Notes on the ReplayMod File Format

[Enchoseon]

In preparation for adding support for dumping raw packets into the ReplayMod file format, I am sharing my brief and uninformed notes. These notes aren't extensive and exist more to just document what I'm thinking and what other people have done.

Format

https://www.replaymod.com/forum/thread/609

• ReplayMod files use the .mcpr file extension, which is just a zip file with a different extension.
  • From my limited testing, I was able to unzip an .mcpr file into a folder, re-zip the contents, and play back the re-zipped file (I used the Ark compression utility)--so it looks like compression level isn't hardcoded or anything and it literally is just a zip file.
    • In fact, it looks like ReplayMod uses a very low amount of compression for higher throughput.
• The meat of the ReplayMod recording is the .tmcpr file, which contains a Java DataOutputStream that follows the following pattern until EOF:
  1. Integer telling the packet's timestamp
  2. Integer telling the byte length of the packet
  3. A byte array that represents the packet

private static void writePacket(DataOutputStream os, long time, Packet packet) throws IOException {
	ByteBuf buf = PooledByteBufAllocator.DEFAULT.buffer();
	NetUtils.writeVarInt(buf, REVERSE_PACKET_MAP.get(packet.getClass()));
	packet.write(buf);
	int length = buf.readableBytes();

	os.writeInt((int) time);
	os.writeInt(length);
	buf.readBytes(os, length);
	buf.release();
}

https://github.com/ReplayMod/ReplayStudio/

• "These packet dumps contain all packets sent from the server to the client (except login phase)"
• "They may also contain additional packets added by the recording software in order to display the client whose connection is recorded."
• There is some removal of redundant packets somehow, not sure what they mean by this. Do they mean that redundant data is reduced because the packets are compressed?
  • Update: https://github.com/ReplayMod/ReplayStudio/tree/master/src/main/java/com/replaymod/replaystudio/filter
    • Of note is the SquashFilter.java, which drops some packets (e.g. "ConfirmTransaction" packet) and does a lot of translating between versions.

The Plan

1. Look at the ReplayStudio repository: https://github.com/ReplayMod/ReplayStudio
2. Create a parser for the ReplayMod file format that can convert everything into PrismarineJS packets (https://github.com/PrismarineJS/node-minecraft-protocol)
3. Reverse it, convert PrismarineJS packets back into raw Minecraft packets and turn those packets into ReplayMod recordings.
4. Record a replay and successfully play it back in ReplayMod.
5. Record a replay with 2based2wait and ReplayMod at the same time, compare speed and filesize.

Forseen Problems

1. This needs to be async. The synchronous world downloader is slow as hell.
2. Moreover, this needs to be performant. Node.js is good at I/O operations, I'm bad at coding. Let's see which side is the victor.
3. I haven't used ReplayMod like, at all, so I have no clue how big the files will get.