ModelParserUtils used by full WildFly's ParseAndMarshalModelsTestCase isn't really testing much. It sets up controllers that kind of seem like the real ones, then parses the config, then executes the boot ops one at a time, ignoring failures. Then it repeats and compares the model created the first round to the one created the second round.

All this proves is that the same errors get ignored in each round.

When I try to correct this it turns out all sorts of errors are happening.

This should be redone to launch an embedded server or hc using the given config, then use a client to read the model. The modify the config, then do a second op to reverse the mod. This forces persistence. Then reload and read the model again. Compare the two models.

This approach is a test of the real software.