Bump protobuf bazel_dep 33.4 -> 33.6 (refs #339)

[tinder-maxwellelliott]

Summary

Refs #339. Bumps the bzlmod protobuf dep from 33.4 to 33.6, the latest stable release on the 33.x line. The Java runtime used by the deploy jar's java_proto_library targets ([@protobuf]) tracks the module version directly, so this is the single source of truth for what ships inside bazel-diff.jar — there is no separate Maven dep to update (we don't pull com.google.protobuf:protobuf-java from maven_install.json).

Why not 34.x

I first pushed this PR targeting protobuf 34.1 (latest stable on BCR). CI's test-jre11-run-example matrix runs against Bazel 7.x and failed analysis with:

Error: _create_proto_info() got unexpected keyword arguments:
  option_deps, extension_declarations

protobuf 34.0 began passing option_deps and extension_declarations to ProtoInfo(...), and Bazel 7.x's builtin ProtoInfo provider does not accept those. That's a hard incompatibility — 34.x requires Bazel 8+. Staying on the 33.x line preserves the existing Bazel 7.x CI matrix; a follow-up PR that drops 7.x from CI can move us to 34.x (and then to whatever lands the Unsafe migration).

Honest disclosure: does this fix the Unsafe warning?

No, not on its own. I verified by grepping the bytes of the freshly-built bazel-bin/cli/bazel-diff_deploy.jar:

$ unzip -p bazel-bin/cli/bazel-diff_deploy.jar com/google/protobuf/UnsafeUtil.class | grep -ao 'arrayBaseOffset\|sun/misc/Unsafe' | sort -u
arrayBaseOffset
sun/misc/Unsafe

protobuf v33.6's UnsafeUtil$MemoryAccessor.arrayBaseOffset still calls sun.misc.Unsafe.arrayBaseOffset(Class) directly — the same call site producing the JDK 24+ "terminally deprecated method" warning in #339. I also checked the protobuf main branch source: the call is still there, unchanged. v34.1 has it too.

The migration off sun.misc.Unsafe (likely to VarHandle) is tracked upstream at protocolbuffers/protobuf#20760. It has not landed at any released tag yet.

Bumping to 33.6 keeps us on the latest 33.x release so we'll pick up smaller fixes in this line. We'll pick up the Unsafe migration in a subsequent bump once it ships upstream and we move off Bazel 7.x.

I'd recommend leaving #339 open until upstream lands the fix and we bump again, rather than closing on this PR.

Verification

• bazel build //cli:bazel-diff_deploy.jar — succeeds with protobuf 33.6.
• bazel test over the 15 cli unit-test targets — all pass:
  • BazelClientTest, BazelModServiceTest, BazelRuleTest, BuildGraphHasherTest
  • CalculateImpactedTargetsInteractorTest, CalculateImpactedTargetsInteractorIssue335Test, ContentHashProviderTest
  • DeserialiseHashesInteractorTest, ModuleGraphParserTest
  • NormalisingPathConverterTest, OptionsConverterTest
  • ProcessStdinHangRegressionTest, SourceFileHasherTest
  • StderrPollutionRegressionTest, TargetHashTest
• E2E tests not run locally due to a pre-existing JDK-env sandbox issue on this machine (Unable to locate a Java Runtime inside the nested bazel server, present on master without these changes). CI will run them.

Test plan

• bazel build //cli:bazel-diff_deploy.jar passes.
• All 15 cli unit-test targets pass under bazel test.
• First attempt at 34.1 failed CI on Bazel 7.x; downgraded target to 33.6 and pushed.
• CI green across the full matrix.
• Once protocolbuffers/protobuf#20760 ships a fix, follow-up PR to bump (likely past 34.x, requires dropping Bazel 7.x first) and verify the warning is gone.

🤖 Generated with Claude Code