ledger: convert FC unmarshalled nil value to empty byte slice on DB write KVs

[ahangsu]

Summary

This is an improvement for DB on write, though it does not have effect on current sqlite3, it might matter if we take new DB backend (some Key-value based DB).

Since encoded.KVRecordV6 is codec:,omitempty,omitemptyarray, when we have an instance of encoded.KVRecordV6, WLOG we talk about empty box, then an empty byte slice []byte{} would be unmarshalled to nil, and written to DB.

The current status is, writing to DB with a nil value tied to a key doesn't really matter, for sqlite3 takes nil and []byte{} as same empty BLOB. But for the sake of consistency, we should convert the value to []byte{}.

Also, when we recheck with the logic of a round-by-round catchup, a box deletion would incur kvmod.val = nil, and accountsNewRoundImpl in acctdelta.go would treat this as a KV pair deletion in DB.

Thus, I think it would be good to convert a nil value unmarshal result to an empty byte slice []byte{}.

Test Plan

Consider the following test:

1. bring up a tracker DB instance of version 9
2. jam a bunch of keys with nil values into table kvstore
3. ensure that the DB read for nil values gets all the keys
4. start tracker DB migration from version 9 to version 10
5. ensure that the DB read for nil values gets 0 keys, i.e., all the nil values are converted to empty byte slices.

[codecov]

Codecov Report

Merging #5225 (586ecbf) into master (14ce659) will decrease coverage by 0.03%.
The diff coverage is 44.44%.

@@            Coverage Diff             @@
##           master    #5225      +/-   ##
==========================================
- Coverage   53.73%   53.71%   -0.03%     
==========================================
  Files         444      444              
  Lines       55671    55680       +9     
==========================================
- Hits        29913    29906       -7     
- Misses      23426    23440      +14     
- Partials     2332     2334       +2     

Impacted Files	Coverage Δ
ledger/catchupaccessor.go	62.94% <0.00%> (-0.21%)	⬇️
ledger/store/trackerdb/sqlitedriver/trackerdbV2.go	0.00% <0.00%> (ø)
ledger/store/trackerdb/sqlitedriver/schema.go	47.54% <100.00%> (+0.23%)	⬆️
ledger/store/trackerdb/sqlitedriver/testing.go	51.02% <100.00%> (+2.08%)	⬆️

... and 6 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[cce]

Do we want to do the migration? I guess it is probably cheap to do because there aren't a lot of KVs out in the wild yet?

[ahangsu]

I am not really sure, for I am unaware of how many nil values out there. But the migration should be cheap, for I guess the nil values in the table caused by FC should be sparse. (just guessing)

[jannotti]

Looks good, thanks.