Don't all comment_objects need to exist in memory (at least with minimal data of author and permlink) anyway? The blockchain needs to check whether a post operation has a unique author/permlink so that it knows whether it is creating a new post/comment with a payout period, or just doing an edit.
So, we are talking about keeping maybe a couple more fields in comment_objects for validating nodes. And if you compare that to the amount of memory full nodes need to keep (whether the post is archived or not), that additional memory overhead is very small.
Memory access patterns for validating nodes are more of a concern, but I doubt most old posts will continue to remain active. And isn't this one of the advantages the ChainBase upgrade provided? That the database can support a much larger amount of infrequently accessed memory (since they remain on disk rather than in R AM)?
Also, I supposed it doesn't have to be indefinite. But archiving a post/comment after just 1 week is really short. (By the way, does archiving a post/comment mean that no new child comments are allowed? Because otherwise you still have the limited attention problem since new comments with active payouts can exist nested in the discussion thread of a months old post.) So what about up to 52 1-week long payout periods (which are only activated after the first payout if there is a new vote), and then a year after the post/comment was created, no new payout periods can be started?
You can compress comment objects down to HASH( author + permlink ) and check for existence, potentially using a bloom filter to detect non-existence. Memory access patterns can also be optimized.
The reality is that a single vote normally results in 0 payout and incurs a cost both at the time it is voted and later when it is rejected for insufficient payout.
We know that 99.85% of all votes (by rshares) are cost within 7 days. We also know that 99.5% of all votes (counted equally) are cast within 7 days. Actual usage shows that old posts don't get votes. In fact, we could get 99% of all votes within 3 days based upon actual data since HF 16.
This is a circular argument based on the current rule set (and UI). It can't be used to evaluate a different proposed rule set.
A similar (bogus) argument would be that comments only get 1% of rewards therefore rewarding comments isn't needed.
If it would be easier to find/filter "old" contend, then they would get MUCH more votes... The same would be true if the system would give more rewards to curators of posts older than 24 hours... So many gems not "mined" yet! Why throwing cement above our rich property ?
Old posts don't get votes because they don't feature. They are hard to find.
UI needs an advanced search page, by the way :)
Of course 99% occur in that time window.... who wants to waste today's voting power on something that will get paid out in 27 days at probably .001 SP?