Spring Petclinic feedback comparison

The independent spice-framework/petclinic consumer owns a pinned, reproducible comparison with canonical Spring Petclinic rather than coupling an application benchmark to the core framework repository. Its benchmarks/spring-petclinic.json manifest selects Spring Petclinic commit 88e37c15cf6fc8490b01bc3e8e2c800cec1ac272, Spring Boot 4.1.0, and Java 25.

Prepare the external checkout explicitly:

git clone https://github.com/spring-projects/spring-petclinic.git <spring-checkout>
git -C <spring-checkout> checkout 88e37c15cf6fc8490b01bc3e8e2c800cec1ac272
<spring-checkout>/mvnw -DskipTests compile

Then run the comparison without network access:

cd <spice-petclinic-checkout>
make benchmark-spring SPRING_PETCLINIC=<spring-checkout>

The Go harness verifies the exact Spring commit, performs two warmups and seven measured edits, restores both source files on every success or failure path, and emits the complete samples plus p90 values as JSON. Maven runs with -o; missing dependencies fail with an explicit preparation instruction.

The current body-edit compile budget is Spice p90 below five seconds and no slower than the same Spring p90. On the reviewed Windows workstation, the first recorded run measured Spice at 2.160 seconds p90 and Spring at 6.174 seconds p90 (ratio 0.350). These are host observations, not universal product claims; the committed thresholds are the contract.

This scenario measures the compiler feedback portion of a save. The separate spice dev event stream reports generation reuse, build, graceful restart, and application-ready phases so restart regressions can be diagnosed instead of hidden inside one aggregate number.