[Account Storage Maps] Reduce storage register reads when using StorageFormatV2Enabled

[fxamacker]

Updates #3584

Currently in PR #3678, all domain registers are being read for a scenario that does not require it. This PR reduces register reads as discussed.

Problem

In some cases, we can reduce 10 register reads to only 2 register reads. 🚀

When StorageFormatV2 feature flag is enabled, Storage.GetDomainStorageMap() reads all domain registers to determine if account is new account or in v1 format.

These register reads are unnecessary when:

• the given domain register exists, or
• the given domain register doesn't exist and createIfNotExists param is false.

Solution

This PR modifies Storage.GetDomainStorageMap() to read the given domain register first (before iterating all domain registers) to reduce register reads.

• If the given domain register exists, then account format is v1 and no more register reading is needed.

• If the given domain register doesn't exist and createIfNotExists is false, then nil domain storage map is returned and no more register reading is needed.

• Otherwise, account format (v1 account or new account) is determined by iterating all domain registers.

Reduced Reads

When using StorageFormatV2Enabled:

For new accounts (neither v1 or v2 yet):

• If domain storage map doesn't exist and createIfNotExists is false, old approach has 10 unique register reads, while new approach has 2 unique register reads.

For v1 accounts:

• If domain storage map doesn't exist and createIfNotExists is false, old approach has 4 unique register reads in tests, while new approach has 2 unique register reads.

Caveat

This change increases code complexity in order to avoid reading all domain registers for this scenario. I added more comments to offset some of this tradeoff. For more context, see comment thread at #3678.

UPDATE:

I can add more tests for register reading after sync with Cadence team about this approach.
I added more tests in case we don't get chance to sync with Cadence team this week.

---

• Targeted PR against master branch
• Linked to Github issue with discussion and accepted design OR link to spec that describes this work
• Code follows the standards mentioned here
• Updated relevant documentation
• Re-reviewed Files changed in the Github PR explorer
• Added appropriate labels

[github-actions]

Cadence Benchstat comparison

This branch with compared with the base branch onflow:feature/combine-domain-payloads-and-domain-storage-maps commit 157d322
The command for i in {1..N}; do go test ./... -run=XXX -bench=. -benchmem -shuffle=on; done was used.
Bench tests were run a total of 7 times on each branch.

Collapsed results for better readability

[turbolent]

Nice work improving this!

The changes look good, I left some very minor suggestions.

I'm having a bit of a harder time understanding when/where we have unnecessary register reads currently, and how the changes of the PR remove the unnecessary register reads, especially because it seems to add an additional read in one of the existing tests.

Could we maybe add a test case to feature/combine-domain-payloads-and-domain-storage-maps before this PR that demonstrates the inefficiency (unnecessary reads), and then this PR could would show more clearly how the unnecessary read(s) are removed?

[fxamacker]

@turbolent there were some questions related to GetDomainStorageMap() and getAccountStorageFormat(), so I refactored the code to improve readability.

Also, the updated storage_tests.go shows the reduced register reads (expectedReadsSet).

PTAL. 🙏

[turbolent]

Great work! The optimization of reducing the register reads is really clearly evident in the tests now, well done 👏