39 lines
1.7 KiB
Markdown
39 lines
1.7 KiB
Markdown
# Test format: shuffling
|
|
|
|
The runner of the Shuffling test type has only one handler: `core`.
|
|
|
|
However, this does not mean that testing is limited.
|
|
Clients may take different approaches to shuffling, for optimizing,
|
|
and supporting advanced lookup behavior back in older history.
|
|
|
|
For implementers, possible test runners implementing testing can include:
|
|
1) Just test permute-index, run it for each index `i` in `range(count)`, and check against expected `mapping[i]` (default spec implementation).
|
|
2) Test un-permute-index (the reverse lookup; implemented by running the shuffling rounds in reverse, from `round_count-1` to `0`).
|
|
3) Test the optimized complete shuffle, where all indices are shuffled at once; test output in one go.
|
|
4) Test complete shuffle in reverse (reverse rounds, same as #2).
|
|
|
|
## Test case format
|
|
|
|
### `mapping.yaml`
|
|
|
|
```yaml
|
|
seed: bytes32
|
|
count: int
|
|
mapping: List[int]
|
|
```
|
|
|
|
- The `bytes32` is encoded as a string, hexadecimal encoding, prefixed with `0x`.
|
|
- Integers are validator indices. These are `uint64`, but realistically they are not as big.
|
|
|
|
The `count` specifies the validator registry size. One should compute the shuffling for indices `0, 1, 2, 3, ..., count (exclusive)`.
|
|
|
|
The `seed` is the raw shuffling seed, passed to permute-index (or optimized shuffling approach).
|
|
|
|
The `mapping` is a look up array, constructed as `[spec.compute_shuffled_index(i, count, seed) for i in range(count)]`
|
|
I.e. `mapping[i]` is the shuffled location of `i`.
|
|
|
|
## Condition
|
|
|
|
The resulting list should match the expected output after shuffling the implied input, using the given `seed`.
|
|
The output is checked using the `mapping`, based on the shuffling test type (e.g. can be backwards shuffling).
|