Incremental Compression

[Eyal Lotem]

Hey.

I have a problem in some network code. I want to send my packets compressed,
but I don't want to compress each packet separately (via .encode('zlib') or
such) but rather I'd like to compress it with regard to the history of the
compression stream. If I use zlib.compressobj and flush it to get the
packet data, I cannot continue to compress with that stream.

I cannot wait until the end of the stream and then flush, because I need to
flush after every packet.

Another capability I require is to be able to copy the compression stream.
i.e: To be able to create multiple continuations of the same compression
stream. Something like:

a = compressobj()
pre_x = a.copy()
x = a.compress('my_packet1')
# send x
# x was not acked yet, so we must send another packet via the pre_x
compressor
y = pre_x.compress('my_packet2')

Is there a compression object that can do all this?

[Adam DePrince]

On Sat, 2006-03-25 at 03:08 +0200, Eyal Lotem wrote:

Hey.

I have a problem in some network code. I want to send my packets compressed,
but I don't want to compress each packet separately (via .encode('zlib') or
such) but rather I'd like to compress it with regard to the history of the
compression stream. If I use zlib.compressobj and flush it to get the
packet data, I cannot continue to compress with that stream.
Yes, you can.

Help on built-in function flush:

flush(...)
flush( [mode] ) -- Return a string containing any remaining
compressed data.
mode can be one of the constants Z_SYNC_FLUSH, Z_FULL_FLUSH,
Z_FINISH; the
default value used when mode is not specified is Z_FINISH.
If mode == Z_FINISH, the compressor object can no longer be used
after
calling the flush() method. Otherwise, more data can still be
compressed.

you want to call

mycompressor.flush( zlib.Z_SYNC_FLUSH )

The difference between the flushes is this:

1. Z_SYNC_FLUSH. This basically send enough data so that the receiver
will get everything you put in. This does decerase your compression
ratio (however, in weird case when I last played with it, it helped.)

2. Z_FULL_FLUSH. This sends enough data so that the receiver will get
everything you put in. This also wipes the compressors statistics, so
the when you pick up where you left of, the compressor will compress
about as well as if you had just started, you are wiping its memory of
what it saw in the past.

3. Z_FINISH. This is the default action, this is what is killing you.

Good luck - Adam DePrince

I cannot wait until the end of the stream and then flush, because I need to
flush after every packet.

Another capability I require is to be able to copy the compression stream.
i.e: To be able to create multiple continuations of the same compression
stream. Something like:

a = compressobj()
pre_x = a.copy()
x = a.compress('my_packet1')
# send x
# x was not acked yet, so we must send another packet via the pre_x
compressor
y = pre_x.compress('my_packet2')

Is there a compression object that can do all this?

Ahh, you are trying to "pretune" the compressor before sending a little
bit ... I think C-zlib does this, but I don't know for sure.

- Adam DePrince

[Eyal Lotem]

Adam DePrince wrote:

On Sat, 2006-03-25 at 03:08 +0200, Eyal Lotem wrote:

Hey.

I have a problem in some network code. I want to send my packets
compressed, but I don't want to compress each packet separately (via
.encode('zlib') or such) but rather I'd like to compress it with regard
to the history of the
compression stream. If I use zlib.compressobj and flush it to get the
packet data, I cannot continue to compress with that stream.
Yes, you can.

Help on built-in function flush:

flush(...)
flush( [mode] ) -- Return a string containing any remaining
compressed data.
mode can be one of the constants Z_SYNC_FLUSH, Z_FULL_FLUSH,
Z_FINISH; the
default value used when mode is not specified is Z_FINISH.
If mode == Z_FINISH, the compressor object can no longer be used
after
calling the flush() method. Otherwise, more data can still be
compressed.

you want to call

mycompressor.flush( zlib.Z_SYNC_FLUSH )

The difference between the flushes is this:

1. Z_SYNC_FLUSH. This basically send enough data so that the receiver
will get everything you put in. This does decerase your compression
ratio (however, in weird case when I last played with it, it helped.)

2. Z_FULL_FLUSH. This sends enough data so that the receiver will get
everything you put in. This also wipes the compressors statistics, so
the when you pick up where you left of, the compressor will compress
about as well as if you had just started, you are wiping its memory of
what it saw in the past.

3. Z_FINISH. This is the default action, this is what is killing you.

Good luck - Adam DePrince

Thanks! That really helps.

I cannot wait until the end of the stream and then flush, because I need
to flush after every packet.

Another capability I require is to be able to copy the compression
stream. i.e: To be able to create multiple continuations of the same
compression stream. Something like:

a = compressobj()
pre_x = a.copy()
x = a.compress('my_packet1')
# send x
# x was not acked yet, so we must send another packet via the pre_x
compressor
y = pre_x.compress('my_packet2')

Is there a compression object that can do all this?

Ahh, you are trying to "pretune" the compressor before sending a little
bit ... I think C-zlib does this, but I don't know for sure.

Yeah, but I don't need a powerful tuning, just a means to copy the
compressor's state. I guess I'll need to write some C for this.

Thanks again!

- Adam DePrince