Patches have correct dtype by construction instead of normalised during array construction#8626
Open
robert3005 wants to merge 3 commits into
Open
Patches have correct dtype by construction instead of normalised during array construction#8626robert3005 wants to merge 3 commits into
robert3005 wants to merge 3 commits into
Conversation
`Patches::cast_values` was a crutch that papered over patch values not matching the parent array's dtype, and ALPRD/Sparse each carried a construction-time normalization step (`canonicalize_patches`, `normalize_patches_dtype`) to fix patches up after the fact. Remove all three: patch-producing operations now yield the correct dtype directly, and constructors assert rather than normalize. - `Patches::mask` widens the surviving values to nullable. Masking an array always yields a nullable result, so the masked patches now carry the correct dtype and callers (ALP mask) no longer re-cast them. - ALP take/mask: `Patches::take`/`mask` already produce values matching the taken/masked encoded array, so build via `ALP::try_new`, which asserts the patch dtype instead of forcing it. - bitpacking take: `PrimitiveArray::patch` (via `Validity::patch`) already asserts matching nullability, so no cast is needed. - ALPRD: drop `canonicalize_patches`. Exceptions are always the non-nullable left-parts dtype; `validate_parts` now requires exactly that, `take` casts the taken exceptions back to it (folded into the take call), and the other construction paths already satisfy it. Neither `ALPRD::try_new` nor `RDEncoder::encode` needs an `ExecutionCtx` anymore (the RD split does no execution); their callers that only plumbed a ctx to pass it down are cleaned up too. - Sparse: drop `normalize_patches_dtype`. Both `Sparse::try_new` and the patches-based paths now assert that the values/patches already match the fill dtype instead of casting; callers construct matching values. Signed-off-by: Robert Kruszewski <robert@spiraldb.com> Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UDa4BYuhQyRHWCUGTanVzE Signed-off-by: Robert Kruszewski <github@robertk.io>
robert3005
added
changelog/chore
github-actions
Bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 0.923x ➖ How to read Verdict and Engines
datafusion / vortex-file-compressed (0.923x ➖, 3↑ 0↓)
No file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.002x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.005x ➖, 0↑ 0↓)
datafusion / parquet (0.996x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.007x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.017x ➖, 0↑ 1↓)
duckdb / parquet (1.001x ➖, 0↑ 0↓)
File Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.991x ➖, 2↑ 5↓)
datafusion / vortex-compact (1.037x ➖, 2↑ 5↓)
datafusion / parquet (1.078x ➖, 1↑ 8↓)
datafusion / arrow (0.938x ➖, 6↑ 1↓)
duckdb / vortex-file-compressed (1.058x ➖, 0↑ 7↓)
duckdb / vortex-compact (0.985x ➖, 0↑ 0↓)
duckdb / parquet (1.083x ➖, 0↑ 6↓)
duckdb / duckdb (1.122x ❌, 0↑ 15↓)
File Size Changes (10 files changed, -0.2% overall, 4↑ 6↓)
Totals:
|
Merging this PR will degrade performance by 15.17%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | chunked_bool_canonical_into[(1000, 10)] |
16 µs | 26.4 µs | -39.51% |
| ⚡ | Simulation | rebuild_naive |
109.1 µs | 91.7 µs | +18.98% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing claude/remove-cast-values-crutch-7u8efe (9597529) with develop (79c6b0f)
Footnotes
-
4 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.043x ➖, 1↑ 7↓)
datafusion / vortex-compact (1.038x ➖, 0↑ 4↓)
datafusion / parquet (1.049x ➖, 1↑ 11↓)
duckdb / vortex-file-compressed (1.055x ➖, 0↑ 8↓)
duckdb / vortex-compact (1.047x ➖, 0↑ 8↓)
duckdb / parquet (1.026x ➖, 0↑ 2↓)
duckdb / duckdb (1.045x ➖, 0↑ 8↓)
File Size Changes (7 files changed, -0.0% overall, 4↑ 3↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.024x ➖, 0↑ 1↓)
datafusion / vortex-compact (0.956x ➖, 0↑ 0↓)
datafusion / parquet (1.136x ➖, 0↑ 3↓)
duckdb / vortex-file-compressed (0.939x ➖, 1↑ 0↓)
duckdb / vortex-compact (0.981x ➖, 0↑ 0↓)
duckdb / parquet (0.963x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) How to read Verdict and Engines
duckdb / vortex-file-compressed (1.023x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.999x ➖, 0↑ 0↓)
duckdb / parquet (1.021x ➖, 0↑ 0↓)
File Size Changes (1 files changed, +0.0% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: Clickbench Sorted on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.984x ➖, 1↑ 2↓)
datafusion / parquet (1.013x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.018x ➖, 0↑ 1↓)
duckdb / parquet (1.001x ➖, 0↑ 0↓)
duckdb / duckdb (1.000x ➖, 0↑ 0↓)
File Size Changes (201 files changed, -0.0% overall, 91↑ 110↓)
Totals:
|
robert3005
commented
Jun 29, 2026
robert3005
commented
Jun 29, 2026
robert3005
commented
Jun 29, 2026
robert3005
commented
Jun 29, 2026
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.003x ➖, 0↑ 0↓)
datafusion / parquet (0.995x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (1.028x ➖, 0↑ 7↓)
duckdb / parquet (1.000x ➖, 0↑ 0↓)
duckdb / duckdb (0.988x ➖, 0↑ 0↓)
File Size Changes (106 files changed, -0.0% overall, 49↑ 57↓)
Totals:
|
robert3005
commented
Jun 29, 2026
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.127x ❌, 0↑ 18↓)
datafusion / vortex-compact (1.110x ❌, 0↑ 12↓)
datafusion / parquet (1.116x ❌, 0↑ 15↓)
datafusion / arrow (1.188x ❌, 0↑ 22↓)
duckdb / vortex-file-compressed (1.108x ❌, 0↑ 12↓)
duckdb / vortex-compact (1.078x ➖, 0↑ 6↓)
duckdb / parquet (1.066x ➖, 0↑ 6↓)
duckdb / duckdb (1.031x ➖, 0↑ 2↓)
File Size Changes (26 files changed, +0.0% overall, 10↑ 16↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.881x ➖, 2↑ 0↓)
datafusion / vortex-compact (0.929x ➖, 0↑ 0↓)
datafusion / parquet (0.935x ➖, 3↑ 1↓)
duckdb / vortex-file-compressed (0.903x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.873x ➖, 0↑ 0↓)
duckdb / parquet (1.004x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Appian on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.060x ➖, 0↑ 2↓)
datafusion / parquet (1.037x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.030x ➖, 0↑ 0↓)
duckdb / parquet (1.023x ➖, 0↑ 0↓)
duckdb / duckdb (1.032x ➖, 0↑ 0↓)
File Size Changes (4 files changed, -0.0% overall, 1↑ 3↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.956x ➖, 2↑ 0↓)
datafusion / vortex-compact (0.875x ➖, 1↑ 0↓)
datafusion / parquet (0.952x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.890x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.827x ➖, 0↑ 0↓)
duckdb / parquet (0.824x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Random AccessVortex (geomean): 0.954x ➖ How to read Verdict and Engines
unknown / unknown (0.978x ➖, 5↑ 4↓)
|
Contributor
Benchmarks: CompressionVortex (geomean): 0.995x ➖ How to read Verdict and Engines
unknown / unknown (1.001x ➖, 2↑ 2↓)
|
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
2 participants
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.
Patches::cast_values was a crutch and sign that in too many places we were
producing wrong dtypes. This pr removes this function and fixes up dtypes where
necessary. We still need a cast in ALP-RD take as ALP-RD patches must be non
nullable