Release Notes - v0.11.3
v0.11.3
Section titled “v0.11.3”Title: Improve evaluation speed and allocation efficiency
Version: v0.11.3
Release: v0.11.3
Comparison: v0.11.2…v0.11.3
Recommended Network Compatibility ⚠️
Section titled “Recommended Network Compatibility ⚠️”| Network | Compatible |
|---|---|
| mainnet | ⛔ |
| preprod-testnet | ✅ |
| preview-testnet | ✅ |
| musashi | ✅ |
This release improves evaluator speed and allocation efficiency across integer arithmetic, data construction, constant creation, and reusable evaluation resources.
Optimize Large-Operand Division
Section titled “Optimize Large-Operand Division”- Improved evaluation of large
divideIntegeroperands by calculating the quotient and remainder together and reusing a read-only constant to adjust the floor result. Tests cover floor semantics and input immutability, with about 7.5% fewer allocations and 5% less time on a 1,360-validator corpus.
Build Dynamic Integer Constants Efficiently
Section titled “Build Dynamic Integer Constants Efficiently”- Reduced allocations for dynamic
int64constants by constructing each value with one allocation. Each evaluation machine now keeps its own cache, so parallel evaluation no longer contends on a process-wide lock.
Improve Typed Data List Construction
Section titled “Improve Typed Data List Construction”- Enhanced typed list construction so adding data heads to empty
list(data)constants uses data-list representations, while adding data-pair heads tolist(pair data data)values uses data-map representations. The evaluator validates declaredProtoPairtypes and rejects mismatched types, reducing allocations by about 28% and time by 9% on a recent mainnet corpus.
Standardize Go Formatting
Section titled “Standardize Go Formatting”- Standardized Go formatting with
golinesandgci, and resolved a formatter and lint failure indata/walk.gowithout changing runtime behavior.
Retain Value Arena Chunks
Section titled “Retain Value Arena Chunks”- Preserved CEK value arena chunks across script size changes with fixed 512-slot chunks, per-arena retention limits, bounded retention, and 16-run shrink hysteresis. This reduces bytes allocated by about 75% and single-threaded time by 6% per corpus pass.
Docs authored by Doc Holiday