviour before the fix was already pretty confusing. JDK 19 and
older do not check the presense of '{' in the constructor, so the
URL object got created, but when converting to file the result would
be e.g. '/../../src/test/remote-repo' which is completely wrong.
But it seems the affected tests did not really care, as all of them
were passing
* Remove forgotten println
Co-authored-by: Yeikel <email@yeikel.com>
* Test with latest JDK
* Do not run ITs with jdk 20, but just the "build itself"
---------
Co-authored-by: Yeikel <email@yeikel.com>
Co-authored-by: Guillaume Nodet <gnodet@gmail.com>
This commit introduces three properties:
* project.rootDirectory: the project's directory or parent directory containing a .mvn subdirectory or a pom.xml flagged with the root="true" attribute. If no such directory can be found, accessing the rootDirectory property will throw an IllegalStateException.
* session.topDirectory : the directory of the topmost project being built, usually the current directory or the directory pointed at by the -f/--file command line argument. The topDirectory is similar to the executionRootDirectory property available on the session, but renamed to make it coherent with the new rootDirectory and to avoid using root in its name. The topDirectory property is computed by the CLI as the directory pointed at by the -f/--file command line argument, or the current directory if there's no such argument.
* session.rootDirectory : the rootDirectory for the topDirectory project.
The topDirectory and rootDirectory properties are made available on the MavenSession / Session and deprecate the executionRootDirectory and multiModuleProjectDirectory properties. The rootDirectory should never change for a given project and is thus made available for profile activation and model interpolation (without the project. prefix, similar to basedir). The goal is also to make the rootDirectory property also available during command line arguments interpolation.
A root boolean attribute is also added to the model to indicate that the project is the root project. This attribute is only supported if the buildconsumer feature is active and removed before the pom is installed or deployed. It can be used as an alternative mechanism to the .mvn directory.
This is the port of https://github.com/apache/maven/pull/1075 to the master branch.
The good thing is that this is ONLY the tests because the maven 4 does not have the problem of MNG-7750.
A few small changes in the tests were needed to handle the internal differences between maven 3.x and maven 4.x.
---
https://issues.apache.org/jira/browse/MNG-7750
This is general rework of current Maven 3.9.x line how it handles plugin and mojo validations.
Changes:
* added plugin validations for dependencies
* introduced pluginValidationManager that gathers violations
* manager creates a report at build end, with dense and non repeating data
* this is in spirit to lessen already too verbose logging, as current solution would report violations as many times plugin is used in reactor, and that can be many (ie. a plugin from parent for each module)
Example report of Maven 3.9.x build: https://gist.github.com/cstamas/b62fdcd53eaf316123cf183f5a24e6a5
---
https://issues.apache.org/jira/browse/MNG-7754
* [MNG-6825] Replace FileUtils with Commons IO
Co-authored-by: Moderne <team@moderne.io>
* Minor fixes and reverts where cases are not handled yet
* Drop explicit dependencies on plexus-utils
---------
Co-authored-by: Moderne <team@moderne.io>
The original fix MNG-7672 matched the "scope" but missed the "order". `project.collectedProjects` are in order as loaded (POM order), is not topologically sorted.
Fix is to use DAG of projects, ask for downstream projects (will return more then we need but sorted) and narrow that list to only contain collected projects.
---
https://issues.apache.org/jira/browse/MNG-7720
If ConcurrencyDependencyGraph#getRootSchedulableBuilds returns an empty
list then MultiThreadedBuilder is locked forever as it never gets a
build result (because nothing is scheduled).
This changes the method, that in such case just the first project is
returned, this might not give the best performance, but ensures that
there is at least one build scheduled and the build-loop can proceed.
This closes#1029
There is really no need for it, and there is a resolver option if really must (but really should not, local repo is and should be considered as transient).
---
https://issues.apache.org/jira/browse/MNG-7713
@Deprecated annotatio is helpful to prevent usage of deprecated API.
For example: developer can see crossed out method/field name if this item
is deprecated.
This closes#985
Currently, if a project is returned from the workspace reader, but is not part
of the reactor, a NPE occurs. In this case, the repositories are read from
the model instead.
This closes#981
Guillaume Nodet
Tamas Cservenak
huazhongming
Slawomir Jaranowski
Karl Heinz Marbaise
sebastien-doyon
Hervé Boutemy
Henning Schmiedehausen
huazhongming
Petr Široký
Tim te Beek
Giovanni van der Schelde
Niels Basjes
Tim te Beek
Elliotte Rusty Harold
Andrey Bruykhov
Christoph Läubrich
Alexey Venderov