Now prunes tries properly (and more aggressively!) without hack #97
Conversation
- However it will only not hash the child if the key matches the key piece of the extension.
- Fixes overall deal with calling the new pruning query in `mpt_trie` and changing the order of calls when generating proof IR.
- Realized that we needed a bit more logic to get the key to the remaining child after a branch collapse. - Ended up adding a bool to `TriePathIter`.
- Also includes one core fix that got left out.
github-actions
bot
added
crate: trace_decoder
Not able? |
| @@ -26,6 +26,11 @@ pub type Nibble = u8; | |||
| /// Used for the internal representation of a sequence of nibbles. | |||
| pub type NibblesIntern = U512; | |||
There was a problem hiding this comment.
Just noticed it now, but why do we need U512? For the 65-Nibbles edge case? We probably would want to refactor it at some point, seems a bit wasteful to leverage U512 everywhere here (though out of scope for this PR).
There was a problem hiding this comment.
Yeah exactly. It's for the 65 Nibble edge case.
I don't think I have a Github issue open for this (there was one in Jira), so I'll create one. Also worth having a discussion on the best way to implement this. If we care about performance, using a byte array might perform worse. I know H256 uses a [u8; 32], but U256s use [u64; 4]. If I had to guess, since H256 don't support common operators (eg. +, -), they probably found that working with u64s allowed them to implement a lot of operations with 64-bit instructions (vs. doing a bunch of operations to individual bytes). Probably worth talking to @dvdplm about this, as IIRC he worked on ethereum-types a good amount.
There was a problem hiding this comment.
Created an issue for this (#104).
Robin's suggested changes for PR #97 Co-authored-by: Robin Salen <30937548+Nashtare@users.noreply.github.com>
Sorry just really want to correct myself as I accidentally negated myself here. 😅 There's just one block that does not work so far (out of ~10). |
This PR is the last piece to tie together all of the pruning optimizations/fixes that have been going on over the past month. I ended up having to cherry-pick from a testing branch, so apologies if this is a bit messy (eg.
Requested PR changes for #39somehow never got merged and was on the tip of another in progress branch that I needed, but I'm just going to leave that here for now).The critical bits in this PR are:
into_txn_proof_gen_irhas changed in order to usedelta_out(just contains info on collapsed nodes) when creating partial tries. I'm going to admit upfront that this section got a lot more complex and will get a nice refactor pretty soon in an upcoming PR.apply_deltas_to_trie_state. It really had no reason to be in here.Almost all blocks that I've tested are not able to generate proofs. I was hoping to say all, but
19240780currently fails.