Op 29 sep. 2017, om 05:18 heeft Peter Todd <pete@petertodd.org> het volgende geschreven:
> 
> On Thu, Sep 28, 2017 at 03:43:05PM +0300, Sjors Provoost via bitcoin-dev wrote:
>> Peter Todd wrote:
>> Perhaps outside the scope of BIP173, but what about baking it into the protocol? That way a transaction that's sent too late, simply won't get confirmed. This removes the need for refund logic or asking a customer to pay just a few extra cents. You could also disallow a second payment.
>> 
>> Two downsides I can think of:
>> * privacy, as differences in expiration policy would be visible on chain
>> * miners might be able to game it in their interaction with brokers
> 
> This has been discussed many times before; there are *severe* downsides to
> making it possible for transactions to become invalid after the fact.

I've heard of that general principe, but I'm having trouble finding a good resource that describes it more precisely.

Is it a peer to peer or mempool issue? E.g a transaction might be accepted into the mempool and relayed at one point in time and suddenly become invalid before they're committed to a block? Or that a node receives a transaction, thinks it's invalid because the address already expired, but then receives an older block later which contains that transaction?

Once in a block, I don't see how it would become invalid later. But as a miner tries to find a block and updates the timestamp, they would have toss the transaction out at some point.

Another objection I can think of, is that the soft fork introducing this change would have to use a transaction type that's non-standard at the moment. This would make it difficult for a non-upgraded node to broadcast such a transaction. The recipient would have to know if the sender has upgraded before communicating an address, which sounds impractical at best.

>>> Being just an expiration time, seconds-level resolution is unnecessary, and
>>> may give the wrong impression. I'd suggest either:
>>> 
>>> 1) Hour resolution - 2^24 hours = 1914 years
>>> 2) Month resolution - 2^16 months = 5458 years
>> 
>> So that's 4.8 characters for hours, or 3.2 for years, plus checksum space? The shorter the better. Perhaps one or two bits can be used to specify an exponent; a large range seems more useful than high precision. For instance a merchant doesn't care if the customer pays within 10:00:00 minutes or 10:00:01 minutes and you wouldn't care if your address is valid 50 years or 50 years and 3 minutes. This point may be mute if minute level resolution is not practical.
> 
> Note that "large range" is a requirement driven by the fact that expiry times
> will inevitably be specified absolutely, not relatively: when the range runs
> out you need to upgrade the standard. Better to use another character and use a
> range that won't run out any time soon.
> 
> This wouldn't create a need for more checksum space.

You're right, relative time makes no sense. So it would take 5 characters to get roughly two minute resolution that lasts for 100 years.

Sjors