A DESCRIPTION OF THE PROBLEM :
This seems to be related to the changes related to JDK-8264160. In JDK 17 and earlier the following test would pass
        var pattern= Pattern.compile("(?:\\b|\\d)"+"äst");
        assertTrue("äst".matches(pattern.pattern()));

so a boundary match would match on the a-umlaut character. This behaviour seems to have received a breaking change with the bug earlier. Now (JDK 20) the test fails.

However as specified in the documentation the UNICODE_CHARACTER_CLASS should use the Unicode characters. I would expect this code to pass

var pattern= Pattern.compile("(?U)(?:\\b|\\d)"+"äst", Pattern.UNICODE_CHARACTER_CLASS);
assertTrue("äst".matches(pattern.pattern()));

but it fails in JDK 20 (and JDK 21 ea).

However "The UNICODE_CHARACTER_CLASS mode can also be enabled via the embedded flag expression (?U)."

var pattern2= Pattern.compile("(?U)(?:\\b|\\d)"+"äst");
assertTrue("äst".matches(pattern2.pattern()));

when using the flag, the test passes

REGRESSION : Last worked in version 17.0.8

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
run the code below

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
this code runs without exception

ACTUAL -
the test fails with an IllegalStateException


---------- BEGIN SOURCE ----------
        var pattern= Pattern.compile("(?:\\b|\\d)"+"äst", Pattern.UNICODE_CHARACTER_CLASS);
        if(!"äst".matches(pattern.pattern()))
            throw new IllegalStateException("äst should match pattern1");
---------- END SOURCE ----------

FREQUENCY : always