Correctness regression on x86 architecture

[deenp03]

Overview of the Issue

After commit 74afbcecf5e15244c1eca605a2a2c5dc95c41f0a, hyperscan-java switched from Hyperscan 5.4.0 to Vectorscan 5.4.11 which means that hyperscan-java now supports ARM. However, a critical x86 bug was recently discovered in Vectorscan 5.4.11: VectorCamp/vectorscan#317.

This bug prevents us from using hyperscan-java version 5.4.11-3.0.0 because it's giving us incorrect results on x86.

Impact

We had to revert to 5.4.0-2.0.0 in order to get correct results on our x86 machines in production, but this comes with several issues:

• We can no longer run tests on our ARM developer laptops
• We are missing several hyperscan-java bug fixes including: Fix a race condition in Scanner (closes #158) #159 Fix Database and Scratch instances not reclaimable by GC (closes #161) #162

Proposed Solution

1. While we wait for a Vectorscan fix/release: cut a new alternative release of hyperscan-java using the latest hyperscan release which works on x86. This release would include hyperscan-java's other latest changes including: Fix a race condition in Scanner (closes #158) #159 Fix Database and Scratch instances not reclaimable by GC (closes #161) #162. This alternative release could use a label (e.g. "-x86").
2. After a Vectorscan fix/release: cut a new release of hyperscan-java using the latest Vectorscan release.

[gliwka]

Thanks for finding this bug.

I will cut a release candidate build based on hyperscan-java main and the latest vectorscan release with the patch applied.

Once vectorscan has a new release upstream, I will publish a proper release.