Yes, accumulating the data will be backward-compatible.

But what if add a special flag for zlib.decompress, which makes bz2 and lzma compatible decoder?  I.e.:

1) unconsumed_tail is always empty (the unconsumed data accumulated in internal buffer, no need manually add it to next chunk of data).  Or even non-existent.
2) EOFError raised if decompress() is called after end-of-stream is reached.

Of course, it is a new feature.