false positive non null return with gradle and jetbrains annotations

[xenoterracide]

/home/xeno/IdeaProjects/untitled/src/main/java/org/example/Main.java:10: error: [NullAway] returning @Nullable expression from method with @NonNull return type

it's talking about the map lambda

    project.getObjects().property(String.class).map( s -> null);

but the gradle implementation very clearly allows a nullable return.

copypasta of gradles source

public interface Transformer<OUT extends @org.jetbrains.annotations.Nullable Object, IN> {
    OUT transform(IN in);
}

tasks.withType<JavaCompile>().configureEach {
  options.errorprone {
    option("NullAway:AnnotatedPackages", listOf("com", "org", "net", "io", "dev", "graphql").joinToString(","))
    error("NullAway")
  }
}

bug.tar.gz

reproducer uses dynamic versions, so here's the at time of writing resolutions that I think are important.

  id("net.ltgt.errorprone") version "4.1.0"

+--- com.google.errorprone:error_prone_core:2.+ -> 2.36.0
\--- com.uber.nullaway:nullaway:0.+ -> 0.12.2

afaik, gradle api is comprehensively annotated with non null as the default. It's used by kotlin consumers with null safety in mind.

[msridhar]

Will take me a bit of time to dig into this one. There are some places where we initially special-cased JSpecify annotations, and I think that was a mistake; we should just accept any @Nullable annotation like we do elsewhere in NullAway, until we see it causing problems. So we can fix that.

But beyond that, I think we will need proper support for inference on generic methods (#1075) before we can support code like this. In this case I assume map is a generic method, and we need to infer that it's ok to treat its output type as @Nullable.

[xenoterracide]

it is, here's the source for that https://github.com/gradle/gradle/blob/5cb81304ee750fd32c943fefba19b5387000d226/subprojects/core-api/src/main/java/org/gradle/api/provider/Provider.java#L127

note: worth saying that I believe that gradle will be moving to jspecify in version 9, there's a ticket related to that ... somewhere, linked in another ticket.