ChangeLog

Version 0.5.0beta3

A bucketload of API changes, Javadoc's overview.html updated accordingly; API not stabilized yet.
A lot more syntax/validation tests.
The utc-millisec format specifier didn't validate decimal numeric instances even though it should: fixed.
Remove support for color and style format specifiers, pom.xml dependency pruning (and update).
Some more Javadoc, still incomplete.
Syntax validation for type and disallow keywords missed the fact that array elements (if the value is an array) had to be unique: fixed.
In schemas, an id with a non normalized URI is considered illegal.
Fix issue #7: the validation context now carries information about the current schema's URI (even if empty), allowing $ref resolution to also work in such a case.

Fix an embarrassing bug with numeric instance validation: Jackson's .canConvertToLong() will also return true for decimal values which integral part fits into a long, but we always want a BigDecimal for decimal values. That'll teach me not to read the Javadoc completely...
Improve testing.
Dependencies update, except for TestNG (my IDE doesn't handle TestNG 6.4+...)

This is the first iteration of a complete rewrite.

Validators are now thread-safe, at least to the best of my knowledge.
Validation is more than twice faster than with 0.4.x.
$ref support is now complete, including id lookup.
JSON Pointer now obeys the newer draft.
Known bug: id lookup can wrongly return wannabe schemas in enum, for instance -- such object instances should not be considered. In the same vein, a schema with two identical id fields is... Well... Awkward to begin with, but the current implementation will grab the first one it finds, and you cannot reliably tell which.

Fix subschema validation issue: addressing a subschema would behave as a ref lookup, and as such you could not validate twice since the second validation would wrongly detect a ref loop error (issue #4).
Fix a discrepancy in JsonPointer between .equals() and .hashCode(): while elements used in both functions were related to each other in any case, they were not the same. This has the added, though invisible, benefit, of a faster .hashCode().

More performance tweaks, but that's not the most important...
Instance-only validation mode: you can now skip schema syntax check and make the validation process assume all your schemas are valid. Use it at your own risk!
Experimental draft v4 support: support for the new definition of the required keyword, and also minProperties and maxProperties.
Account for $schema: if this property is found in a schema, then the implementation automatically switches schemas -- the URI for draft v4 is assumed to be http://json-schema.org/draft-04/schema#.
Ability to change the default schema version: it is draft v3 by default, but it is possible to change it to draft v4 instead.
Ability to register keywords to override a specific schema version.

A 25 times performance improvement thanks to validator caching and other improvements -- see here for an example.
Ability to operate a fail-fast validation, that is a validation which will fail on the first failed item validation.

Fix a bug with uniqueItems validation: when the keyword was seen, its value was always assumed to be true.

Ability to register your own validators for new keywords or existing ones -- see here for an example.
Ability to register handlers for arbitrary URI schemes (the only builtin scheme is still http).
"Subschema" validation: with a JSON document containing several schemas, you can now start validating from one JSON Pointer location in that document (example: validator.validate("#/schema1", instance)).
Very robust JSON Pointer addressing: you can now lookup such exotic paths as #/%2f//a b in your schemas.
Shortcut validation of numeric instances if they fit into Java's long type.
Validated number instances can now be of arbitrary length/precision.
Uses Google's libphonenumber for phone number validation.

Still no support for URIs which are relative but are not JSON Pointers...

Fix a nasty bug with $ref: on a JSON Pointer with mutiple path elements (ie, #/e1/e2), it would try each path from the root instead of in depth

Complete JSON Schema validation (section 5 of the draft), with schema syntax validation
JSON Schema v3 validates itself -- which means you can validate schemas, too.
Looking up JSON documents (schemas or instance to validate) from the file system, from a resource, from URLs.
$ref lookup (JSON Pointer, HTTP and a combination of both only for now), with loop detection.

Numeric types are limited to long or double at a maximum: this is due to using Jackson's ObjectMapper with the default deserialization configuration.