[COMMONSXML-11] Restructure the build matrix around the axes that matter - #24
Merged
Conversation
Add a `native-xalan` profile that builds the tests into a native image, and a GraalVM job to the CI matrix that activates it. This exercises a code path the JVM matrix cannot reach: native-image resolves the JAXP providers at build time under the closed-world assumption, rather than at run time through ServiceLoader. A native binary carries one classpath, so only `test-stockjdk` runs; the other JAXP combinations stay on the JVM matrix. Saxon-HE is dropped, since its JAXP entry points carry no reachability metadata and every ServiceLoader lookup would fail on a missing no-arg constructor. Its `xpath3` group is already excluded from that execution. Apache Xalan supplies TrAX. The stock JDK's XSLTC compiles each stylesheet into translet bytecode and defines the class at run time, which a closed-world image cannot do: with XSLTC the transform tests fail on `No classes have been predefined during the image build`. Predefining them does not help either, even though JDK-8274535 made the generated bytecode deterministic; the image builder rejects the class with `Cannot define class die.verwandlung.fixture`. Interpretive Xalan, the workaround named on that same issue, needs no bytecode at run time. Xalan and its serializer are declared as project dependencies rather than through surefire's additionalClasspathDependencies, which native-image never sees. Reachability metadata comes from the tracing agent rather than a committed config: Xalan reflects on its template setters and reads serializer defaults from property files, none of which static analysis sees, and a hand-written config would rot against every Xalan and JDK update. The attack fixtures are embedded explicitly, since the agent records only the resources a given run happened to open. Assisted-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
java.xml is pure Java, so the OS and JDK axes barely interact. The one OS-specific behavior this library touches is file URI / systemId normalization on Windows, which any single Windows job exercises. Crossing three OSes with two JDK endpoints therefore spent jobs on combinations that cannot differ, while leaving the axes that do differ underexplored. Run the JDK majors on Windows, which covers systemId normalization at every version for free, and keep one Linux canary on the current LTS: a case-sensitive filesystem, and what contributors and downstream CI run. Add distribution variance, one job each. Semeru 8 and 21 cover OpenJ9, whose class library is the most divergent runtime setup-java offers and the most plausible source of surprises in factory lookup and class loading. Zulu is pinned to an old 8 patch level, verifying the hardening degrades gracefully on a runtime predating the later jdk.xml.* backports. Drop macOS: its filesystem is case-insensitive like Windows and POSIX-pathed like Linux, so it is strictly interior to the other two. Vendors that track the OpenJDK Updates projects closely are interchangeable with Temurin here. Narrow the GraalVM job to the `test` lifecycle; native-image is slow and the static analysis goals are already covered by the JVM jobs. Twelve JVM jobs plus the two Android legs, against twenty-one for a full 3x7 matrix, with strictly better coverage of the axes that produce different results. Assisted-By: Claude Opus 4.8 <noreply@anthropic.com>
Member
Author
|
Zulu is pinned to a version that does not enforce |
ppkarwasz
marked this pull request as draft
July 9, 2026 21:03
Member
|
GraalVM should be marked as experimental. We make no effort in Commons towards accommodation for any specific JVM. |
Member
Author
|
GraalVM is marked as experimental in #23 on which PR this is stacked. |
Member
Author
|
After applying #21, the 8.0.201 build is not failing any more. This means the |
ppkarwasz
marked this pull request as ready for review
July 31, 2026 19:46
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
Stacked on #23. The base branch is
feature/ci-matrix, so this PR shows only its own commit. Please merge #23 first; GitHub will then retarget this one atmainautomatically.java.xmlis pure Java, so the OS and JDK axes barely interact. The one OS-specific behavior this library touches is file URI / systemId normalization on Windows, which is old, stable code that any single Windows job exercises. Crossing three OSes with two JDK endpoints therefore spent jobs on combinations that cannot differ, while leaving the axes that do differ underexplored.The new matrix
jdk.xml.*backports-Pnative-xalan(from #23)Twelve JVM jobs plus the two Android legs, against twenty-one for a full 3x7 matrix, with strictly better coverage of the axes that produce different results.
Rationale for the omissions
macOS is intentionally dropped: its filesystem is case-insensitive like Windows and POSIX-pathed like Linux, so it is strictly interior to the other two. If we want to keep the build verified for Mac-based contributors, a scheduled weekly macOS job would be enough.
Corretto, Liberica, Microsoft, SapMachine track the OpenJDK Updates projects closely; for
java.xmlpurposes they are interchangeable with Temurin.Verification
Every distribution and version in the matrix was confirmed to be published before being added:
ibmruntimes/semeru{8,21}-binaries).The suite was run locally against the JDK 8 legs, which are the ones most likely to break: Temurin 8u492 and IBM Semeru (OpenJ9) 8u412 both BUILD SUCCESS, all seven surefire executions green. The GraalVM job's exact command (
mvn ... -Pnative-xalan test) also passes locally on GraalVM 21: 123 native tests, zero failures.Two legs could not be exercised locally and are first tested by this PR's own CI run: Zulu 8.0.201 (no such JDK installed here) and Semeru 21. The Windows jobs likewise run for the first time here; they are the same code paths as the Linux ones apart from systemId normalization, which is precisely what they are there to check.
🤖 Generated with Claude Code