GridSearch with jobs > 1 results in JVM SIGSEGV

[vlcfaria]

Describe the bug
Calling pt.GridSearch with jobs > 1 currently results in a SIGSEGV. Including multiple JDK versions: 17, 21 and 26.

To Reproduce
Any search with jobs > 1 results in the error. Setting backend='ray' prints out an output, but the process still dies unexpectedly.

import pyterrier as pt

dataset = pt.get_dataset("vaswani")

BM25 = pt.terrier.Retriever(dataset.get_index(), wmodel="BM25", controls={"bm25.b" : 0.75})
pt.GridSearch(
    BM25,
    {BM25: {"bm25.b"  : [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1 ],
    }},
    dataset.get_topics().head(5),
    dataset.get_qrels(),
    jobs=2,
)

Expected behavior
Grid search results with optimal parameters

Documentation and Issues

• I have checked the PyTerrier documentation for relevant content
• I have checked for previous relevant PyTerrier issues

Additional context
Log file of the JVM crash is below, not sure if relevant.

hs_err_pid54708.log

[vlcfaria]

By the way, I've noticed test_gridscan_joblib2 does not suffer from the same problem? It is a flaky test that sometimes returns docid out of bounds, but the JVM itself doesn't crash from my experience... no big idea on why this happens, might be simply because JVM is in a better state with the added test overhead?

The above scripts crashes deterministically though. I switched to thread parallelism following Retriever.transform and the script + test works. #676