Report per operation timings in CommandsPerformanceTest

[vogella]

The four binding cache tests each measured one block and reported it through org.eclipse.test.performance. The performance database it reports to has not been configured for years, so commitMeasurements and assertPerformance were no-ops: the tests could neither fail nor report anything. This applies the same treatment OpenCloseEditorTest and EditorSwitchTest already received.

Each test now times its own work and prints min, median, 90th percentile and maximum via UIPerformanceTestUtil.reportTimings. A single cached look-up is far too fast to time individually, so the three hit tests time a batch and report the distribution over batches, warming up first and bounded by time instead of a fixed iteration count. The cache miss test rebuilds the binding set for every sample, since measuring a miss only works while the cache is cold and the previous version could therefore take a single sample and no more.

One real defect surfaced on the way. The reverse look-up asked for the bindings of the null command, which is loop invariant, so the JIT hoisted it out of the measured loop and it reported 0.8ns per look-up. It now cycles through commands that are actually bound and reports 3.1ns, in line with the forward look-up. The tests also assert that a look-up returned something, so they can fail.

Sample output, and note how tight the distributions are compared with the UI tests:

BindingCache hit hard [per 100000 look-ups]      n=200  min=0.45  p50=0.47  p90=0.50  max=0.51 (ms)
BindingCache hit hard reverse [per 100000 look-ups] n=200  min=0.31  p50=0.34  p90=0.41  max=1.09 (ms)
BindingCache hit soft [per 1000 look-ups]        n=200  min=7.56  p50=7.92  p90=8.19  max=12.16 (ms)
BindingCache miss large                          n=20   min=0.72  p50=1.01  p90=1.17  max=1.63 (ms)

[github-actions]

Test Results

   858 files     858 suites   45m 11s ⏱️
 8 160 tests  7 917 ✅ 243 💤 0 ❌
20 382 runs  19 728 ✅ 654 💤 0 ❌

Results for commit 05f42f8.

♻️ This comment has been updated with latest results.