WarningBenchmarks 5000% regressions on 2025-02-11

[Akirathan]

+5000% regressions in WarningBenchmarks on 2025-02-11 caused by warnings.hasWarnings call:

• Good revision: a829bdb
• Bad revision: f419a12
• Range: a829bdb...f419a12
• Most likely culprit: Use BuiltinRootNode.ArgNode to extract argument for a builtin method #12201

To run single benchmark:

sbt:runtime-benchmarks> benchOnly org.enso.interpreter.bench.benchmarks.semantic.WarningBenchmarks.sameWarningVecSum

[JaroslavTulach]

Running

enso/tools/performance/engine-benchmarks$ ./bench_download.py -s engine -b develop wip/jtulach/BuiltinArgNode11827

shows that benchmarks were fine and even faster, but regressed with one of the last commits in the PR:

The culprit must be between

enso$ git log f211c19963357211c65597da92c8f910f940371f...e6af654e514dda263fa1dd4c2c7c015874b1b67b

There is a bunch of commits and a merge in the range, so bisecting will be needed.

[JaroslavTulach]

After bisecting:

e6af654e514dda263fa1dd4c2c7c015874b1b67b is the first bad commit
commit e6af654e514dda263fa1dd4c2c7c015874b1b67b
Author: Jaroslav Tulach <jaroslav.tulach@enso.org>
Date:   Fri Feb 7 09:38:41 2025 +0100

    Let SliceArrayVectorNode handle warnings on its own

Let's see what e6af654 has done: It's warnings.hasWarnings.

[JaroslavTulach]

An example of slowdown: check for hasWarnings in LengthVectorMethodGen node:

VectorBenchmarks_averageOverSliceWrapped10
VectorBenchmarks_averageOverSliceWrapped100