Bringing over lessons learned from the Zcon1 app. There are a lot of changes to make, this is just the beginning of what was necessary to support that app.
Turns out JNI classes are hard to mock due to the way sytem libraries
are loaded. So it became easier to put the JniConverter class behind an
interface. Once that refactor was necessary, it was a good time to
update the name of this class because it retained its original PoC name
for far too long.
Never start downloading blocks prior to sapling activation height, only allow firstrun when dataDb is empty and also add birthday support. Also added corrections to isFirstRun logic.
After experiencing several issues that make it more difficult to test send behavior, including the amount of time required to wait for blocks to get mined when testnet is slow,
it became obvious that it was time to investigate mocking. Most of the behavior in the SDK is driven by channels so the mock only has to focus on putting useful data in the
expected channels at the right time. One tradeoff here was the need to make all the synchronizer properties private, that way any implementation can achieve compatability
without necessasily leveraging the same combinations of building blocks. This tradeoff felt acceptable given that these dependencies can be injected and available as singletons,
if needed. This also had a side effect of elevating several channels into the Synchronizer interface, rather than reaching into the synchronizer to directly access those dependencies.
Another benefit is that it's now easier to see what matters most to the app, particularly which channels are essential.
Send is now functional and shows up in active transactions.
This involved:
- reducing the exposure of the seed
- consistently using Ints for blockheight everywhere to match zcash
- adding the use of spending keys
- adding account initialization on startup
- using accounts but defaulting to account 0
- internalizing birthday so we no longer need a reference outside of the library
- enabling cancellation during send
- cleanup active transaction manager
When a transaction is sent it transitions through a lifecycle, beginning with creating the raw transaction and ending with it being mined and added to the blockchain
The synchronizer now primarily collaborates with a downloader, processor and repository; each with a more focused set of responsibilities.
The downloader streams blocks into a channel, the processor saves blocks from that channel and scans for transactions, the repository
exposes transaction change events.
Kevin Gorham
Kevin Gorham