[wip] SSZ-optimize proof (type 3) #447
Conversation
Signed-off-by: Ignacio Hagopian <[email protected]>
Signed-off-by: Ignacio Hagopian <[email protected]>
| switch { | ||
| case isInsertion(preLen, postLen): | ||
| var newValue [32]byte | ||
| copy(newValue[:], proof.PostValues[i]) |
There was a problem hiding this comment.
Note that we don't have to do the postLen == 32 and all the alignment thing.
If proof.PostValues[i] isn't 32 bytes, then copy will copy whatever amount of bytes there are (in the front of the slice). So, both in aligned and unaligned cases, the right thing will happen.
Same applies to other cases.
| ) | ||
|
|
||
| for _, suffixdiff := range stemstatediff.SuffixDiffs { | ||
| if /* len(suffixdiff.NewValue) > 0 - this only works for a slice */ suffixdiff.NewValue != nil { |
There was a problem hiding this comment.
This previous if (that was kept in Type 2) was removed. This can also be removed in Type 2 as I mentioned in my review in that PR.
Both in Inserts and Updates, we can't have "nil" or empty values in New.
| suffixes := make([]byte, 0, len(ssd.Updates)+len(ssd.Inserts)) | ||
| for i := range ssd.Updates { | ||
| suffixes = append(suffixes, ssd.Updates[i].Suffix) | ||
| } | ||
| for i := range ssd.Inserts { | ||
| suffixes = append(suffixes, ssd.Inserts[i].Suffix) | ||
| } | ||
| sort.Slice(suffixes, func(i, j int) bool { return suffixes[i] < suffixes[j] }) |
There was a problem hiding this comment.
I think in Type 2 there might be a bug since it's only inserting Updates, but prob we need both Updates+Inserts as I did here?
I also sort them since I'd guess that's expected.
There was a problem hiding this comment.
that's possible, I'm only focusing on the SSZ encoding at the moment. We'll pick the best method and fix it if need be.
| ret[i].Updates = make([]UpdateDiff, len(sd[i].Updates)) | ||
| for j := range sd[i].Updates { | ||
| ret[i].Updates[j].Suffix = sd[i].Updates[j].Suffix | ||
| ret[i].Updates[j].Current = sd[i].Updates[j].Current |
There was a problem hiding this comment.
Since we stop having []byte, many of if len(xxx) == 0 cases and similar are gone. We can always copy-by-value.
This PR is grounded on Type 2, but we encapsulate each fields in their own containers.
This should avoid some u32 references as dynamic data, leaving only one per group.
It also changed
[]byteto[32]bytesince thenilcases where already partitioned by Type 2, so those can't happen inCurrentvalues.