Clark Moody via bitcoin-dev
2017-07-14 18:43:37 UTC
(copying from GitHub per jonasschnelli's request)
I can understand the desire to keep all reference strings to the nice
14-character version by keeping the data payload to 40 bits, but it
seems to place artificial limitations on the format (year 2048 & 8191
transactions). I also understand that this might be addressed with
Version 1 encoding. But current blocks are not that far from having
8191 transactions.
You could go with a variable-length encoding similar to Bitcoin's
variable ints and gain the benefit of having a format that will work
for very large blocks and the very far future.
Also, the Bech32 reference libraries allow encoding from byte arrays
into the base-5 arrays native to Bech32. It seems like bit-packing to
these 40 bits might be overkill. As an alternative you could have one
bit-packed byte to start:
# First two bits are the protocol version, supporting values 0-3
V = ((protocol version) & 0x03) << 6
# Next two bits are magic for the blockchain
# 0x00 = Bitcoin
# 0x01 = Testnet3
# 0x02 = Byte1 is another coin's magic code (gives 256 options)
# 0x03 = Byte1-2 is treated as the coin magic code (gives 65280 more options)
M = (magic & 0x03) << 4
# Next two bits are the byte length of the block reference
B = ((byte length of block reference) & 0x03) << 2
# Final two bits are the byte length of the transaction index
T = ((byte length of transaction index) & 0x03)
# Assemble into the first byte
Byte0 = V | M | B | T
This gives you up to 3 bytes for each block and transaction reference,
which is 16.7 M blocks, or year 2336, and 16.7 M transaction slots.
Data part: [Byte0][optional magic bytes 1-2][block reference bytes][tx
reference bytes]
So the shortest data part would have 3 bytes in it, with the reference
version 0 genesis coinbase transaction having data part 0x050000.
I know this is a departure from your vision, but it would be much more
flexible for the long term.
Clark
I can understand the desire to keep all reference strings to the nice
14-character version by keeping the data payload to 40 bits, but it
seems to place artificial limitations on the format (year 2048 & 8191
transactions). I also understand that this might be addressed with
Version 1 encoding. But current blocks are not that far from having
8191 transactions.
You could go with a variable-length encoding similar to Bitcoin's
variable ints and gain the benefit of having a format that will work
for very large blocks and the very far future.
Also, the Bech32 reference libraries allow encoding from byte arrays
into the base-5 arrays native to Bech32. It seems like bit-packing to
these 40 bits might be overkill. As an alternative you could have one
bit-packed byte to start:
# First two bits are the protocol version, supporting values 0-3
V = ((protocol version) & 0x03) << 6
# Next two bits are magic for the blockchain
# 0x00 = Bitcoin
# 0x01 = Testnet3
# 0x02 = Byte1 is another coin's magic code (gives 256 options)
# 0x03 = Byte1-2 is treated as the coin magic code (gives 65280 more options)
M = (magic & 0x03) << 4
# Next two bits are the byte length of the block reference
B = ((byte length of block reference) & 0x03) << 2
# Final two bits are the byte length of the transaction index
T = ((byte length of transaction index) & 0x03)
# Assemble into the first byte
Byte0 = V | M | B | T
This gives you up to 3 bytes for each block and transaction reference,
which is 16.7 M blocks, or year 2336, and 16.7 M transaction slots.
Data part: [Byte0][optional magic bytes 1-2][block reference bytes][tx
reference bytes]
So the shortest data part would have 3 bytes in it, with the reference
version 0 genesis coinbase transaction having data part 0x050000.
I know this is a departure from your vision, but it would be much more
flexible for the long term.
Clark