Issue1859

see FIXMEs in test_zlib.py

Some of these test case failures are caused by passing a start value != 1 to the adler32 method. Since java.util.zip.Adler32 does not support start values, these tests cannot pass with the current jython implementation.

I have separated out these tests so that

A: Tests that can pass on jython are not skipped
B: Tests that cannot pass on jython are still skipped

http://hg.python.org/jython/rev/50966417912b

The jython documentation for the zlib module should also be updated to note that a start value other than 1 is not supported on jython.

Other failures in this test are because they test an inconsistency in cpython, between the zlib.decompress function (which does not accept incomplete streams) and the zlib.decompressobj().decompress method (which does accept incomplete streams).

I have re-enabled the test for jython, but placed an if clause within it to prevent testing for cpython's inconsistent behaviour.

http://hg.python.org/jython/rev/733074a31d77

(There is a detailed comment in the code to this effect)

The tests based on check_big_decompress_buffer do not hang for me, even though they are marked as skipped on jython, because "they hang on jython".

I have checked in some comments to this effect, but made no functional changes.

http://hg.python.org/jython/rev/6b4a1088566e

We should track on which platforms these tests do actually hang.

As noted in the comments, the tests are misleading, because they appear to be testing for compressing and then decompressing data of size 1G + 1M. But the decorator wrapping these tests modifies this value to 5147, based on the value of test_support.real_max_memuse

I have unskipped the check_big_compress_buffer test, which was caused by hex string formatting taking an extremely long time.

More details on 

%x hex formatting takes O(N^2) time.
http://bugs.jython.org/issue2013

Changeset for last message

http://hg.python.org/jython/rev/eef625ddc617

I think we're getting close to closing this one Frank.

If we can get the hex formatting issue resolved (i.e. use a better algorithm for hex formatting), then we can restore the original cpython code path.

Alan: nice!

test_zlip passes now with the known skips Alan placed, so I'm marking this as closed