Spray and Pray - Cost of Overly-Broad Spotless Targets

In the post on the unseasonable Gradle configuration time Spotless was the culprit for the huge increase in the configuration time for Androidify project. As a workaround that post suggested removing Spotless Gradle plugin, but it turns out there is a better way. Our performance issue rests in the following Spotless configuration:

spotless {
    kotlin {
        target("**/*.kt")
        targetExclude("**/build/**/*.kt")
    }
    kotlinGradle {
        target("**/*.kts")
        targetExclude("**/build/**/*.kts")
    }
    format("xml") {
        target("**/*.xml")
        targetExclude("**/build/**/*.xml")
    }
}

From a quick glance it looks reasonable, but if you look at it a bit longer you will spot that with target("**/*.kt") we are telling Spotless that any directory in our project could contain a .kt file, which effectively means Spotless has to check every single directory in the project. Can we be more precise? Yes we can!

spotless {
    kotlin {
        target("src/**/*.kt")
    }
    kotlinGradle {
        target("*.kts")
    }
    format("xml") {
        target("src/**/*.xml")
    }
}

Instead of guessing whether this change improves the situation it is best to use gradle-profiler to validate the results. Based on benchmarks this trivial looking change has a huge impact on the Gradle configuration time dropping it by 67%!

Without this change Gradle profiler benchmark of ./gradlew build --dry-run had 30,102 ms mean with 562 ms standard deviation, but after the change it decreased to 10,076 ms mean with 478 ms standard deviation.

A well configured project should spend less than 100 ms per project in configuration. Even after this change androidify configuration takes about 630 ms per project. We still have more work to do!