Implement Abort() in XZStream and relatives

[ied206]

Summary

• Implement the Abort() method in XZStream and its relatives.
  • Solves the aborting performance issue denoted in Add function to cancel compress. #14

Details

Why current implementation does not have Abort()?

My compression stream implementation follows .NET BCL's official DeflateStream interface. It does not define Abort() or similar method. Many other implementations also have a similar behavior.

My guess is that zlib is a relatively fast compression algorithm, so merely calling Close() was enough even though a user wanted to abort the session.

When and why Close() takes much time?

When closing the XZStream (and its relatives), liblzma flushes its internal buffer.

• Compress mode: The remaining input buffer must be compressed to finalize an xz file into a proper state.
• Decompress mode: Nothing to do before closing.

In threaded compress mode, liblzma aggressively buffers the input data. The compression itself is merely postponed as much as possible. The compression time itself looks shorter, but it goes into the time needed to properly finalize an xz file.

What does the new Abort() do?

It just frees internal liblzma resources. To be exact, it frees LzmaStream, the context for the liblzma operation. It prevents any other liblzma operations from being conducted.

Furthermore, calling Abort() prevents Flush() and FinishWrite() from being called in Dispose()/Close().

It will save up some time on compress mode, especially threaded compression mode. It does not in decompress mode.

liblzma will refuse any operation after calling Abort().

Users must dispose of XZStream right after calling Abort(). The compressed data will be invalid and must be discarded.

Benchmark: How much does Abort() save time compared to Close()?

Taken from the AbortCompress test prints:

• Threaded compression benefits a lot.

# Singlethreaded Compression
A.pdf, -1 = Close() took 3.000ms
A.pdf, -1 = Abort() took 1.000ms
B.txt, -1 = Close() took 1.999ms
B.txt, -1 = Abort() took 0.999ms
C.bin, -1 = Close() took 6.999ms
C.bin, -1 = Abort() took 2.000ms
# Multithreaded Compression
A.pdf, 1 = Close() took 20.002ms
A.pdf, 1 = Abort() took 1.000ms
B.txt, 2 = Close() took 8.000ms
B.txt, 2 = Abort() took 0.000ms
C.bin, 2 = Close() took 222.162ms
C.bin, 2 = Abort() took 8.001ms