Repository navigation
Move aggregates out of vortex-array into vortex-aggregates - #10769
Draft
robert3005 wants to merge 6 commits into
Draft
robert3005 wants to merge 6 commits into
robert3005 wants to merge 6 commits into
Conversation
robert3005
added this pull request to stack #10770
October 24, 2026 12:37
robert3005
force-pushed
the
rk/split-array-04-aggregates
branch
from
October 24, 2026 13:40
350aebd to
4c3f491
Compare
vortex-array keeps the aggregate framework (vtables, accumulators, session
registry, kernels) and the aggregates that back array statistics: min/max,
sum, null/NaN counts, sortedness, constancy, uncompressed size, the all_*
predicates and all_non_distinct. These are wired into `Stat`,
`StatsSetRef::compute_stat` and array validation, so they cannot move up
without first making statistics computation session-dispatched.
Everything else moves to a new `vortex-aggregates` crate: `count`, `first`,
`last`, `mean`, `bounded_min`, `bounded_max` and the `combined` adapter,
together with the tests and benches that exercise them.
`vortex_aggregates::initialize` registers BoundedMin/BoundedMax/First/Last
and the Count grouped kernel, which `AggregateFnSession::default` used to do.
It is called from `vortex_file::register_default_encodings` and from
`VortexSessionDefault`, so sessions built through either keep writing the
bounded string min/max zone stats. Sessions built only from
`vortex_array::array_session()` no longer know these aggregates.
`Min`/`Max::can_satisfy` now recognise the bounded variants by id instead of
by type. A few helpers the moved code needs become public:
`columnar_min_max`, `sum_decimal_dtype`, `DecimalValue::{rescale_i256,
try_from_i256}`, `Accumulator::{dtypes, empty_partial, fold_partial}`, and
the `partial_ord` module (doc-hidden).
Signed-off-by: Claude <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j
…allback
Offset validation (List, ListView, VarBin), `PrimitiveArray::narrow`, the
same-width integer cast range check, ListView's referenced element bounds,
and the debug sortedness checks in `Patches` called the `min_max` and
`is_sorted` aggregates directly, often through `legacy_session()`. That
ties core array construction to aggregates being available in core, so
they could not be resolved by session.
They now go through `arrays::primitive::{integer_bounds, integer_is_sorted,
integer_is_strict_sorted}`, which read the `Min`, `Max`, `IsSorted` and
`IsStrictSorted` statistics: cached when present, otherwise computed by the
session's aggregates and any kernel the encoding provides. Only when a
statistic has no answer, because the session registers no aggregate for
it, is the array executed to a `PrimitiveArray` and scanned once, caching
`Min`, `Max` and `NullCount`. Validation therefore works in every session,
and still benefits from statistics where they exist. Valid counts come from
the validity bitmap and ListView sizes are summed directly.
Signed-off-by: Claude <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j
`StatsSetRef::compute_stat` called the `min_max`, `sum`, `is_sorted`, `is_constant`, `nan_count` and `uncompressed_size_in_bytes` functions directly. It now resolves the aggregate backing each statistic from `ctx.session()`'s `AggregateFnSession` and runs its accumulator, so the aggregate implementations can be registered from any crate. A session with no aggregate registered for a statistic cannot compute it. - New `aggregate_fn::stat_aggregates` holds the IDs of the aggregates backing statistics, the options types those statistics bind (`UncompressedSizeOpts`, `IsSortedOptions`), and the dtype rules (`sum_return_dtype`, `supports_uncompressed_size_in_bytes`) that both the aggregates and `Stat::dtype` use. The old paths re-export them. - `Stat` gains `aggregate_fn_id`, `aggregate_fn_options` and `resolve_aggregate_fn`. `Stat::from_aggregate_fn` matches by ID and serialized options instead of by vtable type, and `StatFn` compares IDs. - `IsSorted` and `IsConstant` are now serializable, so they can be resolved by ID. - The whole-array shortcuts of `is_sorted` and `is_constant` (cached statistics, validity-only answers, cached `min == max`) move into each aggregate's `try_accumulate`, so every caller of the aggregate gets them. - NaN-skipping `Min` and `Max` accumulate through `min_max`, which caches both extrema on the batch, as `compute_stat` did before. `compute_stat` caches sortedness and constancy, which have no stat slot of their own, the same way `is_sorted`/`is_constant` did. Signed-off-by: Claude <noreply@anthropic.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j
robert3005
force-pushed
the
rk/split-array-04-aggregates
branch
from
October 25, 2026 06:10
00ea291 to
d066dd6
Compare
vortex-array now keeps only the aggregate function framework and `all_non_distinct`, which `assert_arrays_eq!` uses to compare arrays. The other concrete aggregates (min/max, sum, counts, sortedness, constancy, uncompressed size and the rest), their Dict kernels, the grouped encoding kernels and their benches move to vortex-aggregates, whose `initialize` registers them on a session. `Stat::aggregate_fn` is replaced by `vortex_aggregates::stat_aggregate_fn`. `ExtensionStoragePushdown` stays in vortex-array. Pushing `is_sorted` and `min_max` down into extension storage becomes `vortex_aggregates::register_storage_ordered`, which `initialize` calls for the temporal types and `vortex_uuid::initialize` calls for UUIDs. Validation and encoding paths that must work with any session no longer go through the registry: integer arrays use `integer_bounds` (which now also reports strict sortedness and caches Min, Max and NullCount), and boolean arrays use the new `bool_bounds`. This covers validity, RunEnd ends, bit-packing, FoR, datetime-parts comparison and the compressor's integer statistics. Tests of aggregate behaviour move alongside the aggregates. Core tests use a test-only row-count aggregate and run the aggregate consistency checks only when aggregates are registered. Test sessions that compute statistics register the aggregates. Signed-off-by: Claude <noreply@anthropic.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j Signed-off-by: Claude <noreply@anthropic.com>
`integer_bounds` read the `Min` and `Max` statistics separately, which is two passes over a cold array. It now resolves the `min_max` aggregate from the session and computes both extrema in one pass, caching `Min`, `Max` and `NullCount` as the separate statistics would. Exact cached statistics are still read first, and the primitive scan remains the fallback when the session registers no `min_max` aggregate. `stat_aggregates::MIN_MAX_ID` names the aggregate in core, and `MinMax` now serializes its `NumericalAggregateOpts` like `Min` and `Max`, so the session can resolve it by ID. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j Signed-off-by: Claude <noreply@anthropic.com>
`valid_count` uses an exact cached `NullCount` first (`len - null count`). Otherwise, for array-backed validity, it now reads the validity array's `Sum` statistic, which is the number of valid rows and can be cached or computed by an encoding kernel, as develop did with the `sum` aggregate. It counts the bitmap only when the session registers no `Sum` aggregate. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ps4mp7KF5c4p66aoSoXx5j Signed-off-by: Claude <noreply@anthropic.com>
robert3005
force-pushed
the
rk/split-array-04-aggregates
branch
from
October 25, 2026 06:42
d066dd6 to
87719b7
Compare
This branch has not been deployed
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
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.
No description provided.