This should speed up synchronization / restoration of large wallets.
Wallets are written only when they switch to up_to_date state, or
when stop_threads() is called when closing the daemon, or when
a command line command finishes.
Bucketing is generalization of coin chooser logic that makes it easy
to implement other algorithms.
- Put core coin chooser functionality in base class.
- Specialize derived class to implement classic electrum algorithm of
oldest coins first. One bucket per output.
No intended change in behaviour.
Coin chooser now sorts the coins as it wants; remove redundant sorting
from get_spendable_coins().
This has several advantages. Fee calculation is now very fast,
as we don't need to keep reserializing the tx. Another is that
we can reason about the fees after adding a change output without
having to add it, recalculate the tx fee, and remove it again.
Proper fix for #1525.
Using python's GC module, I've verified that the daemon, when running,
now releases all verifiers, synchronizers and wallets - all the resources
we care about releasing.
The main problem is, I think, that the synchronizer and wallet
still exist in the daemon process, and updates to things like
TXI and TXO are made but never saved (as client exit is what
saves the wallet).
I suspect fixing the lingering objects is hard; this is a short
term fix to ensure that when internal wallet state is updated,
the wallet is written to disk, so later daemon clients pick up
the correct state.
Previously the verifier job would scan all transactions in
unverified_tx each time it ran.
Nothing was ever removed from this map; it would essentially
be the full set of transactions.
As the job runs about 10 times a second, for a wallet with 500 txs
this would be 5,000 useless loops a second.
This patch makes unverified_tx be simply the set of confirmed
transactions that haven't yet been verified. txs are added once
confirmed, and removed once verified. Hence it will almost always be
empty.
Neil Booth
ThomasV
Chris Beaven
ThomasV