From issue 5210:

amaury.forgeotdarc wrote:
> Hm, I tried a modified version of your first test, and I found another
> problem with the current zlib library;
> starting with the input:
> x = x1 + x2 + HAMLET_SCENE    # both compressed and uncompressed data
>
> The following scenario is OK:
> dco.decompress(x) # returns HAMLET_SCENE
> dco.unused_data   # returns HAMLET_SCENE
>
> But this one:
> for c in x:
>     dco.decompress(x) # will return HAMLET_SCENE, in several pieces
> dco.unused_data   # only one character, the last of (c in x)!
>
> This is a bug IMO: unused_data should accumulate all the extra uncompressed 
> data.

Ideally, I would prefer to raise an EOFError if decompress() is called
after end-of-stream is reached (for consistency with BZ2Decompressor).
However, accumulating the data in unused_data is closer to being backward-
compatible, so it's probably the better approach to take.