Create NOTES.md

NOTES.md

@@ -0,0 +1,45 @@
+### General
+
+- Reduce Qora2 down to core blockchain node with RESTful API access. 
+Other libraries can process name-storage data into websites, or provide web-based wallet UI, etc.
+
+- Trying to reduce number of external dependencies where possible, e.g. avoiding heavy-weight ORM like Hive, Hibernate, etc.
+
+- Trying to reduce duplicated code, especially across transactions with lots of serialisation and signature generation.
+
+- Transaction signatures should really be generated after creating the transaction,
+compared to the old style of generating signature first (and throw-away transaction in the process) 
+then using signature when creating actual transaction!
+
+- Trying to keep most of the source structure, naming, code paths, etc. similar to old Qora to reduce brain load!
+
+- More comments/javadoc
+
+- More JUnit tests
+
+### Differences due to switching from MapDB to HSQLDB
+
+- We might need to maintain more mappings in the database, e.g. Qora address to/from public key,
+as previously public key objects could be stored directly in MapDB.
+
+- The new database tried to store the data in "rawest" form, i.e. raw ```byte[]``` signatures, not base58-encoded.
+
+- The ```Transactions``` table contains ```creator``` column, which duplicates various child table columns, 
+e.g. ```PaymentTransactions.sender```, 
+so that all transactions by a specific Qora account can be quickly found without scanning all child tables.
+
+- Trying to keep all SQL inside respective Java classes, 
+e.g. operations on ```Blocks``` table  only done within ```Block.java```.
+
+- Some MapDB-based objects had Java Map<> obejcts as their values. These will need to be unpicked into separate tables.
+
+### Possible gen2 refactoring already
+
+- We might need to change ```Blocks.generator``` from Qora address in VARCHAR to public key in VARBINARY,
+then derive address from public key as needed.
+
+- Ditto ```Transactions.creator``` and equivalent columns in child tables.
+
+- Extracting values from a ```ResultSet``` by column index is efficient but prone to mistakes 
+as the indexes have to be maintained by a human. There might be a better, general solution to this 
+without having to resort to importing an external ORM library. Maybe simply ```value = resultSet.getInt(columnIndex++)```