sadpandajoe commented on code in PR #43771:
URL: https://github.com/apache/superset/pull/43771#discussion_r3971689298
##########
superset/mcp_service/chart/tool/get_chart_data.py:
##########
@@ -1115,56 +1224,22 @@ async def _query_from_form_data( # noqa: C901
error_type="NoData",
)
- columns = []
- for col_name in raw_columns:
- sample_values = [
- row.get(col_name) for row in data[:3] if row.get(col_name) is
not None
- ]
- data_type = "string"
- if sample_values and all(
- isinstance(v, (int, float)) for v in sample_values
- ):
- data_type = "numeric"
- columns.append(
- DataColumn(
- name=col_name,
- display_name=col_name.replace("_", " ").title(),
- data_type=data_type,
- sample_values=sample_values[:3],
- null_count=sum(1 for row in data if row.get(col_name) is
None),
- unique_count=len({str(row.get(col_name)) for row in data}),
- )
- )
+ columns = _build_data_columns(
+ data, raw_columns, query_result.get("coltypes", [])
+ )
cache_status = get_cache_status_from_result(
query_result, force_refresh=request.force_refresh
)
chart_name = form_data.get("slice_name", "Unsaved chart")
- if request.format in {"csv", "excel"}:
- from superset.models.slice import Slice
-
- # A transient chart supplies export metadata without saving
anything.
- chart = Slice(id=0, slice_name=chart_name, viz_type=viz_type)
- export = (
- _export_data_as_csv
- if request.format == "csv"
- else _export_data_as_excel
- )
- return export(
- chart,
- data[: request.limit] if request.limit else data,
- raw_columns,
- cache_status,
- PerformanceMetadata(query_duration_ms=0,
cache_status="fresh_query"),
- )
summary = (
f"Unsaved chart ({viz_type}). "
f"Contains {len(data)} rows across {len(raw_columns)} columns."
)
await ctx.report_progress(4, 4, "Building response")
- return ChartData(
+ response = ChartData(
Review Comment:
The unsaved `form_data_key` path now always builds `ChartData`, so
`format="csv"` or `"excel"` silently returns JSON with no export payload while
the saved-chart path still exports. Can this preserve the export branch here
and restore coverage for unsaved exports?
##########
superset/mcp_service/app.py:
##########
@@ -395,8 +395,8 @@ def get_default_instructions(
- chart_type="table": Data table for detailed views
- chart_type="table", viz_type="ag-grid-table": Interactive AG Grid table
- chart_type="pie": Pie chart for proportional data (set donut=True for donut)
-- chart_type="gauge": Gauge/dial for one numeric metric, optionally grouped
- into up to 10 dials (native viz_type is "gauge_chart")
+- chart_type="sunburst": Hierarchical part-to-whole chart (hierarchy + metric
Review Comment:
Replacing the Gauge entry with Sunburst removes Gauge from the MCP
capability catalog even though `GaugeChartPlugin` remains registered and
available. Clients following these instructions will not discover a supported
chart type; can the Gauge entry and display-name mapping remain alongside
Sunburst?
##########
superset/mcp_service/chart/query_result.py:
##########
@@ -15,82 +15,1521 @@
# specific language governing permissions and limitations
# under the License.
-"""Helpers for interpreting ChartDataCommand result envelopes."""
+"""Canonicalize and validate ``ChartDataCommand`` result envelopes."""
import math
+import time as system_time
+from bisect import bisect_right
from collections.abc import Mapping
+from dataclasses import dataclass
+from datetime import date, datetime, time, timedelta, timezone
+from decimal import Decimal
+from enum import Enum
from typing import Any
+from uuid import UUID
+from zoneinfo import ZoneInfo
+import numpy as np
+import pandas as pd
+import pytz
+from dateutil import tz as dateutil_tz, zoneinfo as dateutil_zoneinfo
+from pydantic import BaseModel
+from pydantic_core import to_json
+
+from superset.common.chart_data import ChartDataResultFormat
+from superset.common.db_query_status import QueryStatus
+from superset.constants import CACHE_DISABLED_TIMEOUT
from superset.mcp_service.chart.schemas import ChartError
+from superset.utils.core import (
+ ExtraFiltersReasonType,
+ ExtraFiltersTimeColumnType,
+ GenericDataType,
+)
FAILED_QUERY_STATUSES = frozenset(
{"error", "failed", "stopped", "timed_out", "cancelled", "canceled"}
)
+# These are aggregate envelope limits, not per-query allowances. In particular,
+# splitting a result across the maximum number of queries must not multiply the
+# permitted rows, nodes, or encoded bytes.
+MAX_QUERY_RESULTS = 32
+MAX_QUERY_RESULT_ROWS_PER_QUERY = 50_000
+MAX_QUERY_RESULT_ROWS = 100_000
+MAX_QUERY_RESULT_COLUMNS = 4_096
+MAX_QUERY_RESULT_VALUES = 2_500_000
+MAX_QUERY_RESULT_VALUE_BYTES = 16 * 1024 * 1024
+MAX_QUERY_RESULT_METADATA_BYTES = 1024 * 1024
+MAX_QUERY_RESULT_METADATA_ITEMS = 32_768
+MAX_RESULT_VALUE_ITEMS = 4_096
+MAX_RESULT_VALUE_DEPTH = 32
+MAX_RESULT_STRING_LENGTH = 65_536
+MAX_RESULT_KEY_LENGTH = 4_096
+MAX_RESULT_INTEGER_BITS = 4_096
+MAX_RESULT_INTEGER_DIGITS = 1_234
+MAX_RESULT_DECIMAL_DIGITS = 1_024
+MAX_RESULT_DECIMAL_MAGNITUDE = 4_096
+MAX_RESULT_DECIMAL_STORAGE = 2_048
+MAX_QUERY_RESULT_ROWCOUNT = 2**63 - 1
+MAX_QUERY_RESULT_CACHE_TIMEOUT = 2**31 - 1
+MAX_QUERY_RESULT_TIMESTAMP_LENGTH = 64
+
+_ERROR_KEYS = ("error", "errors", "error_message", "message", "detail")
+_MAX_ERROR_TEXT_BYTES = 2_000
+_TRUSTED_TIMEZONE_TYPES = (timezone, ZoneInfo)
+_SAFE_RESULT_ENUM_TYPES = frozenset(
+ {
+ ChartDataResultFormat,
+ QueryStatus,
+ ExtraFiltersReasonType,
+ ExtraFiltersTimeColumnType,
+ GenericDataType,
+ }
+)
+_RESULT_FORMAT_VALUES = frozenset(
+ object.__getattribute__(member, "_value_") for member in
ChartDataResultFormat
+)
+_COLTYPE_VALUES = frozenset(
+ object.__getattribute__(member, "_value_") for member in GenericDataType
+)
+_NUMPY_INTEGER_TYPES = frozenset(
+ type(value)
+ for value in (
+ np.int8(0),
+ np.int16(0),
+ np.int32(0),
+ np.int64(0),
+ np.uint8(0),
+ np.uint16(0),
+ np.uint32(0),
+ np.uint64(0),
+ )
+)
+_NUMPY_FLOAT_TYPES = frozenset(
+ type(value)
+ for value in (np.float16(0), np.float32(0), np.float64(0),
np.longdouble(0))
+)
+_PANDAS_NAT_TYPE = type(pd.NaT)
+_PANDAS_NA_TYPE = type(pd.NA)
+_PANDAS_PERIOD_TYPE = type(pd.Period("2000-01", freq="M"))
+_PANDAS_INTERVAL_TYPE = type(pd.Interval(0, 1))
+_DATEUTIL_FIXED_TIMEZONE_TYPES = frozenset(
+ {type(dateutil_tz.tzoffset(None, 0)), type(dateutil_tz.tzutc())}
+)
+_DATEUTIL_NAMED_TIMEZONE_TYPES = frozenset(
+ {dateutil_tz.tzfile, dateutil_zoneinfo.tzfile}
+)
+_DATEUTIL_TTINFO_TYPE = type(
+ object.__getattribute__(dateutil_tz.gettz("UTC"),
"__dict__")["_ttinfo_std"]
+)
+_DATEUTIL_LOCAL_TIMEZONE_TYPE = type(dateutil_tz.tzlocal())
+_PYTZ_FIXED_TIMEZONE_TYPES = frozenset({type(pytz.FixedOffset(1))})
+_MAX_DATEUTIL_TRANSITIONS = 4_096
+_MAX_DATEUTIL_TTINFOS = 256
+_MAX_DATEUTIL_TRANSITION_MAGNITUDE = 10**12
+
+
+@dataclass
+class _ResultBudget:
+ """Aggregate counters shared by every query and metadata value."""
+
+ rows: int = 0
+ values: int = 0
+ json_bytes: int = 0
+ metadata_items: int = 0
+ metadata_bytes: int = 0
+
+
+@dataclass(frozen=True)
+class _DateutilTimezoneState:
+ """Hook-free subset of a validated exact dateutil tzfile transition
table."""
-def _query_error_text(value: Any) -> str | None:
- """Convert a bounded query error payload into a useful message."""
- if value is None or value is False:
+ transitions: tuple[int, ...]
+ transition_offsets: tuple[int, ...]
+ standard_offset: int
+ before_offset: int | None
+
+
+def _invalid_result(message: str) -> ChartError:
+ return ChartError(
+ error=f"Chart query returned {message}.",
+ error_type="InvalidQueryResult",
+ )
+
+
+def _invalid_metadata(label: str) -> ChartError:
+ return ChartError(
+ error=f"{label} returned hostile or malformed metadata.",
+ error_type="InvalidQueryResult",
+ )
+
+
+def _safe_enum_value(value: Any, expected: frozenset[type[Any]]) -> Any | None:
+ """Read trusted enum storage without invoking public conversion hooks."""
+ if type(value) not in expected or type(value) not in
_SAFE_RESULT_ENUM_TYPES:
return None
- if isinstance(value, Mapping):
- for key in ("error", "error_message", "message", "detail"):
- if text := _query_error_text(value.get(key)):
- return text
+ return object.__getattribute__(value, "_value_")
+
+
+def _bounded_utf8_length(value: str, maximum: int) -> int | None:
+ """Return the exact UTF-8 size while bounding pre-encoding work."""
+ if str.__len__(value) > maximum:
return None
- if isinstance(value, (list, tuple)):
- parts = [text for item in value if (text := _query_error_text(item))]
- return "; ".join(parts[:3]) or None
- text = str(value)
- return text[:2000] if text else None
+ try:
+ encoded = str.encode(value, "utf-8", errors="strict")
+ except UnicodeEncodeError:
+ return None
+ size = bytes.__len__(encoded)
+ return size if size <= maximum else None
-def _failure_for_query_payload(
- payload: Mapping[str, Any], label: str
-) -> ChartError | None:
- """Extract one failure from a top-level or per-query payload."""
+def _json_string_size(value: str, maximum: int) -> int | None:
+ """Return compact UTF-8 JSON string size without serializing the value."""
+ raw_size = _bounded_utf8_length(value, maximum)
+ if raw_size is None:
+ return None
+ escaped_size = raw_size + 2
+ for character in value:
+ codepoint = ord(character)
+ if character in {'"', "\\", "\b", "\t", "\n", "\f", "\r"}:
+ escaped_size += 1
+ elif codepoint < 0x20:
+ escaped_size += 5
+ return escaped_size
+
+
+def _integer_json_size(value: int) -> int:
+ """Return exact decimal JSON size without rendering the bounded integer."""
+ magnitude = -value if value < 0 else value
+ if magnitude == 0:
+ digits = 1
+ else:
+ bits = int.bit_length(magnitude)
+ digits = ((bits - 1) * 30103) // 100000 + 1
+ if magnitude >= 10**digits:
+ digits += 1
+ return digits + (value < 0)
+
+
+def _container_json_syntax_size(item_count: int, *, mapping: bool) -> int:
+ """Return braces/brackets plus compact separators and mapping colons."""
+ if item_count == 0:
+ return 2
+ return 2 + item_count - 1 + (item_count if mapping else 0)
+
+
+def _normalized_scalar_json_size(value: Any) -> int:
+ """Return exact compact JSON size for a normalized scalar."""
+ value_type = type(value)
+ if value is None:
+ return 4
+ if value_type is bool:
+ return 4 if value else 5
+ if value_type is str:
+ size = _json_string_size(value, MAX_RESULT_STRING_LENGTH)
+ assert size is not None
+ return size
+ if value_type is int:
+ return _integer_json_size(value)
+ if value_type is float:
+ return len(float.__repr__(value))
+ if value_type is Decimal:
+ # Pydantic serializes Decimal values as JSON strings so their exact
+ # finite value survives the wire projection without binary rounding.
+ text = Decimal.__str__(value)
+ size = _json_string_size(text, MAX_RESULT_STRING_LENGTH)
+ assert size is not None
+ return size
+ raise AssertionError("result scalar was not normalized")
+
+
+def _pydantic_scalar_json_size(value: Any) -> int:
+ """Return the scalar size emitted by Pydantic's JSON serializer."""
+ if type(value) is float:
+ # pydantic-core uses the shortest exponent (``1e-7``), while Python's
+ # repr retains a leading zero (``1e-07``).
+ return len(to_json(value))
+ return _normalized_scalar_json_size(value)
+
+
+def _charge_json_bytes(
+ budget: _ResultBudget, size: int, *, metadata: bool = False
+) -> str | None:
+ budget.json_bytes += size
+ if budget.json_bytes > MAX_QUERY_RESULT_VALUE_BYTES:
+ return "too many aggregate JSON bytes"
+ if metadata:
+ budget.metadata_bytes += size
+ if budget.metadata_bytes > MAX_QUERY_RESULT_METADATA_BYTES:
+ return "too many aggregate metadata JSON bytes"
+ return None
+
+
+def _charge_value(budget: _ResultBudget, *, metadata: bool = False) -> str |
None:
+ budget.values += 1
+ if budget.values > MAX_QUERY_RESULT_VALUES:
+ return "too many aggregate values"
+ if metadata:
+ budget.metadata_items += 1
+ if budget.metadata_items > MAX_QUERY_RESULT_METADATA_ITEMS:
+ return "too many aggregate metadata values"
+ return None
+
+
+def _charge_text(
+ value: str,
+ budget: _ResultBudget,
+ *,
+ key: bool = False,
+ metadata: bool = False,
+) -> str | None:
+ maximum = (
+ MAX_RESULT_KEY_LENGTH
+ if key
+ else MAX_QUERY_RESULT_METADATA_BYTES
+ if metadata
+ else MAX_RESULT_STRING_LENGTH
+ )
+ size = _json_string_size(value, maximum)
+ if size is None:
+ return "an invalid or oversized object key" if key else "invalid text
data"
+ return _charge_json_bytes(budget, size, metadata=metadata)
+
+
+def _integer_failure(value: int) -> str | None:
+ bits = int.bit_length(value)
+ if bits > MAX_RESULT_INTEGER_BITS:
+ return "an oversized integer"
+ digits = 1 if bits == 0 else ((bits - 1) * 30103) // 100000 + 1
+ if digits > MAX_RESULT_INTEGER_DIGITS:
+ return "an oversized integer"
+ return None
+
+
+def _decimal_failure(value: Decimal) -> str | None:
+ if Decimal.__sizeof__(value) > MAX_RESULT_DECIMAL_STORAGE:
+ return "an oversized Decimal"
+ if not Decimal.is_finite(value):
+ return "a non-finite Decimal"
+ parts = Decimal.as_tuple(value)
+ if tuple.__len__(parts.digits) > MAX_RESULT_DECIMAL_DIGITS:
+ return "an oversized Decimal"
+ exponent = parts.exponent
+ if type(exponent) is not int or abs(exponent) >
MAX_RESULT_DECIMAL_MAGNITUDE:
+ return "an oversized Decimal"
+ return None
+
+
+def _type_mro(value_type: type[Any]) -> tuple[type[Any], ...]:
+ """Read a concrete type's MRO without consulting metaclass overrides."""
+ try:
+ mro = type.__getattribute__(value_type, "__mro__")
+ except (AttributeError, TypeError): # pragma: no cover - defensive
metaclass
+ return ()
+ return mro if type(mro) is tuple else ()
+
+
+def _timezone_name_without_hooks(tzinfo: Any) -> str | None: # noqa: C901
+ """Read common pytz/dateutil zone state without dispatching timezone
hooks."""
+ value_mro = _type_mro(type(tzinfo))
+ if any(base is pytz.tzinfo.BaseTzInfo for base in value_mro):
+ for base in value_mro:
+ try:
+ namespace = type.__getattribute__(base, "__dict__")
+ except (AttributeError, TypeError): # pragma: no cover
+ continue
+ zone = namespace.get("zone")
+ if type(zone) is str and _bounded_utf8_length(zone, 256) is not
None:
+ try:
+ canonical = pytz.timezone(zone)
+ except (KeyError, ValueError):
+ return None
+ # Generated pytz types are trusted; arbitrary subclasses that
+ # inherit their internal fields are not.
+ return zone if type(canonical) is type(tzinfo) else None
+
+ if type(tzinfo) not in _DATEUTIL_NAMED_TIMEZONE_TYPES:
+ return None
+
+ try:
+ namespace = object.__getattribute__(tzinfo, "__dict__")
+ except (AttributeError, TypeError):
+ return None
+ if type(namespace) is not dict:
+ return None
+ filename = dict.get(namespace, "_filename")
+ if type(filename) is not str or _bounded_utf8_length(filename, 4_096) is
None:
+ return None
+ marker = "/zoneinfo/"
+ if (offset := str.find(filename, marker)) >= 0:
+ name = str.__getitem__(filename, slice(offset + len(marker), None))
+ elif not str.startswith(filename, "/") and str.find(filename, "\\") < 0:
+ name = filename
+ else:
+ return None
+ parts = str.split(name, "/")
+ if not parts or any(part in {"", ".", ".."} for part in parts):
+ return None
+ return name if _bounded_utf8_length(name, 256) is not None else None
+
+
+def _object_namespace(value: Any) -> dict[str, Any] | None:
+ """Read exact instance storage without descriptor dispatch."""
+ try:
+ namespace = object.__getattribute__(value, "__dict__")
+ except (AttributeError, TypeError):
+ return None
+ return namespace if type(namespace) is dict else None
+
+
+def _dateutil_ttinfo_offset_without_hooks(value: Any) -> int | None:
+ """Validate one exact dateutil transition record and return its offset."""
+ if type(value) is not _DATEUTIL_TTINFO_TYPE:
+ return None
+ try:
+ offset = object.__getattribute__(value, "offset")
+ delta = object.__getattribute__(value, "delta")
+ isdst = object.__getattribute__(value, "isdst")
+ abbreviation = object.__getattribute__(value, "abbr")
+ is_standard = object.__getattribute__(value, "isstd")
+ is_gmt = object.__getattribute__(value, "isgmt")
+ dst_offset = object.__getattribute__(value, "dstoffset")
+ except (AttributeError, TypeError):
+ return None
+ if type(offset) is not int or not -86_400 < offset < 86_400:
+ return None
+ if type(delta) is not timedelta or delta != timedelta(seconds=offset):
+ return None
+ if type(isdst) is not int or isdst not in {0, 1}:
+ return None
+ if abbreviation is not None and (
+ type(abbreviation) is not str or _bounded_utf8_length(abbreviation,
256) is None
+ ):
+ return None
+ if type(is_standard) is not bool or type(is_gmt) is not bool:
+ return None
+ if type(dst_offset) is not timedelta:
+ return None
+ if not -timedelta(days=1) < dst_offset < timedelta(days=1):
+ return None
+ return offset
+
+
+def _dateutil_named_state_without_hooks( # noqa: C901
+ tzinfo: Any,
+) -> _DateutilTimezoneState | None:
+ """Validate bounded exact dateutil tzfile state without timezone hooks."""
+ if type(tzinfo) not in _DATEUTIL_NAMED_TIMEZONE_TYPES:
+ return None
+ if _timezone_name_without_hooks(tzinfo) is None:
+ return None
+ namespace = _object_namespace(tzinfo)
+ if namespace is None:
+ return None
+ transitions = dict.get(namespace, "_trans_list")
+ utc_transitions = dict.get(namespace, "_trans_list_utc")
+ transition_info = dict.get(namespace, "_trans_idx")
+ info_list = dict.get(namespace, "_ttinfo_list")
+ standard_info = dict.get(namespace, "_ttinfo_std")
+ before_info = dict.get(namespace, "_ttinfo_before")
+ first_info = dict.get(namespace, "_ttinfo_first")
+ if (
+ type(transitions) is not tuple
+ or type(utc_transitions) is not tuple
+ or type(transition_info) is not tuple
+ or type(info_list) is not list
+ or tuple.__len__(transitions) > _MAX_DATEUTIL_TRANSITIONS
+ or tuple.__len__(utc_transitions) != tuple.__len__(transitions)
+ or tuple.__len__(transition_info) != tuple.__len__(transitions)
+ or list.__len__(info_list) == 0
+ or list.__len__(info_list) > _MAX_DATEUTIL_TTINFOS
+ ):
+ return None
+
+ previous_transition: int | None = None
+ previous_utc_transition: int | None = None
+ for index in range(tuple.__len__(transitions)):
+ transition = tuple.__getitem__(transitions, index)
+ utc_transition = tuple.__getitem__(utc_transitions, index)
+ if (
+ type(transition) is not int
+ or type(utc_transition) is not int
+ or abs(transition) > _MAX_DATEUTIL_TRANSITION_MAGNITUDE
+ or abs(utc_transition) > _MAX_DATEUTIL_TRANSITION_MAGNITUDE
+ or (previous_transition is not None and transition <=
previous_transition)
+ or (
+ previous_utc_transition is not None
+ and utc_transition <= previous_utc_transition
+ )
+ ):
+ return None
+ previous_transition = transition
+ previous_utc_transition = utc_transition
+
+ known_info_ids: set[int] = set()
+ for index in range(list.__len__(info_list)):
+ info = list.__getitem__(info_list, index)
+ if _dateutil_ttinfo_offset_without_hooks(info) is None:
+ return None
+ known_info_ids.add(id(info))
+ for info in (standard_info, before_info, first_info):
+ if info is not None and id(info) not in known_info_ids:
+ return None
+ for index in range(tuple.__len__(transition_info)):
+ if id(tuple.__getitem__(transition_info, index)) not in known_info_ids:
+ return None
+ if not transitions:
+ if (
+ standard_info is not list.__getitem__(info_list, 0)
+ or first_info is not standard_info
+ or before_info is not None
+ ):
+ return None
+ else:
+ expected_standard = None
+ expected_dst = None
+ for index in range(tuple.__len__(transition_info) - 1, -1, -1):
+ info = tuple.__getitem__(transition_info, index)
+ is_dst = object.__getattribute__(info, "isdst")
+ if expected_standard is None and not is_dst:
+ expected_standard = info
+ elif expected_dst is None and is_dst:
+ expected_dst = info
+ if expected_standard is not None and expected_dst is not None:
+ break
+ if expected_standard is None:
+ expected_standard = expected_dst
+ expected_before = None
+ for index in range(list.__len__(info_list)):
+ info = list.__getitem__(info_list, index)
+ if not object.__getattribute__(info, "isdst"):
+ expected_before = info
+ break
+ if expected_before is None:
+ expected_before = list.__getitem__(info_list, 0)
+ if standard_info is not expected_standard or before_info is not
expected_before:
+ return None
+ standard_offset = _dateutil_ttinfo_offset_without_hooks(standard_info)
+ if standard_offset is None:
+ return None
+ before_offset = (
+ _dateutil_ttinfo_offset_without_hooks(before_info)
+ if before_info is not None
+ else None
+ )
+ transition_offsets: list[int] = []
+ previous_offset: int | None = None
+ previous_base_offset: int | None = None
+ previous_is_dst: int | None = None
+ previous_dst_offset = 0
+ for index in range(tuple.__len__(transition_info)):
+ info = tuple.__getitem__(transition_info, index)
+ if id(info) not in known_info_ids:
+ return None
+ offset = _dateutil_ttinfo_offset_without_hooks(info)
+ if offset is None:
+ return None
+ is_dst = object.__getattribute__(info, "isdst")
+ dst_offset_seconds = 0
+ if previous_is_dst is not None and is_dst:
+ if not previous_is_dst:
+ assert previous_offset is not None
+ dst_offset_seconds = offset - previous_offset
+ if not dst_offset_seconds and previous_dst_offset:
+ dst_offset_seconds = previous_dst_offset
+ previous_dst_offset = dst_offset_seconds
+ base_offset = offset - dst_offset_seconds
+ adjustment = base_offset
+ if (
+ previous_base_offset is not None
+ and base_offset != previous_base_offset
+ and is_dst != previous_is_dst
+ ):
+ adjustment = previous_base_offset
+ if (
+ tuple.__getitem__(transitions, index)
+ != tuple.__getitem__(utc_transitions, index) + adjustment
+ ):
+ return None
+ transition_offsets.append(offset)
+ previous_offset = offset
+ previous_base_offset = base_offset
+ previous_is_dst = is_dst
+ if transitions and before_offset is None:
+ return None
+ return _DateutilTimezoneState(
+ transitions=transitions,
+ transition_offsets=tuple(transition_offsets),
+ standard_offset=standard_offset,
+ before_offset=before_offset,
+ )
+
+
+def _dateutil_named_offset_without_hooks( # noqa: C901
+ value: datetime, tzinfo: Any
+) -> timezone | None:
+ """Preserve dateutil's source-selected wall offset from validated state."""
+ state = _dateutil_named_state_without_hooks(tzinfo)
+ if state is None:
+ return None
+ epoch_ordinal = date.toordinal(date(1970, 1, 1))
+ wall_timestamp = (
+ (datetime.toordinal(value) - epoch_ordinal) * 86_400
+ + value.hour * 3_600
+ + value.minute * 60
+ + value.second
+ )
+ transitions = state.transitions
+ selected_offset: int | None
+ if not transitions:
+ selected_offset = state.standard_offset
+ else:
+ index = bisect_right(transitions, wall_timestamp) - 1
+
+ def offset_at(transition_index: int | None) -> int | None:
+ if transition_index is None or transition_index + 1 >=
len(transitions):
+ return state.standard_offset
+ if transition_index < 0:
+ return state.before_offset
+ return state.transition_offsets[transition_index]
+
+ if index > 0:
+ selected_offset = offset_at(index)
+ previous_offset = offset_at(index - 1)
+ if selected_offset is None or previous_offset is None:
+ return None
+ is_ambiguous = wall_timestamp < transitions[index] + (
+ previous_offset - selected_offset
+ )
+ if not value.fold and is_ambiguous:
+ index -= 1
+ selected_offset = offset_at(index)
+ if selected_offset is None:
+ return None
+ try:
+ return timezone(timedelta(seconds=selected_offset))
+ except (OverflowError, ValueError):
+ return None
+
+
+def _pytz_named_offset_without_hooks(tzinfo: Any) -> timezone | None:
+ """Return a localized pytz zone's stored offset without calling hooks."""
+ if _timezone_name_without_hooks(tzinfo) is None:
+ return None
+ namespace = _object_namespace(tzinfo)
+ offset = dict.get(namespace, "_utcoffset") if namespace is not None else
None
+ if offset is None:
+ # Static pytz zones store their fixed offset on the generated class.
+ class_namespace = type.__getattribute__(type(tzinfo), "__dict__")
+ offset = class_namespace.get("_utcoffset")
+ if type(offset) is not timedelta:
+ return None
+ try:
+ return timezone(offset)
+ except ValueError:
+ return None
+
+
+def _dateutil_local_offset_without_hooks(
+ value: datetime, tzinfo: Any
+) -> timezone | None:
+ """Select a dateutil local offset using builtin system-time data."""
+ if type(tzinfo) is not _DATEUTIL_LOCAL_TIMEZONE_TYPE:
+ return None
+ namespace = _object_namespace(tzinfo)
+ if namespace is None:
+ return None
+ standard_offset = dict.get(namespace, "_std_offset")
+ daylight_offset = dict.get(namespace, "_dst_offset")
+ has_daylight = dict.get(namespace, "_hasdst")
+ if (
+ type(standard_offset) is not timedelta
+ or type(daylight_offset) is not timedelta
+ or type(has_daylight) is not bool
+ ):
+ return None
+ selected_offset = standard_offset
+ if has_daylight:
+ epoch = datetime(1970, 1, 1)
+ naive = datetime(
+ value.year,
+ value.month,
+ value.day,
+ value.hour,
+ value.minute,
+ value.second,
+ value.microsecond,
+ )
+ timestamp = (naive - epoch).total_seconds()
+ try:
+ is_daylight = bool(
+ system_time.localtime(timestamp +
system_time.timezone).tm_isdst
+ )
+ daylight_saved = daylight_offset - standard_offset
+ previous_is_daylight = bool(
+ system_time.localtime(
+ timestamp
+ - timedelta.total_seconds(daylight_saved)
+ + system_time.timezone
+ ).tm_isdst
+ )
+ except (OverflowError, OSError, ValueError):
+ return None
+ if not is_daylight and is_daylight != previous_is_daylight:
+ is_daylight = not bool(value.fold)
+ selected_offset = daylight_offset if is_daylight else standard_offset
+ try:
+ return timezone(selected_offset)
+ except ValueError:
+ return None
+
+
+def _canonical_timezone(tzinfo: Any) -> timezone | ZoneInfo | None: # noqa:
C901
+ if any(type(tzinfo) is trusted for trusted in _TRUSTED_TIMEZONE_TYPES):
+ return tzinfo
+ if tzinfo is pytz.UTC:
+ return timezone.utc
+ if type(tzinfo) in _DATEUTIL_FIXED_TIMEZONE_TYPES:
+ try:
+ namespace = object.__getattribute__(tzinfo, "__dict__")
+ except (AttributeError, TypeError):
+ return timezone.utc if type(tzinfo) is type(dateutil_tz.tzutc())
else None
+ if type(namespace) is not dict:
+ return None
+ offset = dict.get(namespace, "_offset")
+ if type(offset) is not timedelta:
+ return timezone.utc if type(tzinfo) is type(dateutil_tz.tzutc())
else None
+ if abs(offset) >= timedelta(days=1):
+ return None
+ return timezone(offset)
+ if type(tzinfo) in _PYTZ_FIXED_TIMEZONE_TYPES:
+ try:
+ namespace = object.__getattribute__(tzinfo, "__dict__")
+ except (AttributeError, TypeError):
+ return None
+ if type(namespace) is not dict:
+ return None
+ minutes = dict.get(namespace, "_minutes")
+ if type(minutes) is not int or not -1_440 < minutes < 1_440:
+ return None
+ return timezone(timedelta(minutes=minutes))
+ return None
+
+
+def _timestamp_offset_without_hooks(value: pd.Timestamp) -> timezone | None:
+ """Recover a timestamp's stored wall-clock offset without timezone
hooks."""
+ multipliers = {"s": 1_000_000_000, "ms": 1_000_000, "us": 1_000, "ns": 1}
+ multiplier = multipliers.get(value.unit)
+ if multiplier is None:
+ return None
+ try:
+ instant_ns = int(value.asm8.view("i8")) * multiplier
+ epoch_ordinal = date.toordinal(date(1970, 1, 1))
+ wall_ns = (
+ (
+ (datetime.toordinal(value) - epoch_ordinal) * 86_400
+ + value.hour * 3600
+ + value.minute * 60
+ + value.second
+ )
+ * 1_000_000_000
+ + value.microsecond * 1000
+ + value.nanosecond
+ )
+ offset_ns = wall_ns - instant_ns
+ if offset_ns % 1000:
+ return None
+ return timezone(timedelta(microseconds=offset_ns // 1000))
+ except (OverflowError, TypeError, ValueError):
+ return None
+
+
+def _canonical_datetime(value: datetime) -> tuple[str | None, str | None]:
+ """Serialize an exact datetime through trusted timezone state only."""
+ tzinfo = value.tzinfo
+ canonical_value = value
+ if tzinfo is not None and not any(
+ type(tzinfo) is trusted for trusted in _TRUSTED_TIMEZONE_TYPES
+ ):
+ canonical_tz = (
+ _pytz_named_offset_without_hooks(tzinfo)
+ or _dateutil_local_offset_without_hooks(value, tzinfo)
+ or _dateutil_named_offset_without_hooks(value, tzinfo)
+ or _canonical_timezone(tzinfo)
+ )
+ if canonical_tz is None:
+ return None, "a datetime with an unsupported timezone"
+ canonical_value = datetime(
+ value.year,
+ value.month,
+ value.day,
+ value.hour,
+ value.minute,
+ value.second,
+ value.microsecond,
+ tzinfo=canonical_tz,
+ fold=value.fold,
+ )
+ try:
+ return datetime.isoformat(canonical_value), None
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid datetime"
+
+
+def _canonical_time(value: time) -> tuple[str | None, str | None]:
+ """Serialize an exact time through trusted timezone state only."""
+ tzinfo = value.tzinfo
+ canonical_value = value
+ if tzinfo is not None and not any(
+ type(tzinfo) is trusted for trusted in _TRUSTED_TIMEZONE_TYPES
+ ):
+ if _dateutil_named_state_without_hooks(tzinfo) is not None:
+ canonical_value = time(
+ value.hour,
+ value.minute,
+ value.second,
+ value.microsecond,
+ fold=value.fold,
+ )
+ else:
+ canonical_tz = _canonical_timezone(tzinfo)
+ if canonical_tz is None and type(tzinfo) is
_DATEUTIL_LOCAL_TIMEZONE_TYPE:
+ namespace = _object_namespace(tzinfo)
+ if namespace is not None and dict.get(namespace, "_hasdst") is
False:
+ offset = dict.get(namespace, "_std_offset")
+ if type(offset) is timedelta:
+ try:
+ canonical_tz = timezone(offset)
+ except ValueError:
+ canonical_tz = None
+ if canonical_tz is None:
+ return None, "a time with an unsupported timezone"
+ canonical_value = time(
+ value.hour,
+ value.minute,
+ value.second,
+ value.microsecond,
+ tzinfo=canonical_tz,
+ fold=value.fold,
+ )
+ try:
+ return time.isoformat(canonical_value), None
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid time"
+
+
+def _canonical_timestamp(value: pd.Timestamp) -> tuple[str | None, str | None]:
+ """Preserve a trusted timestamp's instant, offset, nanoseconds, and
fold."""
+ try:
+ tzinfo = value.tzinfo
+ if tzinfo is not None and not any(
+ type(tzinfo) is trusted for trusted in _TRUSTED_TIMEZONE_TYPES
+ ):
+ supported_timezone = (
+ _canonical_timezone(tzinfo) is not None
+ or _pytz_named_offset_without_hooks(tzinfo) is not None
+ or type(tzinfo) is _DATEUTIL_LOCAL_TIMEZONE_TYPE
+ or _dateutil_named_state_without_hooks(tzinfo) is not None
+ )
+ if not supported_timezone:
+ return None, "a timestamp with an unsupported timezone"
+ canonical_tz = _timestamp_offset_without_hooks(value)
+ if canonical_tz is None:
+ return None, "an invalid timestamp"
+ raw_value = value.asm8.view("i8")
+ value = pd.Timestamp(raw_value, unit=value.unit,
tz="UTC").tz_convert(
+ canonical_tz
+ )
+ return pd.Timestamp.isoformat(value), None
+ except (KeyError, OverflowError, TypeError, ValueError):
+ return None, "an invalid timestamp"
+
+
+def _normalize_scalar(value: Any) -> tuple[Any, str | None]: # noqa: C901
+ """Convert one exact trusted producer scalar to a JSON-safe scalar."""
+ value_type = type(value)
+ if value is None or value_type is bool or value_type is str:
+ return value, None
+ if value_type is int:
+ return value, _integer_failure(value)
+ if value_type is float:
+ if math.isnan(value):
+ return None, None
+ return (value, None) if math.isfinite(value) else (None, "a non-finite
number")
+ if value_type is Decimal:
+ return value, _decimal_failure(value)
+ if value_type is datetime:
+ return _canonical_datetime(value)
+ if value_type is time:
+ return _canonical_time(value)
+ if value_type is date:
+ return date.isoformat(value), None
+ if value_type is timedelta:
+ try:
+ return pd.Timedelta(value).isoformat(), None
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid duration"
+ if value_type is UUID:
+ return UUID.__str__(value), None
+
+ if value_type is _PANDAS_NAT_TYPE or value_type is _PANDAS_NA_TYPE:
+ return None, None
+ if value_type is pd.Timestamp:
+ return _canonical_timestamp(value)
+ if value_type is pd.Timedelta:
+ if pd.isna(value):
+ return None, None
+ try:
+ return pd.Timedelta.isoformat(value), None
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid pandas duration"
+ if value_type is _PANDAS_PERIOD_TYPE or value_type is
_PANDAS_INTERVAL_TYPE:
+ # These concrete immutable pandas extension scalars are trusted. Exact
+ # type checks deliberately exclude subclasses with conversion hooks.
+ try:
+ normalized_text = str(value)
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid pandas scalar"
+ if _bounded_utf8_length(normalized_text, MAX_RESULT_STRING_LENGTH) is
None:
+ return None, "an oversized pandas scalar"
+ return normalized_text, None
+ if value_type in _NUMPY_INTEGER_TYPES:
+ normalized = int(value)
+ return normalized, _integer_failure(normalized)
+ if value_type in _NUMPY_FLOAT_TYPES:
+ normalized_float = float(value)
+ if math.isnan(normalized_float):
+ return None, None
+ return (
+ (normalized_float, None)
+ if math.isfinite(normalized_float)
+ else (None, "a non-finite NumPy number")
+ )
+ if value_type is np.bool_:
+ return bool(value), None
+ if value_type is np.str_:
+ return str(value), None
+ if value_type is np.datetime64:
+ if np.isnat(value):
+ return None, None
+ try:
+ return _canonical_timestamp(pd.Timestamp(value))
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid NumPy timestamp"
+ if value_type is np.timedelta64:
+ if np.isnat(value):
+ return None, None
+ try:
+ return pd.Timedelta(value).isoformat(), None
+ except (OverflowError, TypeError, ValueError):
+ return None, "an invalid NumPy duration"
+ return None, "an unsupported or subclassed value"
+
+
+def _normalize_value( # noqa: C901
+ value: Any,
+ budget: _ResultBudget,
+ *,
+ enum_types: frozenset[type[Any]] = frozenset(),
+ metadata: bool = False,
+) -> tuple[Any, str | None]:
+ """Iteratively normalize one bounded exact-container value tree."""
+ stack: list[
+ tuple[Any, list[Any] | dict[str, Any] | None, int | str | None, int,
bool]
+ ] = [(value, None, None, 0, False)]
+ active_containers: set[int] = set()
+ root = value
+
+ while stack:
+ item, parent, slot, depth, leaving = stack.pop()
+ if leaving:
+ active_containers.remove(id(item))
+ continue
+ if depth > MAX_RESULT_VALUE_DEPTH:
+ return None, "excessively nested data"
+ if reason := _charge_value(budget, metadata=metadata):
+ return None, reason
+
+ if type(item) is list:
+ identity = id(item)
+ if identity in active_containers:
+ return None, "cyclic containers"
+ active_containers.add(identity)
+ width = list.__len__(item)
+ if width > MAX_RESULT_VALUE_ITEMS:
+ return None, "an oversized array"
+ if reason := _charge_json_bytes(
+ budget,
+ _container_json_syntax_size(width, mapping=False),
+ metadata=metadata,
+ ):
+ return None, reason
+ stack.append((item, None, None, depth, True))
+ stack.extend(
+ (list.__getitem__(item, index), item, index, depth + 1, False)
+ for index in range(width - 1, -1, -1)
+ )
+ continue
+
+ if type(item) is dict:
+ identity = id(item)
+ if identity in active_containers:
+ return None, "cyclic containers"
+ active_containers.add(identity)
+ width = dict.__len__(item)
+ if width > MAX_RESULT_VALUE_ITEMS:
+ return None, "an oversized object"
+ if reason := _charge_json_bytes(
+ budget,
+ _container_json_syntax_size(width, mapping=True),
+ metadata=metadata,
+ ):
+ return None, reason
+ children: list[tuple[Any, dict[str, Any], str, int, bool]] = []
+ for key, child in dict.items(item):
+ if type(key) is not str:
+ return None, "a non-string object key"
+ if reason := _charge_text(key, budget, key=True,
metadata=metadata):
+ return None, reason
+ children.append((child, item, key, depth + 1, False))
+ stack.append((item, None, None, depth, True))
+ stack.extend(reversed(children))
+ continue
+
+ source_item = item
+ if type(item) in enum_types:
+ normalized = _safe_enum_value(item, enum_types)
+ if normalized is None:
+ return None, "an unsupported enum"
+ item = normalized
+ elif any(base is Enum for base in _type_mro(type(item))):
+ return None, "an enum outside its expected metadata slot"
+
+ normalized, reason = _normalize_scalar(item)
+ if reason is not None:
+ return None, reason
+ max_string_bytes = (
+ MAX_QUERY_RESULT_METADATA_BYTES if metadata else
MAX_RESULT_STRING_LENGTH
+ )
+ if type(normalized) is str:
+ scalar_size = _json_string_size(normalized, max_string_bytes)
+ if scalar_size is None:
+ return None, "invalid text data"
+ else:
+ scalar_size = _normalized_scalar_json_size(normalized)
+ if reason := _charge_json_bytes(
+ budget,
+ scalar_size,
+ metadata=metadata,
+ ):
+ return None, reason
+ if parent is None:
+ root = normalized
+ elif normalized is not source_item:
+ if type(parent) is list:
+ assert type(slot) is int
+ list.__setitem__(parent, slot, normalized)
+ else:
+ assert type(parent) is dict
+ assert type(slot) is str
+ dict.__setitem__(parent, slot, normalized)
+ return root, None
+
+
+def _normalize_metadata_value(
+ payload: dict[str, Any], key: str, budget: _ResultBudget
+) -> str | None:
+ enum_slots: dict[str, frozenset[type[Any]]] = {
+ "status": frozenset({QueryStatus}),
+ "result_format": frozenset({ChartDataResultFormat}),
+ "coltypes": frozenset({GenericDataType}),
+ "applied_filters": frozenset({ExtraFiltersTimeColumnType}),
+ "rejected_filters": frozenset(
+ {ExtraFiltersReasonType, ExtraFiltersTimeColumnType}
+ ),
+ }
+ value = dict.__getitem__(payload, key)
+ normalized, reason = _normalize_value(
+ value,
+ budget,
+ enum_types=enum_slots.get(key, frozenset()),
+ metadata=True,
+ )
+ if reason is None and normalized is not value:
+ dict.__setitem__(payload, key, normalized)
+ return reason
+
+
+def _error_text(value: Any) -> str | None:
+ """Extract a bounded error from an already validated primitive tree."""
+ stack: list[Any] = [value]
+ parts: list[str] = []
+ used = 0
+ while stack and len(parts) < 3 and used < _MAX_ERROR_TEXT_BYTES:
+ item = stack.pop()
+ if type(item) is dict:
+ stack.extend(
+ reversed(
+ [dict.__getitem__(item, key) for key in _ERROR_KEYS if key
in item]
+ )
+ )
+ continue
+ if type(item) is list:
+ stack.extend(
+ list.__getitem__(item, index)
+ for index in range(list.__len__(item) - 1, -1, -1)
+ )
+ continue
+ if item is None or item is False:
+ continue
+ if type(item) is str:
+ remaining = _MAX_ERROR_TEXT_BYTES - used - (2 if parts else 0)
+ encoded = str.encode(item, "utf-8")[:remaining]
+ text = bytes.decode(encoded, "utf-8", errors="ignore")
+ if text:
+ parts.append(text)
+ used += bytes.__len__(encoded) + (2 if len(parts) > 1 else 0)
+ return "; ".join(parts) or None
+
+
+def _failure_for_payload(payload: dict[str, Any], label: str) -> ChartError |
None:
for key in ("error", "errors", "error_message"):
- if message := _query_error_text(payload.get(key)):
+ if key in payload and (message :=
_error_text(dict.__getitem__(payload, key))):
return ChartError(
error=f"{label} failed: {message}", error_type="QueryError"
)
- raw_status = payload.get("status")
- status = str(getattr(raw_status, "value", raw_status) or "")
+ raw_status = dict.get(payload, "status")
+ status = raw_status if type(raw_status) is str else ""
normalized_status = status.strip().casefold().replace("-", "_").replace("
", "_")
if normalized_status in FAILED_QUERY_STATUSES:
message = (
- _query_error_text(payload.get("message"))
- or _query_error_text(payload.get("error_message"))
+ _error_text(dict.get(payload, "message"))
+ or _error_text(dict.get(payload, "error_message"))
or normalized_status
)
return ChartError(error=f"{label} failed: {message}",
error_type="QueryError")
- if payload.get("success") is False:
- message = _query_error_text(payload.get("message")) or "request failed"
+ if dict.get(payload, "success") is False:
+ message = _error_text(dict.get(payload, "message")) or "request failed"
return ChartError(error=f"{label} failed: {message}",
error_type="QueryError")
+ if raw_status is None and "data" not in payload and "queries" not in
payload:
+ if message := _error_text(dict.get(payload, "message")):
+ return ChartError(
+ error=f"{label} failed: {message}", error_type="QueryError"
+ )
+ return None
+
+
+def _metadata_shape_error( # noqa: C901
+ payload: dict[str, Any], label: str, budget: _ResultBudget
+) -> ChartError | None:
+ if "success" in payload and type(dict.__getitem__(payload, "success")) is
not bool:
+ return _invalid_metadata(label)
+ if "status" in payload and type(dict.__getitem__(payload, "status")) is
not str:
+ return _invalid_metadata(label)
if (
- raw_status is None
- and "data" not in payload
- and "queries" not in payload
- and (message := _query_error_text(payload.get("message")))
+ "result_format" in payload
+ and dict.__getitem__(payload, "result_format") not in
_RESULT_FORMAT_VALUES
):
- return ChartError(error=f"{label} failed: {message}",
error_type="QueryError")
+ return _invalid_metadata(label)
+ for count_key in ("rowcount", "sql_rowcount", "total_rows"):
+ if count_key in payload:
+ count = dict.__getitem__(payload, count_key)
+ if count is not None and not (
+ type(count) is int and 0 <= count <= MAX_QUERY_RESULT_ROWCOUNT
+ ):
+ return _invalid_metadata(label)
+ if "is_cached" in payload:
+ cached = dict.__getitem__(payload, "is_cached")
+ if cached is None:
+ if reason := _charge_json_bytes(
+ budget, 1, metadata=True
+ ): # ``false`` vs ``null``
+ return _invalid_result(reason)
+ dict.__setitem__(payload, "is_cached", False)
+ elif type(cached) is not bool:
+ return _invalid_metadata(label)
+ if "cache_timeout" in payload:
+ timeout = dict.__getitem__(payload, "cache_timeout")
+ if timeout is not None and not (
+ type(timeout) is int
+ and (
+ timeout == CACHE_DISABLED_TIMEOUT
+ or 0 <= timeout <= MAX_QUERY_RESULT_CACHE_TIMEOUT
+ )
+ ):
+ return _invalid_metadata(label)
+ if "cache_key" in payload:
+ cache_key = dict.__getitem__(payload, "cache_key")
+ if cache_key is not None and (type(cache_key) is not str or not
cache_key):
+ return _invalid_metadata(label)
+ for timestamp_key in ("cached_dttm", "cache_dttm", "queried_dttm"):
+ if timestamp_key not in payload:
+ continue
+ timestamp = dict.__getitem__(payload, timestamp_key)
+ if timestamp is None:
+ continue
+ if (
+ type(timestamp) is not str
+ or not timestamp
+ or len(timestamp) > MAX_QUERY_RESULT_TIMESTAMP_LENGTH
+ ):
+ return _invalid_metadata(label)
+ normalized = f"{timestamp[:-1]}+00:00" if timestamp.endswith("Z") else
timestamp
+ try:
+ parsed = datetime.fromisoformat(normalized)
+ except ValueError:
+ return _invalid_metadata(label)
+ if parsed.tzinfo is None or parsed.utcoffset() != timedelta(0):
Review Comment:
A normal cache hit supplies `cached_dttm` as Superset's naive UTC timestamp,
but this now rejects every naive timestamp as hostile. That makes the MCP data
and preview paths fail on their second, warm-cache request; can this normalize
the cache timestamp as UTC instead of rejecting it?
--
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]
