Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Raise the hash-table cap to 18 bits at levels 6–8, reducing collisions and retaining more useful matches. Across seven fixtures, output is 0.589% smaller at levels 6/7 and 0.274% smaller at level 8, without a size regression on any fixture at those levels.
The tradeoff is memory and workload-dependent speed. Large-input hash-head storage becomes 512 KiB, previously 64 KiB at levels 6/7 and 128 KiB at level 8. Sunrise and map tiles measured about 1.09x faster. Text timing is unstable across forks. A 2 MiB candidate saved only another 0.067 percentage points overall at level 6.
Linux JVM tests and all 70 corpus/level round trips through JDK and KFlate passed. Raw development JMH samples, sizes, and saved Kompress comparisons are in
performance/2026-09-09-table18timed/RESULTS.md. Independent PR againstdev-1.1.0; results are not additive with other matching changes.Timing calibration: unchanged baseline text varied from 60.37 to 85.43 ms across three fresh forks. Initial single-fork text ratios are provisional; compressed-size results are deterministic.