Changing the transaction version number to be varint

Hi all,

In the transaction today we have a version field which is always 4 bytes. The rest of the integer encoding in a transaction is variable-size because it saves on bytes.

Specifically, in practice this means that almost all of the transaction have bytes 2, 3 & 4 set to zero[1].

The question that I was pondering is that when we accept a new version of transaction format (flextrans uses 4), what would the impact be of also changing the way that the version number is actually serialized to be var int.

The benefit would be that each and every transaction looses 3 bytes. These can be used differently in v1 transactions and are not needed at all to be there for newer transaction formats. The secondairy benefit is that, at least for FlexTrans[2], 100% of all the integers in the transaction are following exactly the same encoding, the var-int encoding.

There is currently no consensus rule that rejects transactions which lie about their version, so obviously this rule should not and can not be introduced retro-actively. It will be from a certain block-height.

The way to do this is that from a certain block-height the current transaction format labels bytes 2, 3 & 4 to be unused. From that same block height the interpretation of the first byte is as varint. Last, we add the rule from that block-height that only transactions that do not lie about their version number are valid. Which means version 1.

Do people see any problems with this? This could be done as a soft fork.

1) It should be 100% because there is no transaction version defined that sets them to non-zero, but there is no consensus rule that rejects transactions that lie about their version number. 2) https://bitcoinclassic.com/devel/Flexible%20Transactions.html

> On 20 Jan 2017, at 22:02, Tom Zander via bitcoin-dev <bitcoin-dev@lists•linuxfoundation•org> wrote:
>
> The way to do this is that from a certain block-height the current
> transaction format labels bytes 2, 3 & 4 to be unused.
> From that same block height the interpretation of the first byte is as
> varint.
> Last, we add the rule from that block-height that only transactions that do
> not lie about their version number are valid. Which means version 1.
>
> Do people see any problems with this?
> This could be done as a soft fork.

Yes, because:

a) what you are talking is a hardfork, because existing nodes will not be able to deserialise the transaction. They will forever interpret the first 4 bytes as nVersion.

b) it is not a “lie” to use non-version 1 txs. It is permitted since v0.1. And version 2 txs is already used due to BIP68.

c) if you are talking about changing the tx serialisation just for network transfer, it’s just a p2p protocol upgrade, not softfork nor hardfork

-------------------------

There are 3 ways to introduce new tx formats:

1. through a softfork, and make the old clients blind to the new format. That’s the segwit approach

2. through a hardfork. Forget the old clients and require new clients to understand the new format. That’s the FlexTran approach (in my understanding)

3. p2p only, which won’t affect consensus. No one could stop you if you try to copy a block by writing in your native language and pass to your peer.

In either way, one could introduce whatever new format one wants.