starknet_patricia: use OnceLock for write-once filled-tree output maps

[yoavGrs]

The filled-tree output maps are written exactly once per node and then
reclaimed, yet they used Mutex<Option<T>> and locked on every write and on
every reclaim - even though after Arc::into_inner the map is uniquely owned
and can never be contended.

Replace the output maps with OnceLock<T>, the primitive that matches this
write-once access pattern: write_to_output_map becomes a lock-free set
(a failed set is the double-update error), and reclaiming via
OnceLock::into_inner drops the locking entirely. Pre-allocate the collected
map with the known capacity. Leaf::Output gains a Sync bound, required for
OnceLock<L::Output>: Sync. The leaf input map keeps Mutex<Option<_>>,
since its values are moved out (take-once), not written.

Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

[reviewable-StarkWare]

This change is 

[yoavGrs]

• starknet_patricia: collapse the let-else + re-match in node_from_edge_data into a single match #14623
• starknet_patricia: use OnceLock for write-once filled-tree output maps #14622 👈 (View in Graphite)
• starknet_patricia: assert unmodified-subtree binary child is present in the skeleton #14614
• starknet_patricia: extract finalize_binary_child helper from node_from_binary_data #14613
• starknet_patricia: thread the leaf output map separately from LeafSource #14600
• starknet_patricia: collapse leaf handling to a single match #14583
• starknet_patricia: introduce LeafSource enum for filled-tree leaf handling #14582
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Medium Risk
Touches core merkle tree fill concurrency and tightens the Leaf::Output trait bound; behavior should be equivalent but double-write and missing-placeholder semantics now depend on OnceLock instead of mutex take.

Overview
Filled-tree and leaf output maps now use OnceLock<T> instead of Mutex<Option<T>>, matching the write-once-per-slot pattern during parallel tree fill.

write_to_output_map uses lock-free OnceLock::set (a failed set still maps to DoubleUpdate). Final assembly goes through renamed collect_output_map, which uses OnceLock::into_inner without mutexes and pre-sizes the result HashMap. Leaf inputs still use Mutex<Option<_>> for take-once consumption.

The Leaf::Output associated type gains a Sync bound so shared OnceLock<L::Output> slots are valid across concurrent tasks.

^{Reviewed by Cursor Bugbot for commit e5f7ec4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[nimrod-starkware]

@nimrod-starkware reviewed 2 files and all commit messages, and made 1 comment.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on yoavGrs).

---

crates/starknet_patricia/src/patricia_merkle_tree/filled_tree/tree.rs line 112 at r1 (raw file):

                    existing_value_as_string: format!("{existing_node:?}"),
                }
            }),

I think that should work, the error is the value that was already set

Suggestion:

            Some(slot) => slot.set(output).map_err(|exisitng_node| {
                FilledTreeError::DoubleUpdate {
                    index,
                    existing_value_as_string: format!("{existing_node:?}"),
                }
            }),

[yoavGrs]

@yoavGrs made 1 comment.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on nimrod-starkware).

---

crates/starknet_patricia/src/patricia_merkle_tree/filled_tree/tree.rs line 112 at r1 (raw file):

Previously, nimrod-starkware wrote…

I think that should work, the error is the value that was already set

The Err contains the value you just tried to set (the rejected output), not the value already
stored in the lock.

[nimrod-starkware]

@nimrod-starkware made 1 comment.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on yoavGrs).

---

crates/starknet_patricia/src/patricia_merkle_tree/filled_tree/tree.rs line 112 at r1 (raw file):

Previously, yoavGrs wrote…

The Err contains the value you just tried to set (the rejected output), not the value already
stored in the lock.

try_insert returns a reference to the stored value. you can use that

[nimrod-starkware]

@nimrod-starkware made 1 comment and resolved 1 discussion.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on yoavGrs).

---

crates/starknet_patricia/src/patricia_merkle_tree/filled_tree/tree.rs line 112 at r1 (raw file):

Previously, nimrod-starkware wrote…

try_insert returns a reference to the stored value. you can use that

nvm it's nightly only, unblocking

[gkaempfer]

Security scan complete — no issues detected.

---

Generated by Claude Code