When the transaction offset is wrong and doesn't parse we turn the
generic runtime_exception into one that goes back to the API caller.
Also catch those generic exceptions on the server level to avoid this
kind of error from disconnecting a peer.
This implements the sendTransaction command using the async framework,
allowing clients to benefit from multi-core systems to process
transactions in parallel when being submitted to the Hub.
The current design of the API is that all questions are answered in-
order. This implies that handling of one connection will also just use
one thread.
If we want to improve performance of things like "sendTransaction" we
want an async option, which runs the action and generates the answer
when its done in a separate thread (a new one for each async-request).
This breaks the in-order reply sequence for clients and as such it
requires them to set a new header-option to enable this.
When you request matching transactions from the mempool for certain
addreses, the API now returns the matched output-index as a separate
tag, which makes it easier to, for instance, use the 'amount' field in
the reply.
The mempool search now shares its serialization code with the blockchain
transaction fetcher, allowing all the filtering and include flags to work
here too.
This also fixes several bugs and reverts the mempool API change.
Additionally I return the FirstSeenTime for each mempool entry.
There is a slight API breakage in the LiveTransactions service, which as
far as I know is still unused outside of Flowee.
These enum values changed:
Api::LiveTransactions::
Transaction 20 => 25
UnspentState 23 => 26
OutputScript 24 => 23
The reason for this change is to have the Blockchain namespace in
sync with the LiveTransactions namespace for similar features.
This supplies a push notification on txid becoming known to the Hub.
Additionally this changes its close relative the AddressMonitorService
to use a different tag for the transaction data and the
double-spend-proof data so as to make it obvious which one is being
sent.
From now on, make sure that the minor always has at least 2 digits
which allows us to use string compare of versions even if we have
more than 9 releases in a year.
The API call for GetBlock has a filter-on-address functionality which is
now ported to no longer use ripe160 addresses but instead uses output-
script hashes.
This avoids problems for transactions not using p2pkh and generally is a
cleaner solution.
This also adds a unit test to test this feature.
Remove addresses of type ripe160 to be used in the API of AddresMonitor
API service. Instead use a sha256 hash of the entire output script.
This is mostly internal changes not affecting the operations much.
The usage of a ripe160 for bitcoin addresses in the API and in the
Indexer loses some info, specifically what kind of script it is.
Additionally not all types of scripts fit this mold. At best that means
its not future-proof.
This adds a method to the API in order to select from a Tx the hashed
outscript (thats singlehashed sha256) and refactor the address indexer
to use that instead of the ripe160 address.
The API enums broke a little, so I used the opportunity to break it a
lot and clean up the enums in order to make them more future-proof.
But, yeah, software from before this commit is protocol incompatible
with software after this commit.
In the address monitor service this detects specifically if a double
spend detected by the Hub is applicable for our subscriber, and if it is
then a Message will be sent to the subscriber.
This introduces a new message type DoubleSpendFound
and changes the API for the TransactionFound message while dropping the
TransactionRejected message.
This is an overload of message() specifically meant to build
replies to incoming messages.
The behaviour changed here is that incoming messages with
headers other than "RequestId" now get copied to the outgoing
message by default.
This also includes a little blockchain of 114 blocks (the first 100 just
coinbase to reach coin-maturity) with transactions generated by the
txVulcano. So lots of outputs.
I adjusted the BlackBoxTest to load this data into any hub with ease.
The old algorithm could end up doubling the size every time it
ran out, which eventually consumes all memory.
This way we reset to the user-default size when useful and increase
only based on need.
This means that the programmer needs to make a better estimation on
maximum size of messages, but it avoids behavior similar to memory
leaks.
Double check our message-calculation is proper by checking the result.
Notice that asserts and logDebug both are turned off in release builds,
so this is just added to debug builds.
tomFlowee
Justaphf