Investigate string allocations and array generation across gherkin flavours

[luke-hill]

🤔 What's the problem you've observed?

Some work was done in .NET here and here. This work should be mimicked (Or at least qualified and checked off that an equivalent piece cannot be done elsewhere).

✨ Do you have a proposal for making it better?

Check equivalent work to PR #336 -> Improved parsing time

• Java
• JavaScript
• Ruby
• Go
• Python
• C
• Objective-C
• Perl
• PHP
• Dart
• C++

Check equivalent work to PR #344 -> Avoid allocation and improve parsing time

• Java
• JavaScript
• Ruby
• Go
• Python
• C
• Objective-C
• Perl
• PHP
• Dart
• C++

📚 Any additional context?

Some languages it may not be relevant for

[jkronegg]

Java

Reading the very_long.feature using OpenJDK21 with JMH micro-benchmark gives the following result (the parser receives the file content as a String):

Benchmark                  Mode  Cnt     Score     Error  Units
MyClassBenchmark.original  avgt   25  6601.674 ± 433.332  us/op

When reading 1000 times the very_long.feature, IntelliJ's Profiler gives the following flame graph:

Most of time is passed on String trimming (about 50% of total duration). I already worked on the point in the past (#84), but some improvement still needs to be done (I have some ideas on how to improve that😉). Otherwise there is no noticeable performance hot spot.

On one of my real-life project with about 100 rules and 1000 test scenarios, the Parser.parse() takes 340 ms (with about 100 ms of String trimming, that is: only about 30% of the parsing duration).

I'll create an issue on that point.

JMH benchmark code:

public class MyClassBenchmark {
    @Benchmark
    @BenchmarkMode(Mode.AverageTime)
    @OutputTimeUnit(TimeUnit.MICROSECONDS)
    public GherkinDocument original(MyClassPlan plan) {
        return plan.parser.parse(plan.featureContent, plan.matcher, "very_long.feature");
    }

    @Test
    void test_for_profiler() {
        MyClassPlan plan = new MyClassPlan();
        for (int i=0; i<1000; i++) plan.parser.parse(plan.featureContent, plan.matcher, "very_long.feature");
    }
}

@State(Scope.Benchmark)
public class MyClassPlan {
    TokenMatcher matcher = new TokenMatcher("en");
    IdGenerator idGenerator = new IncrementingIdGenerator();
    Path path = Paths.get("../testdata/good/very_long.feature");
    Parser<GherkinDocument> parser = new Parser<>(new GherkinDocumentBuilder(idGenerator, path.toString()));
    String featureContent;
    {
        try {
            featureContent = new String(Files.readAllBytes(path));
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    }
}