LinkedCaseInsensitiveMap does not properly support Java 8's merge() and compute() methods [SPR-15026]

[spring-projects-issues]

John Mark opened SPR-15026 and commented

LinkedCaseInsensitiveMap does not properly implement many of the new Map methods that have been added in Java 1.8 such as merge() and the compute* methods.

Using those methods on a LinkedCaseInsensitiveMap results in an entry that can never be retrieved.

---

Affects: 4.3.4

Issue Links:

• LinkedCaseInsensitiveMap doesn't implement getOrDefault properly [SPR-13981] #18553 LinkedCaseInsensitiveMap doesn't implement getOrDefault properly
• LinkedCaseInsensitiveMap doesn't override HashMap.clone() [SPR-14509] #19078 LinkedCaseInsensitiveMap doesn't override HashMap.clone()
• HttpHeaders.keySet() is no longer case-insensitive [SPR-15087] #19653 HttpHeaders.keySet() is no longer case-insensitive

Referenced from: commits e9a87de, 505480c, 8147c11, 50e5a65

[spring-projects-issues]

Juergen Hoeller commented

We're unfortunately constrained in 4.3.x there since we cannot have hard dependencies on the java.util.function interfaces - as used in those merge and compute signatures - in our core codebase. However, we can certainly revisit this for 5.0 where we have a Java 8 baseline now.

[spring-projects-issues]

Juergen Hoeller commented

Looks like there is a straightforward way out: rearchitecting LinkedCaseInsensitiveMap to implement Map directly, delegating to an internal LinkedHashMap for specific operations only and therefore not inheriting HashMap's overrides of those new Java 8 default methods. Instead, we can transparently reuse the original default method implementations from the Map interface which seem to work for our purposes. This is even compatible with Java 6, 7 and 8 (and forward with 9) since we do not have to implement any of the new signatures in LinkedCaseInsensitiveMap itself, so we can technically backport this to 4.3.6.

From a positive perspective, this is also aligned with our existing LinkedMultiValueMap implementation style now. At the same time, it is a significant change in the class hierarchy which we usually wouldn't do in a maintenance release. Since I don't expect anybody to cast a LinkedCaseInsensitiveMap to LinkedHashMap or HashMap, this shouldn't be an issue in practice... and we can even keep removeEldestEntry overrides in custom subclasses working. All in all, it is more important that all of those new Map operations work properly; I therefore see this change as acceptable for the 4.3.x line as well.

[spring-projects-issues]

Andy Wilkinson commented

This has a rather unpleasant side-effect on HttpHeaders. Consider the following setup:

HttpHeaders headers = new HttpHeaders();
headers.add("X-Foo", "bar");

With 5.0 M3 and earlier, this is true:

headers.keySet().contains("x-foo")

For the latest 5.0 snapshots, it's false as LinkedCaseInsensitiveMap.containsKey(Object) is no longer used. This breaks Spring REST Docs with Spring Framework 5.0 snapshots as I currently rely on the case-insensitive behaviour of the key set.

[spring-projects-issues]

Andy Wilkinson commented

On Stephane's advice, I've opened #19653