UnboundConfigurationPropertiesException is no longer thrown from IndexedElementsBinder

[winfriedgerlach]

Short story:

IndexedElementsBinder from Spring 3.5.0 doesn't throw UnboundConfigurationPropertiesException on missing target field any more, even on first element in a list.

Long story:

We are using Spring Boot to bind a list of configuration properties from a file to a @ConfigurationProperties class.

Our configuration looks like this:

prefix.list[0].a=value1
prefix.list[0].b=value2
prefix.list[1].a=value3
prefix.list[1].b=value4

(in fact, we are mapping lists of lists, but I think that our problem would also occur for a simple list)

With prior Spring Boot versions, both prefix.list[5].a=someValue and prefix.list[0].foo=someValue would throw UnboundConfigurationPropertiesException:

• the former due to a "binding gap" in the list
• the latter because the field "foo" does not exist in the target object (cf. IndexedElementsBinder 3.4.x: if (value == null) { break; }). This validation is now broken in Spring Boot 3.5.0.

Deeper analysis showed that our problem is caused by the changes in #44867. We are not particularly happy about these changes, because

1. for us, full validation is more important than performance (sure, other users have other needs, cf. Provide a possibility to extend/override configuration properties binding process #42361) and
2. @philwebb's assessment "with YAML configuration lists are unlikely to have missing elements" does not apply to us, as we are using plain old properties files

But even if we accept that configuration validation is only performed on the first 10 items in the list, IMHO Spring Boot should still detect if the target field does not exist. This was probably just an oversight when implementing #93113a4 ?

We would be glad if you had a look into this issue. Thanks everyone!

[philwebb]

Thanks for raising the issue @winfriedgerlach. I've decided to pretty much revert 93113a4 since I hadn't anticipated how much it would break things.

[winfriedgerlach]

Thanks @philwebb for the incredibly quick response and the effort you put in to make binding more efficient! I am looking forward to 3.5.1 then!

One more thing I noticed with the implementation in 3.5.0: Even if the list contained only one element, IndexedElementsBinder#bindIndexed would still go through the loop ~10 times, which seemed odd to me. But I guess this observation is now obsolete anyway.

[philwebb]

bindIndexed would still go through the loop ~10 times, which seemed odd to me. But I guess this observation is now obsolete anyway.

Indeed, that caused a great deal of pain for nested lists.