There's code in the inflate() method of classlib/modules/archive/src/main/java/java/util/zip/Inflater.java (r926871 and earlier) which short-cuts the inflation if the parameter 'nbytes' is 0 - ie if the caller is actually requesting zero bytes:
if (nbytes == 0)
return 0; --- Don't do any work: DON'T UPDATE THE *FINISHED* FLAG!
While it seems reasonable, it is in fact incompatible with the RI (which does attempt to process the deflated source data in the nbytes, and will update the finished() flag) - and so causes problems with a common inflation loop scenario, where the code loops to populate a destination byte-array of known size while inflater.finished() is not true. Even if the destination byte-array is known to be zero-length (e.g. we're deflating a zero-length file), the coder might still reasonably expect calling inflate() to set the finished() flag to true. However, unlike the RI, Harmony implementation does not update the finished() flag, which leads to an infinite loop or some other bad outcome.
Bit short on time right now, will try to provide a better testcase and possibly a patch at a later date.