(don't merge now) Fix CI tests for OpenJDK 21 #305
Conversation
Wenyu says we don't do class unloading. This is fine.
OpenJDK no longer reserves 4 words for every frame, but instead let each GC barrier set decide the reserved size. We reserve 3 words for MMTk's C1 barrier set. See: openjdk/jdk@36a0822
|
The good thing is that most test cases passed. There is a Clippy warning about We should also check if the CI scripts in mmtk-core, especially minimal test, work. |
|
There are failed test cases that involve DaCapo Chopin benchmarks. Because we already saw them failing on the master branch, I will not attempt to fix them for now. But we should switch from the current DaCapo Chopin 23.9 RC3 to DaCapo Chopin 23.11 MR2 because there may be problems in DaCapo, too. |
wks
force-pushed
the
fix/ci-jdk21
branch
6 times, most recently
from
15730df to
f2b9a1a
Compare
This reverts commit 6086363.
This follows the upstream change
This changes paths build options, etc., to make the CI scripts work with OpenJDK 21. We now use a single source of truth for the OpenJDK path instead of defining OPENJDK_PATH in multiple scripts.
Use jdk21 workarounds for Cassandra and H2O
Point to the newer version of dacapo
We just use the name defined in running-ng, i.e. "dacapochopin". This allows us to use modifiers such as "dacapochopin_jdk21" which expects the benchmark suite name "dacapochopin".
|
Closed in favor for #307 |
This PR is for testing the CI scripts for OpenJDK 21. Don't merge it now.