[timtebeek]

Testcontainers2Migration changes DockerComposeContainer to ComposeContainer but keeps .withLocalCompose(true). Testcontainers 2 removed withLocalCompose(boolean) from both classes, so the call no longer compiles. In Testcontainers 2 a ComposeContainer built from compose files alone, without a DockerImageName, already uses the local Docker Compose binary, so withLocalCompose(true) has no replacement and can go.

Before:

DockerComposeContainer env = new DockerComposeContainer(new File("docker-compose.yml"))
        .withLocalCompose(true)
        .withTailChildContainers(true);

After the migration today (does not compile):

ComposeContainer env = new ComposeContainer(new File("docker-compose.yml"))
        .withLocalCompose(true)
        .withTailChildContainers(true);

After this change:

ComposeContainer env = new ComposeContainer(new File("docker-compose.yml"))
        .withTailChildContainers(true);

This surfaced while validating the Spring Boot 4.1 migration over open source Spring Boot 3.x repositories, in adorsys/open-banking-gateway (opba-protocols/sandboxes/xs2a-sandbox/.../BasicTest.java): cannot find symbol: method withLocalCompose(boolean) on ComposeContainer.

This adds RemoveLocalCompose, which removes withLocalCompose(true) on DockerComposeContainer and ComposeContainer, in a chain or as a statement, using RemoveMethodInvocationsVisitor with isTrueArgument(). It runs in Testcontainers2Migration before the type change. withLocalCompose(false) and non-literal arguments are left alone: their Testcontainers 2 form is a constructor that takes a DockerImageName, which needs a choice of image.

Is this the right change?

• Do we already have it? No.
• Could it be declarative? Not quite: RemoveMethodInvocations removes the call whatever its argument, which would turn withLocalCompose(false) into local compose without a word. The recipe is a thin wrapper around RemoveMethodInvocationsVisitor with an argument predicate.
• Is it worth it? Yes, withLocalCompose(true) is the usual way to use Docker Compose v2 from Testcontainers 1.x, and every such call breaks.
• Does it fix the problem, or move it? It fixes it for true, which is the Testcontainers 2 default for these constructors.
• Does it do no harm? Only the literal true is removed; false and expressions stay, for a person to decide.

Tests: new RemoveLocalComposeTest (removal in a chain and as a statement; false and a variable unchanged) and TestcontainersBestPracticesTest#composeContainerWithLocalCompose (the full migration on the shape from open-banking-gateway; failed before). TestcontainersBestPracticesTest and Testcontainers2LocalStackTest pass.