sadpandajoe commented on code in PR #43870:
URL: https://github.com/apache/superset/pull/43870#discussion_r4224320545
##########
superset/views/datasource/views.py:
##########
@@ -87,6 +95,109 @@ def _load_dataset_for_samples(
return dataset, None
+def _partition_transform_error(
+ orm_datasource: Any,
+ datasource_dict: dict[str, Any],
+) -> str | None:
+ """
+ The first column whose incoming value transform may not be stored, if any.
+
+ Only datasets carry a partition mapping, and only a transform the request
+ actually changes is worth refusing over -- re-sending a value already in
+ storage is what a GET-then-PUT client does, and a transform stored before
+ this check existed is disarmed at probe time instead.
+ """
+ incoming = [
+ column
+ for column in datasource_dict.get("columns") or []
+ if column.get("partition_value_transform")
+ ]
+ database = getattr(orm_datasource, "database", None)
+ if not incoming or database is None:
+ return None
+
+ stored = {
+ column.column_name: column.partition_value_transform
+ for column in orm_datasource.columns
+ }
+ for column in incoming:
+ transform = column["partition_value_transform"]
+ if transform == stored.get(column.get("column_name")):
+ continue
+ if reason := stored_expression_error(
Review Comment:
The legacy save only runs `stored_expression_error` on a changed transform,
while the REST path and the import path also reach `validate_transform`, where
the nondeterministic-function check is blocking. A transform such as `random()
+ :value` is saved here with a 200, and then even a description-only `PUT
/api/v1/dataset/<pk>` fails validation on the persisted transform, so the
dataset can't be saved again until the transform is edited. Should this also
reject a changed transform that `validate_transform` flags as blocking?
##########
superset/models/helpers.py:
##########
@@ -4292,36 +4389,644 @@ def dttm_sql_literal(self, dttm: datetime, col:
"TableColumn") -> str:
return f"""'{dttm.strftime("%Y-%m-%d %H:%M:%S.%f")}'"""
- def get_time_filter( # pylint: disable=too-many-arguments # noqa: C901
+ def _mirror_probe_value(self, dttm: datetime, col:
Optional["TableColumn"]) -> Any:
+ """
+ ``dttm`` in the mapped column's own stored representation.
+
+ The predicate this mirror stands in for compares the column against
+ `dttm_sql_literal(dttm, col)`, so the transform has to be probed with
+ the same value. On a column stored as an epoch integer, ``T`` is a
+ function of an integer: probing it with a `datetime` either returns
+ something unrelated to the partition keys or raises, and a raise costs
+ the dataset its pruning silently.
+
+ Only the ``python_date_format`` branches of `dttm_sql_literal` are
+ reproduced, because only those produce a *value*. `convert_dttm` yields
+ engine-specific SQL text, which cannot be bound as a parameter; its own
+ divergence from the probe is handled by widening the bounds instead
+ (`_engine_literal_resolution`).
+
+ In `dttm_sql_literal`'s own order, though: it asks ``convert_dttm``
+ first and reaches the format only when the engine answers nothing. A
+ column can carry both -- a TIMESTAMP with a ``%Y%m%d``
+ ``python_date_format``, or an entry in
+ ``python_date_format_by_column_name`` -- and applying the format
+ unconditionally probed a value the real predicate never compares.
+ ``event_time >= '2026-01-01 00:00:00'`` mirrored as
+ ``partition_col >= '20260101'``, and a January 15 row keyed
+ ``2026-01-15`` sorts below that string, so the chart lost a row the
+ filter keeps.
+
+ Where the engine does answer, the `datetime` is returned unchanged and
+ the widening path takes over -- which is exactly what already happens
+ for every temporal column that carries no format at all.
+ """
+ if col is None:
+ return dttm
+
+ if col.type and self.db_engine_spec.convert_dttm(
+ col.type, dttm, db_extra=self.db_extra
+ ):
+ # The format is unreachable in `dttm_sql_literal` for this column,
+ # so reproducing it here would describe a different bound.
+ return dttm
+
+ tf = col.python_date_format
+ if not tf and self.db_extra:
+ tf = self.db_extra.get("python_date_format_by_column_name",
{}).get(
+ col.column_name
+ )
+ if not tf:
+ return dttm
+
+ if tf in EPOCH_FORMATS:
+ dttm_tz_aware = dttm
+ if dttm_tz_aware.tzinfo is None:
+ dttm_tz_aware = dttm_tz_aware.replace(tzinfo=timezone.utc)
+ return int(dttm_tz_aware.timestamp()) * EPOCH_FORMATS[tf]
+ return dttm.strftime(tf)
+
+ def _engine_literal_resolution(
+ self, dttm: datetime, col: Optional["TableColumn"]
+ ) -> LiteralResolution:
+ """
+ How much of ``dttm`` survives into this engine's literal for ``col``.
+
+ The real predicate compares `dttm_sql_literal(dttm, col)`, so whatever
+ that literal throws away is a bound the engine never sees. A mirror
+ derived from the full precision then describes a *narrower* range than
+ the predicate it stands in for -- and narrower means it drops rows the
+ filter keeps.
+
+ Two things get thrown away in practice:
+
+ - the sub-second part. SQLite renders a timestamp with
+ ``timespec="seconds"``.
+ - the time of day entirely. Every engine spec with a ``types.Date``
+ branch renders ``dttm.date().isoformat()``, so Presto, Trino,
+ BigQuery and some two dozen others answer `DATE '2026-01-01'` for a
+ bound of 10:00 -- and `T(DATE '2026-01-01')` is below the mirror's
+ own lower bound, so the whole first day is dropped.
+
+ Detected rather than enumerated per engine, because the branch depends
+ on the column's type as well as the spec: instants differing only in
the
+ part under test render as the same text exactly when the engine has
+ thrown that part away.
+
+ The day test uses two fixed times of day rather than ``dttm`` and one
+ alternative, so a bound that happens to sit on the hour the alternative
+ uses cannot read as "no time at all". Neither probe sits at midnight on
+ purpose: SQLite renders a DATE column at exactly midnight as a bare
+ date, so a midnight probe would differ from its alternative for a
+ reason that has nothing to do with resolution.
+
+ Likewise both sub-second probes carry a non-zero remainder. Comparing
+ against ``dttm`` truncated to the second looks like the obvious test
and
+ is wrong for the same formatting-boundary reason.
+ """
+ if col is None or not col.type:
+ return LiteralResolution.FULL
+
+ def render(moment: datetime) -> Optional[str]:
+ return self.db_engine_spec.convert_dttm(
+ col.type, moment, db_extra=self.db_extra
+ )
+
+ try:
+ early = render(dttm.replace(hour=1, minute=2, second=3,
microsecond=444444))
+ late = render(
+ dttm.replace(hour=22, minute=33, second=44, microsecond=555555)
+ )
+ if early is not None and early == late:
+ return LiteralResolution.DAY
+
+ # Flooring to the second is a no-op on a bound that has no
+ # sub-second part, so there is nothing to detect.
+ if not dttm.microsecond:
+ return LiteralResolution.FULL
+
+ first = render(dttm.replace(microsecond=111111))
+ second = render(dttm.replace(microsecond=222222))
+ except Exception: # pylint: disable=broad-except # noqa: BLE001
+ return LiteralResolution.FULL
+
+ if first is not None and first == second:
+ return LiteralResolution.SECOND
+ return LiteralResolution.FULL
+
+ def _round_bounds_outward(
self,
- time_col: "TableColumn",
+ start_dttm: Optional[datetime],
+ end_dttm: Optional[datetime],
+ col: Optional["TableColumn"],
+ ) -> tuple[Optional[datetime], Optional[datetime]]:
+ """
+ The bounds widened to whatever the engine's literal actually resolves
to.
+
+ Rounding outward is never narrower than the predicate being mirrored,
+ which is the only property the mirror has to hold.
+ """
+ resolutions = {
+ self._engine_literal_resolution(bound, col)
+ for bound in (start_dttm, end_dttm)
+ if bound is not None
+ }
+
+ if LiteralResolution.DAY in resolutions:
+ # The engine compares a bare date, so the real predicate's lower
+ # bound is the start of that day and its upper bound admits the
+ # whole of the day before. Anything finer excludes rows it keeps.
+ if start_dttm is not None:
+ start_dttm = start_dttm.replace(
+ hour=0, minute=0, second=0, microsecond=0
+ )
+ if end_dttm is not None and end_dttm != end_dttm.replace(
+ hour=0, minute=0, second=0, microsecond=0
+ ):
+ end_dttm = end_dttm.replace(
+ hour=0, minute=0, second=0, microsecond=0
+ ) + timedelta(days=1)
+ elif LiteralResolution.SECOND in resolutions:
+ if start_dttm is not None:
+ start_dttm = start_dttm.replace(microsecond=0)
+ if end_dttm is not None and end_dttm.microsecond:
+ end_dttm = end_dttm.replace(microsecond=0) +
timedelta(seconds=1)
+
+ return start_dttm, end_dttm
+
+ def _literal_is_coarser_than(
+ self, dttm: datetime, col: Optional["TableColumn"]
+ ) -> bool:
+ """
+ Whether the engine's literal for ``dttm`` throws away part of it.
+
+ Unlike a range, an equality cannot be widened outward, so the only safe
+ answer when the engine compares less than the full value is to decline
+ the mirror.
+ """
+ resolution = self._engine_literal_resolution(dttm, col)
+ if resolution is LiteralResolution.DAY:
+ return dttm != dttm.replace(hour=0, minute=0, second=0,
microsecond=0)
+ if resolution is LiteralResolution.SECOND:
+ return bool(dttm.microsecond)
+ return False
+
+ def _column_literal_resolution(
+ self, col: Optional["TableColumn"]
+ ) -> LiteralResolution:
+ """
+ How much of *any* value this engine compares on ``col``.
+
+ `_engine_literal_resolution` answers about a particular instant, which
+ is what a mirror derived from one needs. Two callers have no instant to
+ ask about -- the summary the Explore glyph reads, and a filter value
+ `_instant_from_filter_value` cannot read as an instant at all -- and
+ both need the column's own answer. Through one helper so that the glyph
+ and the query path cannot disagree about which column is coarse, which
+ is the same reason `mirrorable_operators` is computed once.
+
+ A non-temporal column has no answer, and asking anyway gets a wrong
one:
+ SQLite's ``convert_dttm`` renders for ``types.String`` as well as for
the
+ date types, so a reference instant makes a ``VARCHAR`` mapped column
look
+ second-resolution -- and ``country = 'US'``, the text half of this
+ feature, reads as a value the engine compares coarsely. The per-instant
+ callers never meet this, because they arrive only with a value that
+ parsed as an instant.
+ """
+ if col is None or not col.is_dttm:
+ return LiteralResolution.FULL
+ return self._engine_literal_resolution(_RESOLUTION_REFERENCE_INSTANT,
col)
+
+ def _round_bound_outward(
+ self,
+ instant: datetime,
+ col: Optional["TableColumn"],
+ *,
+ is_upper: bool,
+ ) -> datetime:
+ """
+ One bound widened to whatever the engine resolves ``col`` to.
+
+ `_round_bounds_outward` answers for a since/until pair, because a time
+ range is the shape it was written for. A single comparison is one end
of
+ such a pair with the other end absent, and the widening is per-end
+ already -- a lower bound moves back, an upper bound forward.
+
+ Narrowing the return type is the other half of the point: a bound
handed
+ in non-`None` comes back non-`None`, and saying so here saves every
+ caller from re-establishing it.
+ """
+ start_dttm, end_dttm = self._round_bounds_outward(
+ None if is_upper else instant,
+ instant if is_upper else None,
+ col,
+ )
+ rounded = end_dttm if is_upper else start_dttm
+ return rounded if rounded is not None else instant
+
+ def _mirror_probe_input(
+ self,
+ operator: utils.FilterOperator,
+ value: Any,
+ col: Optional["TableColumn"],
+ ) -> Any:
+ """
+ ``value`` in the representation the mirror may probe the transform at.
+
+ A simple filter on a temporal column carries a *string*.
+ `filter_values_handler` rewrites only an all-digit value, which it
reads
+ as epoch milliseconds, so ``'2026-07-06 10:08:11'`` reaches here
exactly
+ as the chart author typed it and is bound as text. The engine compares
+ that text against the column under its own cast, and on a ``DATE``
+ column the cast drops the time of day: the filter keeps the whole of
+ 2026-07-06 while a mirror probed at 10:08:11 keeps one instant of it,
+ which is narrower than the predicate it stands in for.
+
+ So the text is read back into the instant it denotes -- the only way
+ `_engine_literal_resolution`, which answers about a `datetime`, reaches
+ a filter value at all -- and where the engine compares less of that
+ instant than the filter carries, what to do about it is a property of
+ the *operator*:
+
+ * a bound rounds outward, through the helper the range collector uses.
+ Reading one extra bucket at the edge is the trade this feature makes
+ everywhere else too -- see `mirror_operator` -- and the original
+ predicate is still in the query to exclude the rows that bucket adds.
+ * an equality, or one member of an ``IN`` list, has nowhere to widen
to.
+ These entries are ``AND``-ed onto the query, so "or the day either
+ side" is not something this sink can express, and the request is
+ declined with `_UNMIRRORABLE`. The query is then correct and
unpruned,
+ which is the one failure mode this feature is allowed.
+
+ An instant carrying a UTC offset is declined for every operator, a
+ bound included. Rounding one outward would first have to place it in
the
+ frame the partition column was written in, and guessing that frame is
+ how a mirror ends up a whole offset narrower than the filter -- see the
+ load-bearing assumption in
+ `superset.connectors.sqla.partition_mapping`.
+
+ A *string* `_instant_from_filter_value` cannot read at all is declined
on
+ the same terms, and for the same reason one more step back: there is no
+ instant to round. ``'07/06/2026 10:08:11'`` is not ISO 8601, so the
+ parser leaves it alone -- and Postgres casts that same text to a DATE
and
+ keeps the whole of July 6 while the probe, handed the string verbatim,
+ evaluates the transform at 10:08:11 and asks for a partition key no row
+ holds. That is exactly the narrowing the ISO-only parser was written to
+ prevent, reached through a value it declines to read, so the decline
has
+ to travel with it. Which costs a non-ISO *date* its pruning too,
because
+ telling July 6 from June 7 is the guess that started this.
+
+ Only a string. `_as_probe_input` has already run, so the engine SQL
+ `filter_values_handler` returns for an epoch-millisecond bound arrives
+ here as a `RawProbeValue` rather than text -- and that literal was
+ rendered by the engine at the engine's own resolution, so it is the one
+ thing on this path with nothing left to resolve.
+
+ Where the engine compares the whole instant, a string is returned
+ exactly as it arrived -- down to the type the probe binds and the cache
+ key it lands on, because every query that was already correct has to
+ stay byte-for-byte correct. A `datetime` is still converted, which it
+ has always needed: on a column stored as an epoch integer the transform
+ is a function of an integer, and a `datetime` is not the value the real
+ predicate compares.
+ """
+ parsed = _instant_from_filter_value(value)
+ if parsed is None:
+ if (
+ isinstance(value, str)
+ and self._column_literal_resolution(col) is not
LiteralResolution.FULL
+ ):
+ return _UNMIRRORABLE
+ return value
+
+ if (
+ self._engine_literal_resolution(parsed.replace(tzinfo=None), col)
Review Comment:
This rounds the bound for any mapped column whose engine renders datetimes,
not only temporal ones. SQLite's `convert_dttm` renders for VARCHAR as well
(the note on `_column_literal_resolution` calls this out and that helper guards
on `is_dttm`). With VARCHAR `code` and `p`, a monotonic `:value` transform, a
row `code = p = '2026-07-06T09:00:00'` and the filter `code <
'2026-07-06T10:08:11.500000'`, the bound is rounded to 10:08:12 and `p <=
'2026-07-06 10:08:12.000000'` is added; the space sorts before `T`, so the
matching row is dropped from the mirrored query. Should this skip the
resolution/rounding path for a non-temporal column the same way
`_column_literal_resolution` does?
##########
superset/commands/dataset/importers/v1/utils.py:
##########
@@ -582,6 +734,27 @@ def import_dataset( # noqa: C901
if dataset.id is None:
db.session.flush()
+ # A bundle can name a mapped column and still carry transforms on other
+ # columns; the editor's client-side guard never runs here. Left in place, a
+ # transform on an unmirrored column is invisible and still stored, ready to
+ # go live the moment the mapped column resolves back to it.
+ DatasetDAO.clear_unmapped_partition_transforms(dataset)
+
+ # And repair a reference the import itself broke: a `sync` deletion can
+ # remove the column `partition_column` names, which otherwise leaves the
+ # dataset failing `UpdateDatasetCommand`'s validation on every later edit
--
+ # including a description-only PUT, which carries no columns payload and so
+ # cannot reach the forgiveness branch that tolerates a stored-only dangle.
+ #
+ # Flushed and expired first: the `sync` deletion is still pending here, and
+ # reading the relationship without flushing loads the removed rows back and
+ # un-deletes them.
+ db.session.flush()
+ db.session.expire(dataset, ["columns"])
+ DatasetDAO.clear_dangling_partition_mapping(
Review Comment:
This cleanup only drops references to columns that no longer exist. A bundle
that sets both `partition_column` and `partition_mapped_column` to `event_time`
imports fine, but the mapping validation treats an explicit self-mapping as
blocking, so a later description-only PUT on that dataset fails until someone
repairs the mapping. Should import clear (or reject) a mapping where the two
columns are equal, in line with the dangling-reference repair here?
##########
tests/unit_tests/connectors/sqla/partition_mapping_test.py:
##########
@@ -0,0 +1,2512 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""Unit tests for ``superset.connectors.sqla.partition_mapping``."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from decimal import Decimal
+from importlib import import_module
+from typing import Any, cast
+from unittest.mock import MagicMock, patch, PropertyMock
+
+import pandas as pd
+import pytest
+from dateutil.relativedelta import relativedelta
+from flask import Flask
+from sqlalchemy.dialects import mysql, postgresql
+
+from superset.config import DISALLOWED_SQL_FUNCTIONS
+from superset.connectors.sqla.models import SqlaTable, TableColumn
+from superset.connectors.sqla.partition_mapping import (
+ _mirror_verdict_cache_key,
+ _placeholder_is_bindable,
+ _probe_cache_key,
+ _reads_as_temporal,
+ _render_literal,
+ build_mirrored_predicates,
+ build_probe_sql,
+ contains_executable_comment,
+ contains_jinja,
+ contains_value_placeholder,
+ drop_unmapped_value_transforms,
+ equality_mirrors_safely,
+ evaluate_transform,
+ find_non_deterministic_functions,
+ grain_bucket_width,
+ GRAIN_BUCKET_WIDTHS,
+ is_bare_expression,
+ is_parseable,
+ is_transform_active,
+ is_unfinished,
+ known_mirror_verdict,
+ MappingValidationIssue,
+ mirror_operator,
+ MIRRORABLE_ALWAYS,
+ MIRRORABLE_IF_MONOTONIC,
+ mirrorable_operators,
+ parse_error_detail,
+ placeholder_is_executable,
+ probe_sql_is_evaluable,
+ probed_value_type_error,
+ RawProbeValue,
+ resolve_partition_mapping,
+ stored_expression_error,
+ validate_partition_mapping,
+ validate_transform,
+)
+from superset.constants import TimeGrain
+from superset.db_engine_specs.base import BaseEngineSpec
+from superset.db_engine_specs.oracle import OracleEngineSpec
+from superset.models.core import Database
+from superset.sql.parse import SQLStatement
+from superset.utils.core import FilterOperator, override_user
+
+
[email protected](autouse=True)
+def real_probe_cache(app: Flask) -> Any:
+ """
+ The test app runs a null cache, which would make every cache assertion here
+ vacuously pass. Swap in a real in-memory one for the duration.
+ """
+ from flask_caching import Cache
+
+ from superset.extensions import cache_manager
+
+ cache = Cache(config={"CACHE_TYPE": "SimpleCache",
"CACHE_DEFAULT_TIMEOUT": 300})
+ cache.init_app(app)
+ original = cache_manager._cache # noqa: SLF001
+ cache_manager._cache = cache # noqa: SLF001
+ yield
+ cache_manager._cache = original # noqa: SLF001
+
+
[email protected](autouse=True)
+def enable_partition_filter_mapping(app: Flask) -> Any:
+ """The feature ships off; every test here exercises it on."""
+ original =
app.config["DEFAULT_FEATURE_FLAGS"].get("PARTITION_FILTER_MAPPING")
+ app.config["DEFAULT_FEATURE_FLAGS"]["PARTITION_FILTER_MAPPING"] = True
+ yield
+ if original is None:
+ del app.config["DEFAULT_FEATURE_FLAGS"]["PARTITION_FILTER_MAPPING"]
+ else:
+ app.config["DEFAULT_FEATURE_FLAGS"]["PARTITION_FILTER_MAPPING"] =
original
+
+
+def _table(**kwargs: Any) -> SqlaTable:
+ database = Database(database_name="test_db", sqlalchemy_uri="sqlite://")
+ defaults: dict[str, Any] = {
+ "table_name": "web_events",
+ "database": database,
+ "main_dttm_col": "event_time",
+ "columns": [
+ TableColumn(column_name="event_time", is_dttm=True,
type="TIMESTAMP"),
+ TableColumn(
+ column_name="dt_epoch",
+ type="BIGINT",
+ partition_value_transform=None,
+ ),
+ TableColumn(column_name="country", type="VARCHAR"),
+ TableColumn(column_name="region_key", type="VARCHAR"),
+ ],
+ }
+ defaults.update(kwargs)
+ return SqlaTable(**defaults)
+
+
+def _mapped_table(
+ *,
+ partition_column: str = "dt_epoch",
+ mapped_column: str = "event_time",
+ transform: str | None = "unix_timestamp(:value)",
+ monotonic: bool = True,
+ partition_mapped_column: str | None = None,
+ main_dttm_col: str | None = "event_time",
+) -> SqlaTable:
+ table = _table(main_dttm_col=main_dttm_col)
+ table.partition_column = partition_column
+ table.partition_mapped_column = partition_mapped_column
+ for column in table.columns:
+ if column.column_name == mapped_column:
+ column.partition_value_transform = transform
+ column.partition_transform_is_monotonic = monotonic
+ return table
+
+
+# ---------------------------------------------------------------------------
+# §2 — operator safety matrix
+# ---------------------------------------------------------------------------
+
+
+def test_equality_and_in_are_always_mirrorable() -> None:
+ """``=`` and ``IN`` are safe for any function ``T``."""
+ assert MIRRORABLE_ALWAYS == {FilterOperator.EQUALS, FilterOperator.IN}
+
+
+def test_range_operators_require_a_monotonic_transform() -> None:
+ assert MIRRORABLE_IF_MONOTONIC == {
+ FilterOperator.GREATER_THAN,
+ FilterOperator.GREATER_THAN_OR_EQUALS,
+ FilterOperator.LESS_THAN,
+ FilterOperator.LESS_THAN_OR_EQUALS,
+ FilterOperator.TEMPORAL_RANGE,
+ }
+
+
+def test_mirrorable_operators_excludes_ranges_when_not_monotonic() -> None:
+ assert mirrorable_operators(is_monotonic=False) == MIRRORABLE_ALWAYS
+
+
+def test_mirrorable_operators_includes_ranges_when_monotonic() -> None:
+ assert mirrorable_operators(is_monotonic=True) == (
+ MIRRORABLE_ALWAYS | MIRRORABLE_IF_MONOTONIC
+ )
+
+
[email protected](
+ "operator",
+ [
+ FilterOperator.NOT_EQUALS,
+ FilterOperator.NOT_IN,
+ FilterOperator.LIKE,
+ FilterOperator.ILIKE,
+ FilterOperator.NOT_LIKE,
+ FilterOperator.NOT_ILIKE,
+ FilterOperator.IS_NULL,
+ FilterOperator.IS_NOT_NULL,
+ FilterOperator.IS_TRUE,
+ FilterOperator.IS_FALSE,
+ ],
+)
+def test_negations_and_pattern_matches_are_never_mirrorable(
+ operator: FilterOperator,
+) -> None:
+ """
+ ``T`` is not injective, so ``col != v`` does **not** imply ``T(col) !=
T(v)``:
+ mirroring it would drop rows the original filter keeps.
+ """
+ assert operator not in mirrorable_operators(is_monotonic=True)
+
+
+def test_equality_is_dropped_when_the_engine_does_not_compare_byte_exactly()
-> None:
+ """
+ "Safe for any function ``T``" reasons about value equality: ``T`` is a
+ function, so ``col = v`` gives ``T(col) = T(v)``. The engine reasons about
+ *SQL* equality, and the two part company under a case-insensitive
+ collation -- a stored ``'us'`` satisfies a filter for ``'US'`` while the
+ mirror ``hex('US')`` excludes the row, and the chart loses it silently.
+ """
+ assert mirrorable_operators(is_monotonic=False, equality_is_safe=False) ==
set()
+
+
+def test_ranges_survive_an_unsafe_equality() -> None:
+ """
+ Range mirroring already rests on the owner declaring ``T``
+ order-preserving with respect to the column's own order -- an assertion
+ about the very comparison semantics this gate checks for. Equality has no
+ such declaration behind it, which is why only equality is withdrawn.
+ """
+ assert (
+ mirrorable_operators(is_monotonic=True, equality_is_safe=False)
+ == MIRRORABLE_IF_MONOTONIC
+ )
+
+
[email protected](
+ "operator,expected",
+ [
+ (FilterOperator.GREATER_THAN, FilterOperator.GREATER_THAN_OR_EQUALS),
+ (FilterOperator.LESS_THAN, FilterOperator.LESS_THAN_OR_EQUALS),
+ (
+ FilterOperator.GREATER_THAN_OR_EQUALS,
+ FilterOperator.GREATER_THAN_OR_EQUALS,
+ ),
+ (FilterOperator.LESS_THAN_OR_EQUALS,
FilterOperator.LESS_THAN_OR_EQUALS),
+ (FilterOperator.EQUALS, FilterOperator.EQUALS),
+ (FilterOperator.IN, FilterOperator.IN),
+ ],
+)
+def test_a_strict_bound_mirrors_non_strictly(
+ operator: FilterOperator, expected: FilterOperator
+) -> None:
+ """
+ Order-preserving means non-decreasing, not injective, so ``col < v``
+ implies only ``T(col) <= T(v)``. A day-key transform maps a whole day onto
+ one value and a strict mirror drops the boundary bucket. The non-strict
+ cases are already safe and pass through, which is what makes applying this
+ twice harmless.
+ """
+ assert mirror_operator(operator) == expected
+
+
+def test_relaxing_a_bound_stays_inside_the_mirrorable_set() -> None:
+ """
+ The substitution is about strictness, not about which filters mirror. The
+ Explore indicator reads `mirrorable_operators`, so leaving that set alone
is
+ what keeps the glyph from promising pruning the SQL does not do -- and it
is
+ also what guarantees `handle_comparison_filter` is never handed an operator
+ it has no case for.
+ """
+ assert {
+ mirror_operator(operator) for operator in MIRRORABLE_IF_MONOTONIC
+ } <= MIRRORABLE_IF_MONOTONIC
+
+
+def _column(type_: str) -> TableColumn:
+ database = Database(database_name="t", sqlalchemy_uri="sqlite://")
+ table = SqlaTable(table_name="t", database=database)
+ return TableColumn(column_name="c", type=type_, table=table)
+
+
[email protected](
+ "type_, binary, expected",
+ [
+ ("VARCHAR", True, True),
+ ("VARCHAR", False, False),
+ # Numeric and temporal comparison is exact everywhere.
+ ("BIGINT", False, True),
+ ("TIMESTAMP", False, True),
+ ("DOUBLE", False, True),
+ ],
+)
+def test_equality_mirrors_safely_only_guards_string_columns(
+ app: Flask, type_: str, binary: bool, expected: bool
+) -> None:
+ class _Spec(BaseEngineSpec):
+ binary_string_comparison = binary
+
+ with app.app_context():
+ assert equality_mirrors_safely(_column(type_), _Spec) is expected
+
+
+def test_equality_mirrors_safely_fails_closed_on_an_unresolvable_type(
+ app: Flask,
+) -> None:
+ """
+ The gate exists to stop a silent wrong answer, so a column whose type it
+ cannot read is treated as the risky case rather than waved through.
+ """
+ column = MagicMock()
+ type(column).type_generic = PropertyMock(side_effect=ValueError("no type"))
+
+ with app.app_context():
+ assert equality_mirrors_safely(column, BaseEngineSpec) is False
+
+
+def test_no_mapped_column_leaves_the_matrix_alone(app: Flask) -> None:
+ """Nothing to reason about, and the resolver bails out on it anyway."""
+ with app.app_context():
+ assert equality_mirrors_safely(None, BaseEngineSpec) is True
+
+
[email protected](
+ "module, spec_name",
+ [
+ ("presto", "PrestoEngineSpec"),
+ ("hive", "HiveEngineSpec"),
+ ("trino", "TrinoEngineSpec"),
+ ("impala", "ImpalaEngineSpec"),
+ ("postgres", "PostgresEngineSpec"),
+ ("bigquery", "BigQueryEngineSpec"),
+ ("sqlite", "SqliteEngineSpec"),
+ ],
+)
+def test_the_byte_exact_engines_opt_in(module: str, spec_name: str) -> None:
+ """
+ The engines this feature targets compare text byte-exactly, so the gate
+ must not cost them the second canonical mapping (`lower(:value)` onto a
+ lowercased key).
+ """
+ spec = getattr(import_module(f"superset.db_engine_specs.{module}"),
spec_name)
+ assert spec.binary_string_comparison is True
+
+
[email protected]("module, spec_name", [("mysql", "MySQLEngineSpec")])
+def test_a_case_insensitive_engine_does_not_opt_in(module: str, spec_name:
str) -> None:
+ """
+ MySQL's default collation is `utf8mb4_0900_ai_ci`, which is exactly the
+ case the reviewer raised.
+ """
+ spec = getattr(import_module(f"superset.db_engine_specs.{module}"),
spec_name)
+ assert spec.binary_string_comparison is False
+
+
+def test_the_base_spec_does_not_opt_in() -> None:
+ """A spec that has not said so does not mirror string equality."""
+ assert BaseEngineSpec.binary_string_comparison is False
+
+
+# ---------------------------------------------------------------------------
+# Time grain bucket widths
+# ---------------------------------------------------------------------------
+
+
[email protected]("grain", list(TimeGrain))
+def test_every_builtin_grain_has_a_known_bucket_width(
+ app: Flask, grain: TimeGrain
+) -> None:
+ """
+ The widening is only sound for a grain whose bucket width we know, so a
+ grain added upstream without a width here would silently stop mirroring.
+ This is the test that notices.
+ """
+ with app.app_context():
+ assert grain_bucket_width(grain.value, "sqlite") is not None
+
+
+def test_a_calendar_grain_widens_by_a_calendar_month(app: Flask) -> None:
+ """A month is not thirty days, and the month it lands in decides its
width."""
+ with app.app_context():
+ width = grain_bucket_width(TimeGrain.MONTH.value, "sqlite")
+
+ assert width is not None
+ assert datetime(2026, 1, 31) + width == datetime(2026, 2, 28)
+ assert datetime(2026, 2, 1) + width == datetime(2026, 3, 1)
+
+
+def test_a_fractional_grain_has_a_width(app: Flask) -> None:
+ """
+ ``PT0.5H`` and ``P0.25Y`` are why the widths are written out rather than
+ parsed -- and the anchored week grains are intervals ``parse_duration``
+ rejects outright.
+ """
+ with app.app_context():
+ assert grain_bucket_width(TimeGrain.HALF_HOUR.value, "sqlite") ==
relativedelta(
+ minutes=30
+ )
+ assert grain_bucket_width(
+ TimeGrain.WEEK_ENDING_SATURDAY.value, "sqlite"
+ ) == relativedelta(days=7)
+
+
+def test_an_unknown_grain_has_no_known_width(app: Flask) -> None:
+ with app.app_context():
+ assert grain_bucket_width("P13X", "sqlite") is None
+ assert grain_bucket_width(None, "sqlite") is None
+
+
+def test_an_addon_grain_has_no_known_width(app: Flask) -> None:
+ """An operator-declared grain carries whatever duration string they
typed."""
+ with app.app_context():
+ app.config["TIME_GRAIN_ADDONS"] = {"PT2S": "2 second"}
+ try:
+ assert grain_bucket_width("PT2S", "sqlite") is None
+ finally:
+ app.config["TIME_GRAIN_ADDONS"] = {}
+
+
+def test_a_builtin_grain_an_operator_redefined_has_no_known_width(
+ app: Flask,
+) -> None:
+ """
+ ``TIME_GRAIN_ADDON_EXPRESSIONS`` replaces a built-in grain's SQL for one
+ engine, so its duration no longer bounds the displacement *there* -- but it
+ still does on every other engine.
+ """
+ with app.app_context():
+ app.config["TIME_GRAIN_ADDON_EXPRESSIONS"] = {
+ "sqlite": {"P1D": "DATETIME({col}, 'start of year')"}
+ }
+ try:
+ assert grain_bucket_width(TimeGrain.DAY.value, "sqlite") is None
+ assert grain_bucket_width(TimeGrain.DAY.value, "hive") is not None
+ finally:
+ app.config["TIME_GRAIN_ADDON_EXPRESSIONS"] = {}
+
+
+def test_the_width_table_is_keyed_on_grain_durations(app: Flask) -> None:
+ """Keys are the ISO duration a filter carries, not the enum member name."""
+ assert GRAIN_BUCKET_WIDTHS[TimeGrain.DAY] == relativedelta(days=1)
+ assert set(GRAIN_BUCKET_WIDTHS) == {grain.value for grain in TimeGrain}
+
+
+# ---------------------------------------------------------------------------
+# §4.1 — resolving the effective mapping, and the defensive bail-outs
+# ---------------------------------------------------------------------------
+
+
+def test_resolve_returns_the_mapping_when_everything_lines_up(app: Flask) ->
None:
+ with app.app_context():
+ mapping = resolve_partition_mapping(_mapped_table())
+
+ assert mapping is not None
+ assert mapping.partition_column == "dt_epoch"
+ assert mapping.mapped_column == "event_time"
+ assert mapping.value_transform == "unix_timestamp(:value)"
+ assert mapping.is_monotonic is True
+
+
+def test_effective_mapped_column_follows_main_dttm_col(app: Flask) -> None:
+ """No explicit override: the mapping follows the default datetime
column."""
+ with app.app_context():
+ mapping = resolve_partition_mapping(_mapped_table())
+
+ assert mapping is not None
+ assert mapping.mapped_column == "event_time"
+
+
+def test_explicit_override_wins_over_main_dttm_col(app: Flask) -> None:
+ table = _mapped_table(
+ mapped_column="country",
+ transform="lower(:value)",
+ monotonic=False,
+ partition_mapped_column="country",
+ )
+ table.partition_column = "region_key"
+
+ with app.app_context():
+ mapping = resolve_partition_mapping(table)
+
+ assert mapping is not None
+ assert mapping.mapped_column == "country"
+ assert mapping.partition_column == "region_key"
+ assert mapping.is_monotonic is False
+
+
+def test_resolve_returns_none_when_the_feature_flag_is_off(app: Flask) -> None:
+ table = _mapped_table()
+ with app.app_context():
+ with patch(
+ "superset.connectors.sqla.partition_mapping.feature_flag_manager."
+ "is_feature_enabled",
+ return_value=False,
+ ):
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_without_a_partition_column(app: Flask) -> None:
+ table = _mapped_table()
+ table.partition_column = None
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_when_partition_column_no_longer_exists(
+ app: Flask,
+) -> None:
+ """A column sync can drop the physical partition column out from under
us."""
+ table = _mapped_table()
+ table.partition_column = "dt_epoch_gone"
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_when_the_mapped_column_no_longer_exists(
+ app: Flask,
+) -> None:
+ table = _mapped_table(main_dttm_col="vanished")
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_on_self_mapping(app: Flask) -> None:
+ """
+ ``partition_column == effective mapped column`` mirrors a column onto
itself.
+ Save-time validation rejects it, but rows predating that validation exist.
+ """
+ table = _mapped_table(
+ partition_column="event_time",
+ mapped_column="event_time",
+ )
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
[email protected]("transform", [None, "", " "])
+def test_resolve_returns_none_without_a_transform(
+ app: Flask, transform: str | None
+) -> None:
+ table = _mapped_table(transform=transform)
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_when_the_transform_lacks_the_placeholder(
+ app: Flask,
+) -> None:
+ table = _mapped_table(transform="unix_timestamp(event_time)")
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_for_a_non_deterministic_transform(app: Flask) ->
None:
+ """
+ The query-time gate is `is_transform_active`, the same one the Explore
+ indicator reads. Anything narrower would let a transform that reached
+ storage without passing `UpdateDatasetCommand` -- through import, or a
+ hand-written bundle -- mirror a filter with a value frozen at probe time,
+ while the editor reported the mapping as inactive and nothing on screen
+ said otherwise.
+ """
+ table = _mapped_table(transform="unix_timestamp(:value) + rand()")
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_for_a_multi_expression_transform(app: Flask) ->
None:
+ """A select list is not an expression; the probe cannot align its
columns."""
+ table = _mapped_table(transform="lower(:value), 'x'")
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_for_an_unparseable_transform(app: Flask) -> None:
+ table = _mapped_table(transform="unix_timestamp(:value")
+ with app.app_context():
+ assert resolve_partition_mapping(table) is None
+
+
+def test_resolve_returns_none_when_the_mapped_column_has_an_advanced_data_type(
+ app: Flask,
+) -> None:
+ """
+ ``translate_filter`` builds its own predicate shape from *translated*
values,
+ so the ``(operator, value)`` pair the operator matrix reasons about does
not
+ exist. Mirroring anyway would silently apply the wrong values (§4.1).
+ """
+ table = _mapped_table()
+ for column in table.columns:
+ if column.column_name == "event_time":
+ column.advanced_data_type = "port"
+
+ with app.app_context():
+ with patch.dict(app.config["ADVANCED_DATA_TYPES"], {"port":
MagicMock()}):
+ with patch(
+
"superset.connectors.sqla.partition_mapping.feature_flag_manager."
+ "is_feature_enabled",
+ side_effect=lambda flag: (
+ flag in {"PARTITION_FILTER_MAPPING",
"ENABLE_ADVANCED_DATA_TYPES"}
+ ),
+ ):
+ assert resolve_partition_mapping(table) is None
+
+
+def test_advanced_data_type_does_not_block_when_the_flag_is_off(app: Flask) ->
None:
+ """An inert ``advanced_data_type`` is not a reason to skip mirroring."""
+ table = _mapped_table()
+ for column in table.columns:
+ if column.column_name == "event_time":
+ column.advanced_data_type = "port"
+
+ with app.app_context():
+ with patch(
+ "superset.connectors.sqla.partition_mapping.feature_flag_manager."
+ "is_feature_enabled",
+ side_effect=lambda flag: flag == "PARTITION_FILTER_MAPPING",
+ ):
+ assert resolve_partition_mapping(table) is not None
+
+
+# ---------------------------------------------------------------------------
+# §5 — transform inspection helpers
+# ---------------------------------------------------------------------------
+
+
[email protected](
+ "transform,expected",
+ [
+ ("unix_timestamp(:value)", True),
+ ("lower(:value)", True),
+ ("CAST(:value AS BIGINT)", True),
+ ("unix_timestamp(event_time)", False),
+ (":values", False),
+ ("", False),
+ (None, False),
+ ],
+)
+def test_contains_value_placeholder(transform: str | None, expected: bool) ->
None:
+ assert contains_value_placeholder(transform) is expected
+
+
[email protected](
+ "transform,expected",
+ [
+ ("unix_timestamp(:value)", False),
+ ("{{ current_username() }}", True),
+ ("lower({% if x %}:value{% endif %})", True),
+ ("lower(:value) -- {# comment #}", True),
+ ],
+)
+def test_contains_jinja(transform: str, expected: bool) -> None:
+ assert contains_jinja(transform) is expected
+
+
[email protected](
+ ("transform", "expected"),
+ [
+ ("lower(:value)", False),
+ ("lower(:value) /* an ordinary comment */", False),
+ ("lower(:value) -- a line comment", False),
+ (":value /*!50000 + 1 */", True),
+ (":value /*M!50000 + 1 */", True),
+ (":value /*! + 1 */", True),
+ ("", False),
+ (None, False),
+ ],
+)
+def test_contains_executable_comment(transform: str | None, expected: bool) ->
None:
+ assert contains_executable_comment(transform) is expected
+
+
[email protected](
+ "transform",
+ [
+ "unix_timestamp(:value)",
+ "lower(:value)",
+ "CAST(:value AS BIGINT)",
+ "date_format(:value, 'yyyyMMdd')",
+ ],
+)
+def test_pure_transforms_report_no_non_deterministic_functions(
+ transform: str,
+) -> None:
+ assert find_non_deterministic_functions(transform, "hive") == set()
+
+
[email protected](
+ "transform,expected_name",
+ [
+ ("date_diff(:value, now())", "NOW"),
+ ("CAST(:value AS BIGINT) + rand()", "RAND"),
+ ("CAST(:value AS DATE) - current_date", "CURRENT_DATE"),
+ ],
+)
+def test_non_deterministic_functions_are_reported(
+ transform: str, expected_name: str
+) -> None:
+ """
+ The probe runs at a different moment and in a different session from the
+ chart query, and its result is cached, so anything time- or
+ randomness-dependent freezes a snapshot of probe time into the predicate.
+ """
+ assert expected_name in find_non_deterministic_functions(transform, "hive")
+
+
+def test_niladic_unix_timestamp_is_rejected_but_the_unary_form_is_not() ->
None:
+ """
+ On Hive/Impala ``unix_timestamp()`` means "now" while ``unix_timestamp(x)``
+ -- the canonical temporal transform -- is pure. The distinction is the
whole
+ reason this check inspects arity rather than just the name.
+ """
+ assert find_non_deterministic_functions("unix_timestamp(:value)", "hive")
== set()
+ assert find_non_deterministic_functions(
+ "unix_timestamp(:value) - unix_timestamp()", "hive"
+ )
+
+
+# ---------------------------------------------------------------------------
+# §4.2 — probe evaluation
+# ---------------------------------------------------------------------------
+
+
+def _user(username: str) -> Any:
+ """A stand-in for `override_user`, which only needs `.username` here."""
+ user = MagicMock()
+ user.username = username
+ user.is_anonymous = False
+ return user
+
+
+def _database_returning(values: list[Any]) -> Database:
+ """A real ``Database`` (so the dialect is real) with only ``get_df``
stubbed."""
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ database.get_df = MagicMock( # type: ignore[method-assign]
+ return_value=pd.DataFrame(
+ [values], columns=[f"v{i}" for i in range(len(values))]
+ )
+ )
+ return database
+
+
+def _database_returning_frame(frame: pd.DataFrame) -> Database:
+ """
+ ``_database_returning`` for a probe result whose column dtypes matter.
+
+ The flat-list sibling builds its frame from one Python list, so pandas
+ infers a single dtype and the row-versus-cell distinction cannot be seen.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ database.get_df = MagicMock( # type: ignore[method-assign]
+ return_value=frame
+ )
+ return database
+
+
+def _probe(database: Database) -> MagicMock:
+ """
+ The stubbed ``get_df``, typed for call assertions.
+
+ ``Database.get_df`` is a real method, so mypy reads its declared signature
+ rather than the mock that replaced it and rejects ``.call_args`` and
+ friends.
+ """
+ return cast(MagicMock, database.get_df)
+
+
+def test_evaluate_transform_returns_one_value_per_input(app: Flask) -> None:
+ database = _database_returning([1767225600, 1769904000])
+
+ with app.app_context():
+ result = evaluate_transform(
+ database,
+ None,
+ "default",
+ "unix_timestamp(:value)",
+ ["2026-01-01 00:00:00", "2026-02-01 00:00:00"],
+ )
+
+ assert result == [1767225600, 1769904000]
+ _probe(database).assert_called_once()
+
+
+def test_evaluate_transform_binds_values_rather_than_interpolating(
+ app: Flask,
+) -> None:
+ """
+ Filter values are attacker-controlled (a Gamma user picks them), so they
must
+ be bound and escaped, never pasted into the probe SQL.
+ """
+ database = _database_returning(["o''brien"])
+
+ with app.app_context():
+ evaluate_transform(database, None, None, "lower(:value)", ["O'Brien"])
+
+ sql = _probe(database).call_args.kwargs["sql"]
+ assert "O'Brien" not in sql
+ assert "O''Brien" in sql
+
+
[email protected](
+ ("transform", "expected"),
+ [
+ ("lower(:value)", True),
+ ("CAST(:value AS BIGINT)", True),
+ ("date_format(:value, '%Y%m%d')", True),
+ # The cast shorthand: a trailing colon stops SQLAlchemy's bind scan.
+ (":value::bigint", False),
+ ("(:value::date)::text", False),
+ ],
+)
+def test_a_postgres_cast_shorthand_is_not_bindable(
+ transform: str, expected: bool
+) -> None:
+ """
+ `VALUE_PLACEHOLDER_RE` is `:value\b`, which matches inside
`:value::bigint`.
+ SQLAlchemy's own scan ends with `(?![:\\w$])`, so the following colon stops
+ it: `text(":value::bigint")` reports a parameter named `valu`, and asking
it
+ to bind `value` raises.
+
+ Pinned because the two regexes disagreeing is the whole bug: every
+ write-side gate accepted the transform and the probe then died on an
+ `ArgumentError` that `_probe` swallows.
+ """
+ assert _placeholder_is_bindable(transform) is expected
+
+
+def test_a_cast_shorthand_transform_still_probes(app: Flask) -> None:
+ """
+ The transform is valid SQL, passes every gate, and used to prune nothing:
+ `build_probe_sql` raised, `_probe` swallowed it, and the only symptom was a
+ mapping that silently never mirrored -- with the preview reporting an
engine
+ failure for SQL that was never sent.
+ """
+ database = _database_returning([20260115])
+
+ with app.app_context():
+ result = evaluate_transform(
+ database, None, None, ":value::bigint", ["20260115"]
+ )
+
+ assert result == [20260115]
+ assert _probe(database).call_count == 1
+ assert "::bigint" in _probe(database).call_args.kwargs["sql"]
+
+
+def test_a_cast_shorthand_value_is_still_escaped_by_the_dialect(
+ app: Flask,
+) -> None:
+ """
+ The fallback substitutes where the bind path binds, so it has to earn the
+ same guarantee its sibling above pins: the value is rendered by the
+ dialect's literal processor, not pasted in.
+
+ Binding was only ever the vehicle for reaching that processor --
+ `_compile_literal` inlines the result either way -- so this renders first
+ and substitutes second. Same string, different order.
+ """
+ database = _database_returning(["x"])
+
+ with app.app_context():
+ evaluate_transform(database, None, None, ":value::text", ["O'Brien"])
+
+ sql = _probe(database).call_args.kwargs["sql"]
+ assert "O'Brien" not in sql
+ assert "O''Brien" in sql
+
+
+def test_evaluate_transform_dedupes_repeated_values(app: Flask) -> None:
+ """Three inputs, two distinct: the probe evaluates the transform twice."""
+ database = _database_returning(["us", "us"])
+
+ with app.app_context():
+ result = evaluate_transform(
+ database, None, None, "lower(:value)", ["US", "US", "us"]
+ )
+
+ assert result == ["us", "us", "us"]
+ sql = _probe(database).call_args.kwargs["sql"]
+ assert sql.count("lower") == 2
+
+
+def test_evaluate_transform_fails_open_on_a_short_result_row(app: Flask) ->
None:
+ """A row narrower than the probe asked for means the results cannot be
+ aligned back to their inputs; pruning is skipped rather than guessed at."""
+ database = _database_returning(["us"])
+
+ with app.app_context():
+ assert (
+ evaluate_transform(database, None, None, "lower(:value)", ["US",
"CA"])
+ is None
+ )
+
+
+def test_evaluate_transform_fails_open_on_a_wide_result_row(app: Flask) ->
None:
+ """
+ Too many columns is the dangerous direction. A transform whose select list
+ holds two expressions returns 2N columns for N inputs, and reading the
first
+ N interleaves the expressions instead of taking one per value -- a
predicate
+ built from the wrong values rather than one that is merely short. A frame
of
+ `('us', 'x', 'ca', 'x')` read as the first two columns would have yielded
+ `('us', 'x')` and dropped every CA row.
+
+ The transform is a perfectly ordinary `lower(:value)` and the width comes
+ from the stubbed frame, which is the only way to reach the guard: a
+ transform that really does carry two expressions -- `lower(:value), 'x'` --
+ is refused by `stored_expression_error` for not being a single expression,
+ so `get_df` is never called and `None` comes back for an unrelated reason.
+ Asserting the probe *ran* is what separates the two, and it is what makes
+ relaxing the guard from `!=` to `<` fail here rather than pass: four
columns
+ for two inputs is not fewer than two.
+ """
+ database = _database_returning(["us", "x", "ca", "x"])
+
+ with app.app_context():
+ assert (
+ evaluate_transform(database, None, None, "lower(:value)", ["US",
"CA"])
+ is None
+ )
+
+ assert _probe(database).call_count == 1
+
+
+def test_the_probe_select_carries_the_engine_s_from_clause(app: Flask) -> None:
+ """
+ A `SELECT` with no `FROM` is not universal SQL: Oracle and Db2 need a
+ one-row table to select from. Without this every probe on those engines
+ raises, which `evaluate_transform` swallows -- so the only symptom is a
+ correctly configured mapping that silently never prunes.
+ """
+ assert build_probe_sql("lower(:value)", ["US"]) == "SELECT lower('US') AS
v0"
+ assert build_probe_sql("lower(:value)", ["US"], None, " FROM
DUAL").endswith(
+ " FROM DUAL"
+ )
+ assert OracleEngineSpec.select_without_from_suffix == " FROM DUAL"
+ assert BaseEngineSpec.select_without_from_suffix == ""
+
+
[email protected]("dialect", [mysql.dialect(), postgresql.dialect()])
+def test_the_probe_select_does_not_double_percent_signs(dialect: Any) -> None:
+ """
+ `TextClause` compilation runs the dialect's `post_process_text`, which
+ doubles every `%` on any paramstyle whose DBAPI later interpolates
+ parameters -- MySQL and Postgres, and every spec inheriting from them. The
+ probe has no such interpolation pass: it executes with no bound parameters
+ at all. Left doubled, `date_format(:value, '%Y%m%d')` asks the warehouse
+ for the literal string `%%Y%%m%%d`, so the probe returns that instead of a
+ date and the mirrored predicate drops every row the filter keeps.
+ """
+ sql = build_probe_sql("date_format(:value, '%Y%m%d')", ["2026-01-15"],
dialect)
+
+ assert "'%Y%m%d'" in sql
+ assert "%%" not in sql
+
+
+def test_the_probe_select_is_unchanged_on_a_dialect_that_does_not_double() ->
None:
+ """
+ Pins the no-dialect path, which is also the blind spot: the default dialect
+ does not double percent signs, so no assertion made through it can see the
+ bug the two cases above cover.
+ """
+ assert (
+ build_probe_sql("date_format(:value, '%Y%m%d')", ["2026-01-15"])
+ == "SELECT date_format('2026-01-15', '%Y%m%d') AS v0"
+ )
+
+
+def test_a_raw_probe_value_is_substituted_literally() -> None:
+ r"""
+ `RawProbeValue.sql` is rendered from a filter value, so it reaches
+ `parse_skeleton` as the *replacement* for `:value`. Read as a replacement
+ template, a backslash in it is an escape: `array('a\nb')` -- which is how a
+ dialect renders the two characters `\` and `n` -- would become a real
+ newline, so the engine is asked about a different array than the predicate
+ compares and the mirror can exclude rows the filter keeps.
+ """
+ sql = build_probe_sql("toYYYYMM(:value)",
[RawProbeValue(r"array('a\nb')")])
+
+ assert sql == r"SELECT toYYYYMM(array('a\nb')) AS v0"
+
+
+def test_a_raw_probe_value_does_not_re_inject_the_placeholder() -> None:
+ r"""
+ `\g<0>` in a replacement template expands to the whole match, which would
+ put `:value` back into the probe -- an unbound parameter the engine
refuses.
+ """
+ sql = build_probe_sql("toYYYYMM(:value)", [RawProbeValue(r"'a\g<0>b'")])
+
+ assert sql == r"SELECT toYYYYMM('a\g<0>b') AS v0"
+ assert ":value" not in sql
+
+
+def test_a_raw_probe_value_with_a_group_reference_does_not_raise() -> None:
+ r"""
+ `\1` is an invalid group reference for this pattern, so as a replacement
+ template it raised `re.PatternError` rather than probing anything.
+ """
+ sql = build_probe_sql("toYYYYMM(:value)", [RawProbeValue(r"'a\1b'")])
+
+ assert sql == r"SELECT toYYYYMM('a\1b') AS v0"
+
+
+def test_a_probed_literal_keeps_its_percent_sign(app: Flask) -> None:
+ """
+ The preview panel renders probed values through the same dialect, so a
+ value that merely contains a `%` would be shown doubled.
+ """
+ database = Database(database_name="pct_db",
sqlalchemy_uri="postgresql://h/d")
+
+ with app.app_context():
+ assert _render_literal(database, "100%") == "'100%'"
+
+
+def test_the_probe_reads_each_cell_without_row_wide_dtype_coercion(
+ app: Flask,
+) -> None:
+ """
+ A row is a Series, so it carries one dtype across every probe column: a row
+ mixing an exact int64 with a float unifies to float64 and an integer
+ partition key above 2^53 comes back rounded. The mirror then asks for a key
+ the warehouse does not hold, and the row the filter matched is dropped.
+ """
+ database = _database_returning_frame(
+ pd.DataFrame({"v0": [9007199254740993], "v1": [1.5]})
+ )
+
+ with app.app_context():
+ probed = evaluate_transform(
+ database, None, None, "CAST(:value AS NUMERIC)",
["9007199254740993", "1.5"]
+ )
+
+ assert probed == [9007199254740993, 1.5]
+
+
+def test_evaluate_transform_pins_catalog_and_schema(app: Flask) -> None:
+ """
+ The probe runs in a different session from the chart query; pinning the
+ catalog and schema keeps session settings as close as the pool allows.
+ """
+ database = _database_returning([1])
+
+ with app.app_context():
+ evaluate_transform(database, "prod", "analytics", "lower(:value)",
["x"])
+
+ kwargs = _probe(database).call_args.kwargs
+ assert kwargs["catalog"] == "prod"
+ assert kwargs["schema"] == "analytics"
+
+
+def test_evaluate_transform_fails_open_when_the_probe_raises(app: Flask) ->
None:
+ """A wedged engine must not break the chart — it just stops pruning."""
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ database.get_df = MagicMock( # type: ignore[method-assign]
+ side_effect=RuntimeError("connection reset")
+ )
+
+ with app.app_context():
+ assert evaluate_transform(database, None, None, "lower(:value)",
["x"]) is None
+
+
+def test_evaluate_transform_fails_open_on_an_empty_result(app: Flask) -> None:
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ database.get_df = MagicMock( # type: ignore[method-assign]
+ return_value=pd.DataFrame()
+ )
+
+ with app.app_context():
+ assert evaluate_transform(database, None, None, "lower(:value)",
["x"]) is None
+
+
+def test_evaluate_transform_returns_none_for_no_values(app: Flask) -> None:
+ database = _database_returning([])
+
+ with app.app_context():
+ assert evaluate_transform(database, None, None, "lower(:value)", [])
is None
+
+ _probe(database).assert_not_called()
+
+
[email protected](
+ "transform",
+ [
+ "(SELECT password FROM ab_user LIMIT 1) || :value",
+ "lower(:value) UNION ALL SELECT password FROM ab_user",
+ "lower(:value); DROP TABLE ab_user",
+ "password || :value FROM ab_user",
+ "lower(:value) FROM ab_user WHERE 1 = 1",
+ ],
+ ids=[
+ "subquery",
+ "set-operation",
+ "multi-statement",
+ "bare-from",
+ "from-and-where",
+ ],
+)
+def test_a_transform_that_is_not_a_storable_expression_never_runs(
+ app: Flask, transform: str
+) -> None:
+ """
+ The probe is where a transform stops being text and becomes SQL.
+
+ `build_probe_sql` binds `:value` and splices the rest in verbatim, so a
+ transform that was never a single harmless expression is arbitrary SQL
+ against the dataset's database -- read back out of the emitted predicate
+ by anyone who can open "View query". The write-side gates refuse these,
+ but they can only speak for rows written after they existed, so the last
+ word belongs here.
+ """
+ database = _database_returning(["irrelevant"])
+
+ with app.app_context():
+ assert evaluate_transform(database, None, None, transform, ["us"]) is
None
+
+ _probe(database).assert_not_called()
+
+
[email protected](
+ "transform, unfinished",
+ [
+ ("unix_timestamp(:value", True),
+ ("unix_timestamp(:value))", True),
+ ("unix_timestamp(:value) +", True),
+ ("unix_timestamp(:value)", False),
+ ("unix_timestamp(:value); DROP TABLE ab_user", False),
+ ("unix_timestamp(:value) UNION ALL SELECT password FROM ab_user",
False),
+ ("password || :value FROM ab_user", False),
+ ],
+)
+def test_is_unfinished_separates_not_sql_yet_from_the_wrong_sql(
+ transform: str, unfinished: bool
+) -> None:
+ """
+ The two deserve opposite treatment -- a half-typed transform is stored and
+ parked inactive, the wrong SQL is refused -- and neither `is_parseable` nor
+ parsing as a single statement tells them apart, because the multi-statement
+ form fails both exactly as unfinished text does. Parsing as a script does.
+ """
+ assert is_unfinished(transform, "hive") is unfinished
+
+
+def test_a_transform_smuggling_its_own_from_clause_is_refused(app: Flask) ->
None:
+ """
+ The hole this gate was added for.
+
+ `is_parseable` counts only the projection, so `password || :value FROM
+ ab_user` reads as a single select expression and passed every write-side
+ gate -- including with `ALLOW_ADHOC_SUBQUERY` off, since a top-level FROM
+ is not a sub-query. `build_probe_sql` then emitted `SELECT password || 'us'
+ AS v0 FROM ab_user`, where its own alias is read as a table alias on the
+ FROM clause the transform brought with it: valid SQL, one row, and the
+ secret handed back through the predicate the preview panel renders.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ transform = "password || :value FROM ab_user"
+
+ assert is_parseable(transform, "sqlite") is True
+ assert is_bare_expression(transform, "sqlite") is False
+
+ with app.app_context():
+ reason = stored_expression_error(database, None, None, transform)
+ assert reason is not None
+
+ blocking = _blocking(validate_transform(transform, "sqlite"))
+ assert len(blocking) == 1
+ assert "single SQL expression" in str(blocking[0].message)
+ assert is_transform_active(transform, "sqlite") is False
+
+
+def test_a_subquery_transform_is_refused_even_with_adhoc_subquery_enabled(
+ app: Flask,
+) -> None:
+ """
+ `validate_adhoc_subquery` admits a sub-query under `ALLOW_ADHOC_SUBQUERY`
+ and returns RLS-rewritten SQL -- which this path has no way to propagate,
+ because `build_probe_sql` splices the original transform text. So the probe
+ ran a sub-query with no row-level-security predicates at all and reported
+ the hidden row in `emitted_predicate`. A transform is a scalar function of
+ `:value`, so refusing the sub-query outright is both correct and what makes
+ the missing RLS rewrite moot.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ transform = "(SELECT secret FROM vault LIMIT 1) || :value"
+
+ with app.app_context():
+ app.config["DEFAULT_FEATURE_FLAGS"]["ALLOW_ADHOC_SUBQUERY"] = True
+ try:
+ reason = stored_expression_error(database, None, None, transform)
+ finally:
+ del app.config["DEFAULT_FEATURE_FLAGS"]["ALLOW_ADHOC_SUBQUERY"]
+
+ assert reason is not None
+ assert "sub-query" in reason
+
+
[email protected](
+ "comment",
+ ["/*!50000 + (SELECT secret FROM vault LIMIT 1) */", "/*M!50000 + 1 */"],
+)
+def test_an_executable_comment_is_refused_by_every_gate(
+ app: Flask, comment: str
+) -> None:
+ """
+ The one construct that made the shape gates describe a statement the engine
+ would not run.
+
+ sqlglot reads a MySQL executable comment as comment data: the pinned parse
+ below reports no sub-query and no clause, so `stored_expression_error` and
+ `probe_sql_is_evaluable` both passed it, while MySQL and MariaDB execute
+ what is inside. An owner with dataset write and no SQL Lab could put one in
+ a transform, call the preview and read a table back out of
+ `emitted_predicate`, with no RLS on the hidden query.
+
+ Refused on every engine, not only the two that honour it -- hence SQLite
+ here. There is no legitimate transform carrying one, and a per-engine gate
+ would store on Postgres what detonates after the dataset is re-pointed.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ transform = f"(:value {comment})"
+
+ # The premise: no parse-tree gate can see it.
+ statement = SQLStatement(f"SELECT (1 {comment}) AS v0", "mysql")
+ assert statement.has_subquery() is False
+ assert statement.get_clause_names() == set()
+
+ with app.app_context():
+ assert stored_expression_error(database, None, None, transform) is not
None
+
+ blocking = [
+ issue for issue in validate_transform(transform, "mysql") if
issue.blocking
+ ]
+ assert len(blocking) == 1
+ assert "executable comment" in str(blocking[0].message)
+
+
+def test_an_executable_comment_stops_the_probe_before_the_engine(app: Flask)
-> None:
+ """
+ The gate above is the save path. This is the row an earlier release stored:
+ `_probe` consults `stored_expression_error` on every call, so a transform
+ already in the metadata database never reaches `get_df`.
+ """
+ database = _database_returning(["x"])
+ transform = "(:value /*!50000 + (SELECT secret FROM vault LIMIT 1) */)"
+
+ with app.app_context():
+ assert evaluate_transform(database, None, None, transform, ["US"]) is
None
+
+ assert _probe(database).call_count == 0
+
+
+def test_a_plain_comment_is_still_allowed_through_the_shape_gates(
+ app: Flask,
+) -> None:
+ """
+ Only the executable form is refused. An ordinary comment is inert on every
+ engine, so rejecting it would cost an owner a transform that works -- and
+ the placeholder gate already covers the case that matters, a `:value` the
+ comment swallows.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+
+ with app.app_context():
+ assert (
+ stored_expression_error(database, None, None, "lower(:value) /* ok
*/")
+ is None
+ )
+
+
+def test_a_transform_calling_a_disallowed_function_is_refused(app: Flask) ->
None:
+ """
+ The probe executes the transform, so the operator's function denylist has
+ to apply to it for the same reason it applies in SQL Lab.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
+ transform = "version() || :value"
+
+ with app.app_context():
+ original = app.config["DISALLOWED_SQL_FUNCTIONS"]
+ try:
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = {"sqlite": {"version"}}
+ reason = stored_expression_error(database, None, None, transform)
+ assert reason is not None
+ assert "version" in reason
+
+ # Named, rather than echoing the operator's whole denylist back.
+ assert "current_user" not in reason
+
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = {}
+ assert stored_expression_error(database, None, None, transform) is
None
+ finally:
+ # Restored: `app.config` is shared, and leaving it emptied silently
+ # disabled this gate for every test that ran afterwards.
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = original
+
+
[email protected](
+ "function",
+ [
+ "database_to_xml",
+ "database_to_xmlschema",
+ "database_to_xml_and_xmlschema",
+ "schema_to_xml",
+ "schema_to_xmlschema",
+ "schema_to_xml_and_xmlschema",
+ "table_to_xml",
+ "table_to_xmlschema",
+ "table_to_xml_and_xmlschema",
+ "query_to_xml",
+ "query_to_xmlschema",
+ "query_to_xml_and_xmlschema",
+ ],
+)
+def test_the_whole_postgres_xml_family_is_denied_by_default(function: str) ->
None:
+ """
+ These read tables with no FROM clause and no sub-query, so every gate that
+ reasons about table references is blind to them -- the denylist is the only
+ thing that can refuse them, and it has to name each shape. `schema_to_xml`
+ and three siblings were missing, so a bare scalar call cleared
+ `stored_expression_error` entirely and returned connection-user-readable
+ schema contents through preview.
+
+ Asserted against the shipped default rather than through the gate, because
+ the gap was an incomplete enumeration rather than broken wiring -- and
+ because `app.config` is ambient here, which the behavioural test below
+ handles by setting the denylist itself.
+ """
+ assert function in DISALLOWED_SQL_FUNCTIONS["postgresql"]
+
+
+def test_a_denied_read_capable_function_is_refused_despite_having_no_from(
+ app: Flask,
+) -> None:
+ """
+ The wiring half: a bare scalar call passes the length, single-statement,
+ bare-expression, sub-query and placeholder gates, so the denylist is what
+ has to stop it.
+ """
+ database = Database(database_name="probe_db",
sqlalchemy_uri="postgresql://u@h/d")
+ transform = "schema_to_xml('public', true, false, '') || :value"
+
+ with app.app_context():
+ original = app.config["DISALLOWED_SQL_FUNCTIONS"]
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = {"postgresql":
{"schema_to_xml"}}
+ try:
+ reason = stored_expression_error(database, None, None, transform)
+ finally:
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = original
+
+ assert reason is not None
+ assert "schema_to_xml" in reason
+
+
[email protected]("function", ["ts_rewrite", "ts_stat"])
+def test_the_postgres_full_text_query_functions_are_denied_by_default(
+ function: str,
+) -> None:
+ """
+ `ts_stat('select ...')` and `ts_rewrite(q, 'select ...')` run their *text*
+ argument as a query. The query is a string, so these have the same blind
+ spot the XML family above does: no FROM clause and no sub-query for a
+ table-reference gate to reason about, which leaves the denylist as the only
+ thing that can refuse them.
+ """
+ assert function in DISALLOWED_SQL_FUNCTIONS["postgresql"]
+
+
+def test_a_denied_query_running_text_function_is_refused(app: Flask) -> None:
+ """
+ The wiring half, and the only thing that proves the *name* matching works
+ for these two: neither is a function sqlglot models, so each reaches
+ `get_disallowed_functions` through its ANONYMOUS branch rather than as a
+ typed node. A denylist entry that only matched modelled functions would
+ pass the enumeration test above and still let these through.
+ """
+ database = Database(database_name="probe_db",
sqlalchemy_uri="postgresql://u@h/d")
+ transform = "ts_stat('select 1') || :value"
+
+ with app.app_context():
+ original = app.config["DISALLOWED_SQL_FUNCTIONS"]
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = {"postgresql": {"ts_stat"}}
+ try:
+ reason = stored_expression_error(database, None, None, transform)
+ finally:
+ app.config["DISALLOWED_SQL_FUNCTIONS"] = original
+
+ assert reason is not None
+ assert "ts_stat" in reason
+
+
+def test_the_function_denylist_is_keyed_on_the_engine_spec_name(app: Flask) ->
None:
+ """
+ Every other denylist gate in the codebase keys on the engine spec's own
+ name, which is also the name the config documentation tells an operator to
+ write, so this one does too -- a spec covering several SQLAlchemy backends
+ through `engine_aliases` reports the one canonical name under which its
+ entry exists.
+ """
+ database = Database(database_name="probe_db", sqlalchemy_uri="sqlite://")
Review Comment:
On `sqlite://` both `database.backend` and `database.db_engine_spec.engine`
are `"sqlite"`, so this passes even if the gate keyed on `database.backend`
instead of the engine spec name. Could this use an aliased engine (backend
`postgres`, spec `postgresql`) with a function denied only under `postgresql`,
and assert `stored_expression_error` rejects it? That is the case the test name
promises.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
