sadpandajoe commented on code in PR #43770:
URL: https://github.com/apache/superset/pull/43770#discussion_r4235896143


##########
superset/mcp_service/chart/chart_utils.py:
##########
@@ -3383,7 +3414,620 @@ def _as_column_list(value: Any) -> list[Any]:
     return [value]
 
 
-def analyze_chart_capabilities(viz_type: str | None, config: Any) -> 
ChartCapabilities:
+_NATIVE_TEMPORAL_ROLE_FIELDS: dict[str, frozenset[str]] = {
+    # Typed ``x`` is persisted as native x_axis/granularity_sqla for XY and
+    # Mixed Timeseries. Waterfall exposes the typed field as ``x_axis``.
+    "x_axis": frozenset({"x", "x_axis"}),
+    "granularity_sqla": frozenset({"x", "x_axis", "temporal_column"}),
+    # Chart plugins may designate a chart-specific query role as the implicit
+    # dashboard-time subject.
+    "start_time": frozenset({"start_time"}),
+}
+
+
+def _bullet_token_list(values: Sequence[str | int | float]) -> str:
+    """Serialize typed Bullet controls to the frontend's comma-separated 
form."""
+    tokens: list[str] = []
+    for value in values:
+        if isinstance(value, float):
+            token = repr(value)
+            # ``100`` parses back to the same binary float as ``100.0`` and
+            # preserves the frontend's established compact integer spelling.
+            if token.endswith(".0") and not (
+                value == 0.0 and math.copysign(1.0, value) < 0
+            ):
+                token = token[:-2]
+            tokens.append(token)
+        else:
+            tokens.append(str(value))
+    return ",".join(tokens)
+
+
+def map_bullet_config(config: BulletChartConfig) -> Dict[str, Any]:  # noqa: 
C901
+    """Map typed Bullet config to ``Bullet/buildQuery`` and transformProps.
+
+    The frontend buildQuery replaces the generic query fields with exactly one
+    metric and the groupby hierarchy. Presentation controls stay in native
+    snake_case form_data; the chart plugin camelizes them for transformProps.
+    """
+    if (
+        config.dimensions is None
+        and config._inherited_groupby is None
+        and config.order_by
+    ):
+        # An update resolves its saved hierarchy before mapping. Without one,
+        # creation must validate sort targets against an empty hierarchy.
+        BulletChartConfig.model_validate(
+            {**config.model_dump(exclude_unset=True), "dimensions": []}
+        )
+    metric = create_metric_object(config.metric)
+    form_data: Dict[str, Any] = {
+        "viz_type": "bullet",
+        "metric": metric,
+    }
+
+    # Optional semantic/query fields are emitted only when explicitly supplied.
+    # This lets update_chart and update_chart_preview preserve native saved 
state,
+    # while an explicit empty value still clears it through the generic merge 
path.
+    # The row limit always carries the schema default so a new chart is 
bounded;
+    # update merging restores the saved limit when the caller omits it.
+    if "dimensions" in config.model_fields_set:
+        form_data["groupby"] = [dimension.name for dimension in 
config.dimensions or []]
+    form_data["row_limit"] = config.row_limit
+    if "time_range" in config.model_fields_set:
+        form_data["time_range"] = config.time_range
+
+    if config.order_by:
+        dimensions = config.order_dimensions
+        orderby: list[list[Any]] = []
+        for order in config.order_by:
+            role, index = resolve_bullet_order_target(
+                order.column, dimensions, config.metric
+            )
+            if role == "metric":
+                order_target: Any = metric
+            else:
+                if index is None:  # Defensive: resolver pairs dimensions with 
indexes.
+                    raise ValueError("Bullet dimension order target has no 
index")
+                dimension = dimensions[index]
+                order_target = (
+                    dimension.name if isinstance(dimension, ColumnRef) else 
dimension
+                )
+            orderby.append([order_target, order.ascending])
+        form_data["orderby"] = orderby
+    elif "order_by" in config.model_fields_set:
+        form_data["orderby"] = []
+
+    presentation_fields: dict[str, tuple[str, Any]] = {
+        "ranges": ("ranges", _bullet_token_list(config.ranges)),
+        "range_labels": (
+            "range_labels",
+            _bullet_token_list(config.range_labels),
+        ),
+        "markers": ("markers", _bullet_token_list(config.markers)),
+        "marker_labels": (
+            "marker_labels",
+            _bullet_token_list(config.marker_labels),
+        ),
+        "marker_lines": (
+            "marker_lines",
+            _bullet_token_list(config.marker_lines),
+        ),
+        "marker_line_labels": (
+            "marker_line_labels",
+            _bullet_token_list(config.marker_line_labels),
+        ),
+        "y_axis_format": ("y_axis_format", config.y_axis_format),
+        "show_labels": ("show_labels", config.show_labels),
+        "show_legend": ("show_legend", config.show_legend),
+    }
+    for field_name, (form_key, value) in presentation_fields.items():
+        if field_name in config.model_fields_set:
+            form_data[form_key] = value
+
+    _add_adhoc_filters(form_data, config.filters)
+    if config.filters == [] and "filters" in config.model_fields_set:
+        form_data["adhoc_filters"] = []
+    if config.time_range and config.temporal_column:
+        _ensure_temporal_adhoc_filter(form_data, config.temporal_column)
+        for filter_ in form_data.get("adhoc_filters", []):
+            if (
+                isinstance(filter_, dict)
+                and filter_.get("operator") == 
FilterOperator.TEMPORAL_RANGE.value
+                and filter_.get("subject") == config.temporal_column
+                and filter_.get("comparator") == NO_TIME_RANGE
+            ):
+                filter_["comparator"] = config.time_range
+    return form_data
+
+
+def _normalize_native_filter_aliases(form_data: Mapping[str, Any]) -> 
Dict[str, Any]:
+    """Fold legacy WHERE, HAVING, and filter predicates into adhoc controls."""
+    from superset.utils.core import form_data_to_adhoc, simple_filter_to_adhoc
+
+    normalized = dict(form_data)
+    legacy_filters = [
+        form_data_to_adhoc(normalized, clause)
+        for clause in ("having", "where")
+        if normalized.get(clause)
+    ]
+    legacy_filters.extend(
+        simple_filter_to_adhoc(filter_, "where")
+        for filter_ in normalized.get("filters") or []
+        if filter_ is not None
+    )
+    if legacy_filters:
+        normalized["adhoc_filters"] = [
+            *legacy_filters,
+            *(normalized.get("adhoc_filters") or []),
+        ]
+    for key in ("where", "having", "filters"):
+        normalized.pop(key, None)
+    return normalized
+
+
+def _normalize_bullet_query_aliases(form_data: Mapping[str, Any]) -> Dict[str, 
Any]:
+    """Fold inherited native predicates and ordering into canonical 
controls."""
+    from superset.mcp_service.chart.chart_helpers import _parse_orderby
+
+    normalized = _normalize_native_filter_aliases(form_data)
+    if "groupby" in normalized and not isinstance(normalized["groupby"], list):
+        # Bullet/buildQuery and transformProps read ensureIsArray(groupby).
+        normalized["groupby"] = bullet_groupby_list(normalized["groupby"])
+    if "order_by_cols" in normalized:
+        # Native extractQueryFields concatenates both aliases in key order.
+        ordering: list[Any] = []
+        for key, value in normalized.items():
+            if key == "order_by_cols":
+                ordering.extend(_parse_orderby(value))
+            elif key == "orderby":
+                ordering.extend(value or [])
+        normalized["orderby"] = ordering
+    normalized.pop("order_by_cols", None)
+    return normalized
+
+
+def merge_bullet_form_data(
+    existing_form_data: Mapping[str, Any], new_form_data: Dict[str, Any]
+) -> None:
+    """Preserve omitted native Bullet controls across update tool paths.
+
+    Query roles and every UI control have an explicit typed representation.
+    Mappers emit optional fields only when the caller supplied them, so copying
+    the bounded native keys below preserves omitted state while explicit empty,
+    false, null, and zero-like values remain authoritative.
+    """
+    if (
+        existing_form_data.get("viz_type") != "bullet"
+        or new_form_data.get("viz_type") != "bullet"
+    ):
+        return
+    existing_form_data = _normalize_bullet_query_aliases(existing_form_data)
+    preserved_keys = {
+        "groupby",
+        "adhoc_filters",
+        "time_range",
+        "row_limit",
+        "orderby",
+        "ranges",
+        "range_labels",
+        "markers",
+        "marker_labels",
+        "marker_lines",
+        "marker_line_labels",
+        "y_axis_format",
+        "show_labels",
+        "show_legend",
+        "url_params",
+        # Native query context (dashboard/native filter predicates and time
+        # overrides) that buildQueryObject applies on top of the controls.
+        "extra_form_data",
+        "extra_filters",
+        MCP_DASHBOARD_TIME_FILTER_SUBJECT,
+    }
+
+    # Threshold and label arrays are one frontend control pair. If callers
+    # replace the values without replacing their labels, clear the stale labels
+    # instead of accidentally reassigning them by position.
+    dependent_controls = {
+        "ranges": "range_labels",
+        "markers": "marker_labels",
+        "marker_lines": "marker_line_labels",
+    }
+    for values_key, labels_key in dependent_controls.items():
+        if values_key in new_form_data and labels_key not in new_form_data:
+            new_form_data[labels_key] = ""
+
+    preserve_orderby = (
+        "orderby" not in new_form_data and "orderby" in existing_form_data
+    )
+    for key in preserved_keys:
+        if (
+            key == MCP_DASHBOARD_TIME_FILTER_SUBJECT
+            and "adhoc_filters" in new_form_data
+        ):
+            # The marker describes a mapper-generated temporal filter. Do not
+            # retain stale provenance when an explicit filter update removed 
it.
+            continue
+        if key in existing_form_data and key not in new_form_data:
+            new_form_data[key] = existing_form_data[key]
+    if preserve_orderby:
+        new_form_data["orderby"] = _orderby_for_final_output_roles(
+            existing_form_data, new_form_data
+        )
+
+
+def _bullet_output_labels(
+    form_data: Mapping[str, Any],
+) -> tuple[set[str], dict[str, Any]]:
+    """Return a Bullet state's dimension output labels and metric outputs."""
+    from superset.mcp_service.chart.chart_helpers import _column_label, 
_metric_label
+
+    dimensions = {
+        label
+        for column in bullet_groupby_list(form_data.get("groupby"))
+        if (label := _column_label(column)) is not None
+    }
+    metrics = form_data.get("metrics") or []
+    if not isinstance(metrics, (list, tuple)):
+        metrics = [metrics]
+    metric_outputs = {
+        label: metric
+        for metric in [form_data.get("metric"), *metrics]
+        if (label := _metric_label(metric)) is not None
+    }
+    return dimensions, metric_outputs
+
+
+def bullet_groupby_list(groupby: Any) -> list[Any]:
+    """Normalize a saved Bullet hierarchy like the frontend 
``ensureIsArray``."""
+    if groupby is None:
+        return []
+    return list(groupby) if isinstance(groupby, (list, tuple)) else [groupby]
+
+
+def _orderby_for_final_output_roles(
+    existing_form_data: Mapping[str, Any], new_form_data: Mapping[str, Any]
+) -> Any:
+    """Drop sorts on removed output roles and rebind inherited metric 
expressions.
+
+    Native ordering may also rank by a saved metric or column that is not a
+    displayed output (``get_sqla_query`` resolves it independently). Those
+    sorters never named a Bullet role, so a role change does not remove them.
+    """
+    from superset.mcp_service.chart.chart_helpers import _column_label, 
_metric_label
+
+    saved = existing_form_data.get("orderby")
+    if not isinstance(saved, list):
+        return saved
+    outputs, metric_outputs = _bullet_output_labels(new_form_data)
+    outputs.update(metric_outputs)
+    previous_dimensions, previous_metrics = 
_bullet_output_labels(existing_form_data)
+    previous_outputs = previous_dimensions | set(previous_metrics)
+    retained = []
+    for entry in saved:
+        if isinstance(entry, (list, tuple)) and entry:
+            target = entry[0]
+            label = (
+                _metric_label(target)
+                or _column_label(target)
+                or target.get("metric_name")
+                if isinstance(target, Mapping)
+                else target
+            )
+            if (
+                isinstance(label, str)
+                and label not in outputs
+                and label in previous_outputs
+            ):
+                continue
+            if (
+                isinstance(target, Mapping)
+                and isinstance(label, str)
+                and label in metric_outputs
+            ):
+                # Label equality identifies an output role, not expression
+                # equality: execute the final metric, never the saved 
expression.
+                entry = [metric_outputs[label], *entry[1:]]
+        retained.append(entry)
+    return retained
+
+
+def _filter_identity(filter_: Any) -> tuple[Any, ...] | None:
+    """Return the native identity used when one filter replaces another."""
+    if not isinstance(filter_, Mapping):
+        return None
+    return (
+        filter_.get("clause"),
+        filter_.get("expressionType"),
+        filter_.get("subject"),
+        filter_.get("operator"),
+    )
+
+
+def _temporal_binding_filter(filters: list[Any], subject: Any) -> dict[str, 
Any] | None:
+    """Find the unique filter owned by a recorded MCP temporal marker."""
+    if subject is None:
+        return None
+    if not isinstance(subject, str) or not subject:
+        raise ValueError(
+            "MCP temporal binding provenance subject must be a non-empty 
string"
+        )
+    matches = [
+        filter_
+        for filter_ in filters
+        if isinstance(filter_, dict)
+        and filter_.get("subject") == subject
+        and filter_.get("operator") == FilterOperator.TEMPORAL_RANGE.value
+    ]
+    if len(matches) != 1:
+        raise ValueError(
+            "MCP temporal binding provenance must match exactly one "
+            f"TEMPORAL_RANGE filter for subject {subject!r}; found 
{len(matches)}"
+        )
+    return matches[0]
+
+
+def _append_or_replace_filter(filters: list[Any], filter_: Any) -> None:
+    """Append a filter, replacing the same native role when identifiable."""
+    identity = _filter_identity(filter_)
+    if identity is None:
+        if filter_ not in filters:
+            filters.append(filter_)
+        return
+    filters[:] = [item for item in filters if _filter_identity(item) != 
identity]
+    filters.append(filter_)
+
+
+def _native_temporal_subject_changed(
+    existing_form_data: Mapping[str, Any],
+    new_form_data: Mapping[str, Any],
+    explicit_fields: set[str],
+) -> bool:
+    """Return whether an authoritative native temporal role was replaced.
+
+    Mapping a partial update can propose a dataset fallback binding even when
+    the caller only changed filters. That proposal is not authoritative. A
+    changed x/granularity/chart-specific role is authoritative only when its
+    corresponding typed field was actually supplied.
+    """
+    for native_key, typed_fields in _NATIVE_TEMPORAL_ROLE_FIELDS.items():
+        if explicit_fields.isdisjoint(typed_fields):
+            continue
+        existing_value = existing_form_data.get(native_key)
+        incoming_value = new_form_data.get(native_key)
+        if existing_value != incoming_value:
+            return True
+    return False
+
+
+def _native_temporal_binding(
+    form_data: Mapping[str, Any], filters: list[Any]
+) -> tuple[str | None, dict[str, Any] | None]:
+    """Resolve one binding for a trusted native temporal role, if present."""
+    for native_key in _NATIVE_TEMPORAL_ROLE_FIELDS:
+        subject = form_data.get(native_key)
+        if not isinstance(subject, str) or not subject:
+            continue
+        matches = [
+            filter_
+            for filter_ in filters
+            if isinstance(filter_, dict)
+            and filter_.get("subject") == subject
+            and filter_.get("operator") == FilterOperator.TEMPORAL_RANGE.value
+        ]
+        if len(matches) > 1:
+            raise ValueError(
+                "An authoritative native temporal subject must match at most 
one "
+                f"TEMPORAL_RANGE filter for subject {subject!r}; found "
+                f"{len(matches)}"
+            )
+        if matches:
+            return subject, matches[0]
+    return None, None
+
+
+def merge_update_form_data(  # noqa: C901
+    existing_form_data: Mapping[str, Any],
+    new_form_data: Dict[str, Any],
+    config: ChartConfig,
+) -> None:
+    """Apply the shared omission/provenance contract for chart updates.
+
+    Mapper-generated neutral temporal bindings are infrastructure, not evidence
+    that the caller supplied ``filters`` or changed a saved time-range binding.
+    This helper is used by immediate saves, preview-first saved updates, and
+    cached-preview updates so omission, clear, replacement, and temporal
+    overrides have identical behavior.
+
+    State never crosses a visualization boundary: a viz-type change starts from
+    the mapper's output, so the previous chart's predicates are not restored.
+    """
+    # A missing saved viz_type is a boundary too, matching 
merge_chart_form_data.
+    if existing_form_data.get("viz_type") != new_form_data.get("viz_type"):
+        return
+    existing_form_data = _normalize_native_filter_aliases(existing_form_data)
+    # The initial overlay may carry legacy keys from saved form data. Filter
+    # omission/replacement below owns the complete canonical predicate 
sequence.
+    for key in ("where", "having", "filters"):
+        new_form_data.pop(key, None)
+    existing_filters = list(existing_form_data.get("adhoc_filters") or [])
+    incoming_filters = list(new_form_data.get("adhoc_filters") or [])
+    existing_subject = 
existing_form_data.get(MCP_DASHBOARD_TIME_FILTER_SUBJECT)
+    incoming_subject = new_form_data.get(MCP_DASHBOARD_TIME_FILTER_SUBJECT)
+    existing_binding = _temporal_binding_filter(existing_filters, 
existing_subject)
+    incoming_binding = _temporal_binding_filter(incoming_filters, 
incoming_subject)
+
+    explicit_fields = set(getattr(config, "model_fields_set", set()))
+    filters_explicit = "filters" in explicit_fields
+    range_explicit = "time_range" in explicit_fields
+    subject_explicit = "temporal_column" in explicit_fields
+    native_subject_changed = _native_temporal_subject_changed(
+        existing_form_data, new_form_data, explicit_fields
+    )
+    subject_authoritative = subject_explicit or native_subject_changed
+    if incoming_binding is None:
+        native_subject, native_binding = _native_temporal_binding(
+            new_form_data, incoming_filters
+        )
+        if native_binding is not None:
+            incoming_subject = native_subject
+            incoming_binding = native_binding
+    incoming_user_filters = [
+        filter_ for filter_ in incoming_filters if filter_ is not 
incoming_binding
+    ]
+    temporal_explicit = range_explicit or subject_authoritative
+    if (
+        existing_binding is None
+        and incoming_binding is not None
+        and isinstance(incoming_subject, str)
+        and "filters" not in explicit_fields
+        and temporal_explicit
+    ):
+        # A saved temporal filter that Explore wrote has no MCP provenance
+        # marker. When it is the only native filter for the incoming subject,
+        # the update replaces it in place instead of appending a duplicate.
+        native_matches = [
+            filter_
+            for filter_ in existing_filters
+            if isinstance(filter_, dict)
+            and filter_.get("subject") == incoming_subject
+            and filter_.get("operator") == FilterOperator.TEMPORAL_RANGE.value
+        ]
+        if len(native_matches) == 1:
+            existing_binding = native_matches[0]
+            existing_subject = incoming_subject
+
+    chosen_binding: dict[str, Any] | None = None
+    chosen_subject: Any = None
+    if not filters_explicit:
+        # Omission is byte-faithful: keep the native sequence in its exact 
order,
+        # including SQL/HAVING objects and a provenance-owned binding at any 
index.
+        merged_filters = list(existing_filters)
+        chosen_binding = existing_binding
+        chosen_subject = existing_subject
+        if temporal_explicit:
+            if subject_authoritative:
+                chosen_binding = incoming_binding
+                chosen_subject = incoming_subject
+            elif existing_binding is not None:
+                # A range-only update belongs to the saved subject, even when
+                # mapping the partial config proposed the dataset main_dttm.
+                chosen_binding = dict(existing_binding)
+                chosen_subject = existing_subject
+            else:
+                chosen_binding = incoming_binding
+                chosen_subject = incoming_subject
+
+            if chosen_binding is not None:
+                chosen_binding = dict(chosen_binding)
+                if range_explicit:
+                    chosen_binding["comparator"] = (
+                        getattr(config, "time_range", None) or NO_TIME_RANGE
+                    )
+                elif existing_binding is not None:
+                    # Subject-only replacement preserves the saved active or
+                    # neutral range instead of resetting it to No filter.
+                    chosen_binding["comparator"] = existing_binding.get(
+                        "comparator", NO_TIME_RANGE
+                    )
+            if existing_binding is not None:
+                binding_index = next(
+                    index
+                    for index, filter_ in enumerate(merged_filters)
+                    if filter_ is existing_binding
+                )
+                if chosen_binding is None:
+                    merged_filters.pop(binding_index)
+                else:
+                    # A temporal override changes infrastructure in place 
instead
+                    # of moving it past surrounding native filters.
+                    merged_filters[binding_index] = chosen_binding
+            elif chosen_binding is not None:
+                merged_filters.append(chosen_binding)

Review Comment:
   A range-only update on a chart whose temporal filter was saved from Explore 
(no MCP provenance marker) on a column other than the dataset's main time 
column won't reuse that filter. Here `existing_binding` is `None` (the marker 
lookup returns nothing, and the native-match fallback above only looks for the 
mapper's proposed subject), so the dataset-default binding is appended and the 
saved filter stays in place. E.g. a saved `ShipDate TEMPORAL_RANGE "Last year"` 
with main column `OrderDate`, updated with `time_range: "Last month"`, ends up 
with both filters, and `time_range: null` can't clear the `ShipDate` 
restriction. Should an unmarked, single native TEMPORAL_RANGE filter be adopted 
as the saved binding when `temporal_column` is omitted?



-- 
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]

Reply via email to