Ben Weidig created TAP5-2839:
--------------------------------
Summary: Add JMH benchmarking
Key: TAP5-2839
URL: https://issues.apache.org/jira/browse/TAP5-2839
Project: Tapestry 5
Issue Type: Improvement
Affects Versions: 5.10.0
Reporter: Ben Weidig
Performance changes in Tapestry aren't currently measurable, at least not in a
deterministic fashion.
There is no way to state what a change to the IoC registry, a coercion, or the
DOM rendering actually costs based on hard numbers.
h4. Proposal
Introduce [JMH|https://github.com/openjdk/jmh] as a Gradle convention plugin
({{tapestry.jmh-convention}}), so any module can carry microbenchmarks in
{{src/jmh/java}}.
* JMH {{1.37}}, declared in the version catalog
* Benchmark source set compiles against the module's own classes and its
{{provided}} configuration, mirroring the test source set
* Two tasks per module: {{jmh}} (run) and {{jmhList}} (list without running)
* Run options as {{-P}} properties: {{jmh.include}}, {{jmh.exclude}},
{{jmh.quick}}, {{jmh.forks}}, {{jmh.profilers}}, {{jmh.resultFile}}, and others
* Results are written as JSON, so a before/after pair can be diffed, and easier
visualized
h4. CI
Benchmarks are deliberately *not* wired into {{check}} or {{build}}.
A JMH run can take minutes or even longer, and has therefore no place in CI.
The root task {{jmhCompileAll}} compiles every benchmark source set instead, so
refactorings cannot break them silently.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
