-
v0.0.13 Stable
released this
2024-08-26 21:24:21 +00:00 | 226 commits to main since this releaseBug fixes:
- Previously ConflictSet.h was missing a precondition that versions passed to
check
andsetOldestVersion
must be <= the last version passed toaddWrites
. See https://blog.weaselab.dev/conflict-set-advance-version-even-if-no-writes/
Downloads
- Previously ConflictSet.h was missing a precondition that versions passed to
-
v0.0.12 Stable
released this
2024-08-20 23:58:03 +00:00 | 238 commits to main since this releasePerformance bug fixes:
- GC could run too fast with write-range-heavy workloads
Performance improvements:
- Lowered instruction counts in write paths
- Lower memory usage for Node3 and Node48 (slightly)
Test improvements:
- Accumulate fuzz corpus instead of replacing it (will still need to regen corpus when the test itself changes). The idea is even if a test doesn't add coverage, the fact that it marginally improved coverage in an earlier implementation still makes it "interesting"
- Enforce coverage of avx512f-specific routines
Downloads
-
v0.0.11 Stable
released this
2024-08-08 19:08:34 +00:00 | 291 commits to main since this releaseDisable frame pointers and debug symbols on macos to work around a clang crash
Downloads
-
v0.0.10 Stable
released this
2024-08-08 18:25:57 +00:00 | 294 commits to main since this releaseNotable changes since v0.0.9
- Many minor performance improvements
- Improved test coverage
- Add a new server-like benchmark
Downloads
-
v0.0.9 Stable
released this
2024-07-23 00:24:39 +00:00 | 365 commits to main since this releaseBug fixes
- Using more than one conflict set interleaved in the same thread could cause wrong results (
fb9f5ce6f4
). See https://blog.weaselab.dev/a-bug-finds-its-way-into-a-release/ for the postmortem
Performance fixes
- GC was not outpacing writes (
4182d904c5
)
Features
- Add an experimental interface for getting internal metrics from a conflict set. The specific names and meanings of the metrics are considered implementation details, but may be interesting to curious users
Caveats
- Users may notice a small performance regression from maintaining metrics (even if they never inspect the metrics)
Downloads
- Using more than one conflict set interleaved in the same thread could cause wrong results (
-
v0.0.8 Stable
released this
2024-07-11 01:56:29 +00:00 | 403 commits to main since this releaseNotable changes from v0.0.7
addWrites
no longer has a precondition that the writes are canonical
Downloads
-
v0.0.7 Stable
released this
2024-07-08 23:11:32 +00:00 | 410 commits to main since this releaseNotable changes from v0.0.6
- Versions must now always be >= 0
- We no longer supports tracking a span of more than two billion versions
- Significantly improved performance for checking range reads
- Improved worst-case memory usage from 144 bytes per key to 112 bytes per key
Downloads
-
v0.0.6 Stable
released this
2024-06-12 21:06:42 +00:00 | 500 commits to main since this releaseNotable changes from v0.0.5
- paper will now be included in releases page
- improve convenience of using FetchContent from cmake. This is not officially supported yet.
Downloads
-
v0.0.5 Stable
released this
2024-06-11 20:11:38 +00:00 | 505 commits to main since this releaseNotable changes from v0.0.4
- Explicitly allow writeVersion to be non-decreasing
- Fix build for Apple Silicon with llvm 18
Downloads
-
v0.0.4 Stable
released this
2024-04-18 21:31:48 +00:00 | 517 commits to main since this releaseNotable changes from v0.0.3
- We now target macos 11.0
- Tests now have 100% line coverage
Downloads