initial ? wtf

fpr-bench/tools/README.md

@@ -0,0 +1,55 @@
+# FPR Bench CLI
+
+Run JMH benchmarks via a single Python CLI and export JSON results.
+
+Requires Python 3.4+.
+
+## Run a benchmark
+From the repo root:
+
+```sh
+python fpr-bench/tools/fpr_bench.py run
+```
+
+This builds the shaded JMH jar and writes results to `fpr-bench/results/`.
+You will be prompted for benchmark types and common JMH settings.
+
+Available types: `throughput`, `latency`, `common` (runs the combined bench).
+
+## Pass JMH arguments
+Use `--` to pass flags directly to JMH (this skips the menu):
+
+```sh
+python fpr-bench/tools/fpr_bench.py run -- --wi 5 -i 5 -f 5 -tu us
+```
+
+## Results location
+JSON output is stored in:
+
+```
+fpr-bench/results/YYYYMMDD_HHMMSS__<tag>__<type>.json
+```
+
+If multiple types are selected, one JSON file is produced per type.
+
+## Examples
+Skip the build and use a custom tag:
+
+```sh
+python fpr-bench/tools/fpr_bench.py run --no-build --tag smoke
+```
+
+Run only latency and throughput types non-interactively:
+
+```sh
+python fpr-bench/tools/fpr_bench.py run --types latency,throughput -- --wi 3 -i 3 -f 1
+```
+
+Use custom `mvn` and `java` executables:
+
+```sh
+python fpr-bench/tools/fpr_bench.py run --mvn mvn.cmd --java java
+```
+
+If another JMH instance is detected, the tool will try to terminate it. If that
+fails, it runs with `-Djmh.ignoreLock=true`.