Use BuiltinRootNode.ArgNode to extract argument for a builtin method

[JaroslavTulach]

Pull Request Description

• as Dispatch of to_text, is_a & co. on intersection types #12192 (comment) states:
• there is a need to extract values from EnsoMultiValue in the "builtin method prelude code"
• to enable specializations based on type of arguments, we need to wrap such prelude by a Node
• hence introducing ArgNode & co.
• it kinda replaces the "static code emitted" by annotation processor
• with Truffle @Specializations done inside of ArgNode
• making this all more Truffle-like

Checklist

Please ensure that the following checklist has been satisfied before submitting the PR:

• All code follows the
  Scala,
  Java,
• Unit tests keep running
• Engine benchmarks are OK: slow run, faster run - looks OK, but one more verify run
• Libs benchmarks run is OK

[JaroslavTulach]

Now we attempt to provide a representationType to each Builtin. That way we can deduce from signature methods having argument like boolean that the desired builtin is Boolean and the Enso type should be Standard.Base.Data.Boolean.Boolean. With that information we can extract (via CastToNode) such a type from EnsoMultiValue....

[JaroslavTulach]

Right now there are two dataflow errors:

sbt:enso> runEngineDistribution --run test/Base_Tests/src/Semantic/Warnings_Spec.enso should.be.allowed.in.Vector|should.be.preserved.after.operations.on.Vector

[FAILED] Dataflow Warnings: [0/2, 770ms]

    - [FAILED] should be allowed in Vector [522ms]
        Reason: ['d', 'b', 'a'] did not equal [(Map_Error.Error 1 'b'), (Map_Error.Error 0 'a'), 'd']; first difference at index 0  (at Base_Tests/src/Semantic/Warnings_Spec.enso:272:9-126).


    - [FAILED] should be preserved after operations on Vector [248ms]
        Reason: ['warn4', 'warn3', 'warn2', 'warn1'] did not equal [(Map_Error.Error 3 'warn4'), (Map_Error.Error 2 'warn3'), (Map_Error.Error 1 'warn2'), (Map_Error.Error 0 'warn1')]; first difference at index 0  (at Base_Tests/src/Semantic/Warnings_Spec.enso:284:9-191).

I was debugging thru it, but I don't think I get any idea what can be the problem? Maybe something with wrapping errors? But why?

[JaroslavTulach]

Currently there are two failures:

• RuntimeVisualizationsTest
• RuntimeErrorsTest

Again, I am not really sure why at present.

[JaroslavTulach]

I'd say we have way too many Builtins. Not sure why Polyglot or Debug should be builtin types at all!?

[JaroslavTulach]

The first time ArgNode is requested to "cast", it searches for the ensoType representing the value of the argument. It would be better to do this when the ArgNode is being created, but it may be too early (before Standard.Base types are loaded in), so I haven't even tried.

Opinions?

[JaroslavTulach]

I believe the code is in shape to receive some inception review.

[Akirathan]

Generally looks like a good idea. Before merging, please don't forget to benchmark it. I am interested what will be the impact on performance.

[JaroslavTulach]

Checking benchmark results:

• StringBenchmarks_lengthOfStrings - more than 15x slower
• MultiValueBenchmarks - more than 10x slower
• AtomBenchmarks_benchGenerateList is 4x slower
• AtomBenchmarks_benchGenerateListQualified is 4x slower
• ListBenchmarks_mapAnyOverList by 15%
• EqualsBenchmarks_equalsStrings by 10%
• EqualsBenchmarks_equalsPrimitives by less then 10%
• ListBenchmarks_mapOverList by less then 10%

I had to patch the tool to download new results:

diff --git tools/performance/engine-benchmarks/bench_tool/__init__.py tools/performance/engine-benchmarks/bench_tool/__init__.py
index 918676fd62..11b88f0894 100644
--- tools/performance/engine-benchmarks/bench_tool/__init__.py
+++ tools/performance/engine-benchmarks/bench_tool/__init__.py
@@ -58,9 +58,9 @@ class Source(Enum):
 
     def artifact_names(self) -> List[str]:
         if self == Source.ENGINE:
-            return ["Runtime Benchmark Report"]
+            return ["Runtime Benchmark Report", "benchmark-results.xml"]
         elif self == Source.STDLIB:
-            return ["Enso JMH Benchmark Report"]
+            return ["Enso JMH Benchmark Report", "benchmark-results.xml"]
         else:
             raise ValueError(f"Unknown source {self}")

thus

• to mitigate Fix artifact upload in benchmark jobs #12226
• I am adding 934deae

And also

• Benchmarks made faster with few BranchProfile instances: f211c19
• running engine benchmarks again

[JaroslavTulach]

• Benchmarks made faster with few BranchProfile instances: f211c19
• running engine benchmarks again

There are overall improvements in the engine benchmarks:

ListBenchmarks may need some more work as this result isn't yet as great as it was:

• ListBenchmarks_mapOverList
• but the slowdown is less then 10%

Update on Feb 7, 2025: Engine benchmarks seem to be fine.

[JaroslavTulach]

Using hasWarnings here instead of simple warnings instanceof WithWarnings is causing problems:

• e6af654 is causing benchmarks regression
• WarningBenchmarks 5000% regressions on 2025-02-11 #12258 (comment)
• Use BuiltinRootNode.ArgNode to extract argument for a builtin method #12201 (comment)

Let's see what #12258 can do.