feat(fuzz): prioritize favored corpus entries - #15818
Conversation
…ur/favored-power-schedule-1784381979
mablr
left a comment
mablr
left a comment
There was a problem hiding this comment.
We should probably build the mutation schedule once per new_inputs call and reuse it for both donor picks, rather than rebuilding edge frequencies and weights twice.
The edge-frequency map also uses trusted integer keys, so switching it to FxHashMap (or avoiding hashing entirely) appears worthwhile.
In the same fixed-seed Nerite local run, snapshot reuse improved throughput from 604 to 988 tx/s, and snapshot reuse plus FxHashMap reached 2,867 tx/s.
|
derek bench fuzz compare-ref=master timeout=600 |
|
cc @mablr bench (fuzz) event published. Results will be reported separately. View publisher run Config: subcommand: |
…ur/favored-power-schedule-1784381979 # Conflicts: # Cargo.toml # crates/evm/evm/src/executors/corpus.rs
|
derek bench invariant timeout=10800 |
|
cc @grandizzy bench (invariant) event published. Results will be reported separately. View publisher run Config: subcommand: |
3 participants
Stacked on #15817.
Makes the coverage minset affect donor selection, not just eviction. The scheduler gives energy to favored entries, rare edges, and recent coverage yield, while retaining bounded uniform and disk exploration.
Adds a focused scheduling regression test.
Prompted by: @0xalpharush