[bernardladenthin]

Summary

• spotbugs:check was red on main after Write the calibration paste block with LF line endings on every platform #236. VA_FORMAT_STRING_USES_NEWLINE fires on every String.format pattern containing \n and asks for %n — the platform separator the calibration paste block must not carry, so the rule and the requirement conflict. The newline leaves the pattern instead: a new appendXmlNumber helper appends it as a literal (the shape appendJsonNumber/appendYamlNumber already use, for the same documented reason) and the numbers are formatted at the call site through the FORMAT_* constants, which until now only the JSON/YAML renderers used although their javadoc claimed otherwise. No suppression, rendered text unchanged.
• The srcmorph-selftest profile names its model directory once. All 38 <modelPath> entries carried the absolute X:/Modelle/…, a drive that no longer exists on the owner's machine — so the profile ran nowhere as committed, and a machine-specific path shipped in a published POM 38 times. They read ${ai.modelDir}/<file>.gguf now; another machine uses -Dai.modelDir=/path/to/models.
• Two documentation statements that stopped matching jllama 6.0.0: the gpuLayers = -1 comment ("CPU build stays CPU; GPU build uses its own default" — there is no such split since the natives-jar restructure), and TODO.md's "GPU classifier fat jars … never launched", which now records that cuda13-windows-x86-64 has had a real launch pass.
• Four dead constants and one concatenated literal removed from LlamaCppJniAiGenerationProviderTest (six Error Prone warnings). Their assertions moved to the model-free LlamaCppJniProviderSupportTest with the core-module extraction, so no coverage is lost.

Test plan

• Affected unit / integration tests pass locally — full reactor clean verify BUILD SUCCESS on all three modules, 659 + 39 + 32 tests, 0 failures, spotbugs:check and spotless:check clean. Run against a resolvable llama 6.0.0-SNAPSHOT from the Central snapshot repository in an isolated local repository, which is the only way to get past the pin described below.
• -P srcmorph-selftest interpolation verified with help:effective-pom: the default resolves every path to M:/Modelle/<file>.gguf, -Dai.modelDir=<dir> redirects all 38.
• The gpuLayers claim is measured, not assumed — RTX 3070 Laptop, Qwen3-4B-Instruct-2507-Q4_K_M, identical config, only the classpath differing: 63 t/s prefill / 7.1 t/s decode with ggml registering [RPC, CPU], and 3474 / 86.8 with [CUDA, RPC, CPU] once the cuda13 module jar is present. gpuLayers −1 and 99 within noise in both cases.
• The cuda13 fat jar (-P gpu-cuda, 258 MB) ran a real GenerateFileIndex over the production prompt set through the CLI and through srcmorph:generate.
• CI will be red on this branch, and not because of it. The reactor still dies at the first Maven step on the unresolvable net.ladenthin:llama:6.0.0 (TODO.md, first open item) — which is also why no CI run could ever have shown the SpotBugs failure this PR fixes: nothing reached SpotBugs. Expect green only once jllama 6.0.0 is on Central.
• Docs / CHANGELOG updated

Related issues / PRs

Refs #236 — the first commit here fixes a SpotBugs violation that PR introduced and CI could not see.

Checklist

• I have read CONTRIBUTING.md and CODE_OF_CONDUCT.md
• My commits follow Conventional Commits — they do not, deliberately: the commits on main use plain imperative subjects with a detailed body (e.g. "Write the calibration paste block with LF line endings on every platform"), and these match that. CONTRIBUTING.md documents Conventional Commits, so the written convention and the repository's practice disagree; worth settling one way or the other, and happy to rewrite these subjects if the document is the one that should win.
• No security-sensitive changes

🤖 Generated with Claude Code

https://claude.ai/code/session_01E2h8gXyyE5UeimkL9vQv9G

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 93e92447-99fa-4a4e-a275-4a7468021996

📥 Commits

Reviewing files that changed from the base of the PR and between d334ebe and 09a71b0.

📒 Files selected for processing: 7

• CHANGELOG.md
• TODO.md
• docs/ai-index-benchmark/tools/calibrate-bench-pom.xml
• srcmorph-maven-plugin/pom.xml
• srcmorph/src/main/java/net/ladenthin/srcmorph/engine/CalibrationReport.java
• srcmorph/src/main/java/net/ladenthin/srcmorph/provider/LlamaCppJniAiGenerationProvider.java
• srcmorph/src/test/java/net/ladenthin/srcmorph/provider/LlamaCppJniAiGenerationProviderTest.java

+-----------------------------------------+
| Making HAL 9000 open the pod bay doors. |
+-----------------------------------------+
  \
   \     /)/)
        ( '')
        /  |
      o(___)

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}