A DESCRIPTION OF THE PROBLEM :
The MIME RFC requires that any character outside of the base64 alphabet be ignored.

The MIME decoder does this *except* within padding.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
java.util.Base64.getMimeDecoder().decode("AA=!=");

When two padding characters are appropriate, it should be legal to have characters outside the base64 alphabet between them, as such characters are always ignored.

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
All characters outside of the base64 alphabet should be ignored, wherever they are, so the above should decode without issue.
ACTUAL -
Exception java.lang.IllegalArgumentException: Input byte array has wrong 4-byte ending unit
        at Base64$Decoder.decode0 (Base64.java:733)
        at Base64$Decoder.decode (Base64.java:535)
        at Base64$Decoder.decode (Base64.java:558)

---------- BEGIN SOURCE ----------
java.util.Base64.getMimeDecoder().decode("AA=!=");
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
Input can be sanitised to remove characters outside of the base64 alphabet prior to processing. For example:

input = input.replaceAll("[^A-Za-z0-9+/=]","");

FREQUENCY : always