(This is the same comment I just added to the http://java.net/jira/browse/JSR_283-811 issue.)
While I agree that this was a problem in JSR-283, I don't believe it's necessary or even advisable to fix or address.
First of all, changing the "jcr:requiredType" property definition to use the same case as the constants in javax.jcr.PropertyType will not be backward compatible. This is because a JSR-283 repository can represent the node type definitions, property definitions, and child node definitions within the "/jcr:system/jcr:nodeTypes" subgraph. Changing the constraints would make all these "nt:propertyDefinition" nodes in existing repositories invalid. (Note that the "jcr:requiredType" property is mandatory.
Secondly, this shouldn't be an issue for an implementation. This is because the javax.jcr.nodetype.PropertyDefinition interface defines the "getRequiredType()" method as returning an int, not a string. In other words, the string representation of the required type within the "/jcr:system/jcr:nodeType" subgraph cannot be used within the implementations javax.jcr.nodetype.PropertyDefinition implementation without converting to an integer. Yes, an implementation will not be able to use existing javax.jcr.PropertyType.valueFromName(String) to perform the conversion, and instead needs to use a case-insensitive version. But that is purely an implementation detail.
Perhaps a better alternative change would be to change the existing "javax.jcr.PropertyType.valueFromName(String)" implementation to be case-insensitive. Consider that the CND notation introduced in JSR-283 specifically states in Section 184.108.40.206 "Case Insensitive Keywords" specifically states:
The keywords of CND, though defined above as terminal strings with specific cases, are in fact case-insensitive.
For example, STRING can be written string, String or even StRiNg.
Therefore, an implementation that parses a CND file already has to convert a property definition's type in a case-insensitive manner, and changing the existing "javax.jcr.PropertyType.valueFromName(String)" implementation to be case-insensitive would allow an implementation to use that method in CND parsing rather than its own custom implementation.
However, I'm actually in favor of doing none of these, and marking this issue as 'not a bug'.