Refactoring DataGrid to use DataService.py

docs/DataGrid Refactoring.md

@@ -0,0 +1,116 @@
+# DataGrid Refactoring
+
+## Objective
+
+Clearly separate data management and rendering responsibilities in the DataGrid system.
+The current architecture mixes data mutation, formula computation, and rendering in the
+same `DataGrid` class, which complicates cross-table formula management and code reasoning.
+
+## Guiding Principles
+
+- `DataService` can exist without rendering. The reverse is not true.
+- All data mutations go through `DataService`.
+- Columns have two facets: data semantics (`ColumnDefinition`) and UI presentation (`ColumnUiState`).
+- No more parent hierarchy where avoidable — access via `InstancesManager.get_by_type()`.
+- The persistence key is `grid_id` (stable), not `table_name` (can change over time).
+
+---
+
+## New Classes (`core/data/`)
+
+### `DataServicesManager` — SingleInstance
+
+- Owns the `FormulaEngine` (cross-table formula coordination)
+- Creates `DataService` instances on demand from `DataGridsManager`
+- Provides access to `DataService` instances by `grid_id`
+- Provides the resolver callback for `FormulaEngine`: `grid_id → DataStore`
+
+### `DataService` — companion to `DataGrid`
+
+- Owns `DataStore` and `list[ColumnDefinition]`
+- Holds a reference to `DataServicesManager` for `FormulaEngine` access
+- Methods: `load_dataframe(df)`, `add_row()`, `add_column()`, `set_data(col_id, row_index, value)`
+- Mutations call `mark_data_changed()` → set dirty flag
+- `ensure_ready()` → recalculates formulas if dirty (called by `mk_body_content_page()`)
+- Can exist without any rendering
+
+### `DataStore` — renamed from `DatagridStore`
+
+- Pure persistence: `ne_df`, `ns_fast_access`, `ns_row_data`, `ns_total_rows`
+- `DbObject` with no business logic
+
+### `ColumnDefinition`
+
+- Data semantics: `col_id`, `title`, `type`, `formula`, `col_index`
+
+---
+
+## Modified Classes
+
+### `DataGridsRegistry` — streamlined
+
+- Persistence only: `put()`, `remove()`, `get_all_entries()`
+- **Loses**: `get_columns()`, `get_column_type()`, `get_column_values()`, `get_row_count()`
+
+### `DatagridMetadataProvider` — becomes a concrete SingleInstance
+
+- No longer abstract / interface (only one concrete implementation exists)
+- Reads from `DataServicesManager` and `DataGridsRegistry`
+- Holds: `style_presets`, `formatter_presets`, `all_tables_formats`
+- Exposes: `list_tables()`, `list_columns()`, `list_column_values()`, `get_column_type()`,
+  `list_style_presets()`, `list_format_presets()`
+
+### `DataGridsManager` — pure UI
+
+- **Keeps**: `TreeView`, `TabsManager`, document state, `Commands`
+- **Loses**: `FormulaEngine`, presets, `DatagridMetadataProvider`, `_resolve_store_for_table()`
+
+### `DataGrid` — pure rendering
+
+- **Keeps**: `mk_*`, `render()`, `__ft__()`, `_state`, `_settings`
+- **Keeps**: `_apply_sort()`, `_apply_filter()`, `_get_filtered_df()`
+- **Loses**: `add_new_row()`, `add_new_column()`, `init_from_dataframe()`,
+  `_recalculate_formulas()`, `_register_existing_formulas()`, `_df_store`
+- Accesses its `DataService` via its `grid_id`:
+  `InstancesManager.get_by_type(DataServicesManager).get_service(grid_id)`
+- `mk_body_content_page()` calls `data_service.ensure_ready()` before rendering
+
+### `DatagridState`
+
+- `columns` changes from `list[DataGridColumnState]` → `list[ColumnUiState]`
+- Everything else remains unchanged
+
+### `DataGridColumnState` — split into two classes
+
+| Class | Belongs to | Fields |
+|---|---|---|
+| `ColumnDefinition` | `DataService` | `col_id`, `title`, `type`, `formula`, `col_index` |
+| `ColumnUiState` | `DatagridState` | `col_id`, `width`, `visible`, `format` |
+
+---
+
+## Structural Fix
+
+**Current bug**: `mark_data_changed()` is defined in `FormulaEngine` but is never called
+by `DataGrid`. Formulas are only recalculated defensively at render time.
+
+**After refactoring**:
+- Every mutation in `DataService` calls `mark_data_changed()` → dirty flag set
+- `mk_body_content_page()` calls `data_service.ensure_ready()` → recalculates if dirty
+- Multiple mutations before a render = a single recalculation
+
+---
+
+## Progress Tracking
+
+- [x] Create `DataStore` (rename `DatagridStore`)
+- [x] Create `ColumnDefinition`
+- [x] Create `DataService`
+- [x] Create `DataServicesManager`
+- [x] Refactor `DataGridsRegistry` (streamline)
+- [x] Refactor `DatagridMetadataProvider` (make concrete)
+- [x] Refactor `DataGridsManager` (pure UI)
+- [x] Refactor `DataGrid` (pure rendering, split `DataGridColumnState`)
+- [x] Update tests
+- [ ] Remove `init_from_dataframe` from `DataGrid` (kept temporarily for transition)
+- [ ] Full split of `DataGridColumnState` into `ColumnDefinition` + `ColumnUiState` in `DatagridState`

src/app.py

@@ -1,9 +1,7 @@
-import json
 import logging.config
 
-import pandas as pd
 import yaml
-from dbengine.handlers import BaseRefHandler, handlers
+from dbengine.handlers import handlers
 from fasthtml import serve
 from fasthtml.components import Div
 
@@ -15,7 +13,6 @@ from myfasthtml.controls.InstancesDebugger import InstancesDebugger
 from myfasthtml.controls.Keyboard import Keyboard
 from myfasthtml.controls.Layout import Layout
 from myfasthtml.controls.TabsManager import TabsManager
-from myfasthtml.controls.TreeView import TreeView, TreeNode
 from myfasthtml.controls.helpers import Ids, mk
 from myfasthtml.core.dbengine_utils import DataFrameHandler
 from myfasthtml.core.instances import UniqueInstance
@@ -39,53 +36,6 @@ app, rt = create_app(protect_routes=True,
                      base_url="http://localhost:5003")
 
 
-
-def create_sample_treeview(parent):
-  """
-  Create a sample TreeView with a file structure for testing.
-
-  Args:
-      parent: Parent instance for the TreeView
-
-  Returns:
-      TreeView: Configured TreeView instance with sample data
-  """
-  tree_view = TreeView(parent, _id="-treeview")
-  
-  # Create sample file structure
-  projects = TreeNode(label="Projects", type="folder")
-  tree_view.add_node(projects)
-  
-  myfasthtml = TreeNode(label="MyFastHtml", type="folder")
-  tree_view.add_node(myfasthtml, parent_id=projects.id)
-  
-  app_py = TreeNode(label="app.py", type="file")
-  tree_view.add_node(app_py, parent_id=myfasthtml.id)
-  
-  readme = TreeNode(label="README.md", type="file")
-  tree_view.add_node(readme, parent_id=myfasthtml.id)
-  
-  src_folder = TreeNode(label="src", type="folder")
-  tree_view.add_node(src_folder, parent_id=myfasthtml.id)
-  
-  controls_py = TreeNode(label="controls.py", type="file")
-  tree_view.add_node(controls_py, parent_id=src_folder.id)
-  
-  documents = TreeNode(label="Documents", type="folder")
-  tree_view.add_node(documents, parent_id=projects.id)
-  
-  notes = TreeNode(label="notes.txt", type="file")
-  tree_view.add_node(notes, parent_id=documents.id)
-  
-  todo = TreeNode(label="todo.md", type="file")
-  tree_view.add_node(todo, parent_id=documents.id)
-  
-  # Expand all nodes to show the full structure
-  # tree_view.expand_all()
-  
-  return tree_view
-
-
 @rt("/")
 def index(session):
   session_instance = UniqueInstance(session=session,
@@ -120,19 +70,15 @@ def index(session):
   btn_popup = mk.label("Popup",
                        command=add_tab("Popup", Dropdown(layout, "Content", button="button", _id="-dropdown")))
   
-  # Create TreeView with sample data
-  tree_view = create_sample_treeview(layout)
-  
   layout.header_left.add(tabs_manager.add_tab_btn())
   layout.header_right.add(btn_show_right_drawer)
   layout.left_drawer.add(btn_show_instances_debugger, "Debugger")
   layout.left_drawer.add(btn_show_commands_debugger, "Debugger")
   layout.left_drawer.add(btn_file_upload, "Test")
   layout.left_drawer.add(btn_popup, "Test")
-  layout.left_drawer.add(tree_view, "TreeView")
   
   # data grids
-  dgs_manager = DataGridsManager(layout, _id="-datagrids")
+  dgs_manager = DataGridsManager(session_instance)
   layout.left_drawer.add_group("Documents", Div("Documents",
                                                 dgs_manager.mk_main_icons(),
                                                 cls="mf-layout-group flex gap-3"))

src/myfasthtml/controls/DataGrid.py

@@ -5,10 +5,8 @@ from dataclasses import dataclass
 from functools import lru_cache
 from typing import Optional
 
-import pandas as pd
 from fasthtml.common import NotStr
 from fasthtml.components import *
-from pandas import DataFrame
 
 from myfasthtml.controls.BaseCommands import BaseCommands
 from myfasthtml.controls.CycleStateControl import CycleStateControl
@@ -17,23 +15,28 @@ from myfasthtml.controls.DataGridFormattingEditor import DataGridFormattingEdito
 from myfasthtml.controls.DslEditor import DslEditorConf
 from myfasthtml.controls.IconsHelper import IconsHelper
 from myfasthtml.controls.Keyboard import Keyboard
+from myfasthtml.controls.Mouse import Mouse
 from myfasthtml.controls.Panel import Panel, PanelConf
 from myfasthtml.controls.Query import Query, QUERY_FILTER
-from myfasthtml.controls.datagrid_objects import DataGridColumnState, DataGridRowState, \
-  DatagridSelectionState, DataGridHeaderFooterConf, DatagridEditionState
-from myfasthtml.controls.helpers import mk, column_type_defaults
+from myfasthtml.controls.datagrid_objects import DataGridColumnState, DataGridRowUiState, \
+  DatagridSelectionState, DataGridHeaderFooterConf, DatagridEditionState, DataGridColumnUiState, \
+  DataGridRowSelectionColumnState
+from myfasthtml.controls.helpers import mk
 from myfasthtml.core.commands import Command
-from myfasthtml.core.constants import ColumnType, ROW_INDEX_ID, FooterAggregation, DATAGRID_PAGE_SIZE, FILTER_INPUT_CID, \
-  ROW_SELECTION_ID
+from myfasthtml.core.constants import ColumnType, FooterAggregation, DATAGRID_PAGE_SIZE, FILTER_INPUT_CID
+from myfasthtml.core.data.ColumnDefinition import ColumnDefinition
+from myfasthtml.core.data.DataService import DataService
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
 from myfasthtml.core.dbmanager import DbObject
 from myfasthtml.core.dsls import DslsManager
 from myfasthtml.core.formatting.dsl.completion.FormattingCompletionEngine import FormattingCompletionEngine
+from myfasthtml.core.formatting.dsl.completion.provider import DatagridMetadataProvider
 from myfasthtml.core.formatting.dsl.definition import FormattingDSL
 from myfasthtml.core.formatting.dsl.parser import DSLParser
 from myfasthtml.core.formatting.engine import FormattingEngine
-from myfasthtml.core.instances import MultipleInstance
+from myfasthtml.core.instances import MultipleInstance, InstancesManager
 from myfasthtml.core.optimized_ft import OptimizedDiv
-from myfasthtml.core.utils import make_safe_id, merge_classes, make_unique_safe_id, is_null
+from myfasthtml.core.utils import merge_classes, is_null
 from myfasthtml.icons.carbon import row, column, grid
 from myfasthtml.icons.fluent import checkbox_unchecked16_regular
 from myfasthtml.icons.fluent_p2 import checkbox_checked16_regular, column_edit20_regular, add12_filled
@@ -70,8 +73,8 @@ class DatagridState(DbObject):
       super().__init__(owner, name=f"{owner.get_id()}#state", save_state=save_state)
       self.sidebar_visible: bool = False
       self.selected_view: str = None
-      self.columns: list[DataGridColumnState] = []
-      self.rows: list[DataGridRowState] = []  # only the rows that have a specific state
+      self.columns: list[DataGridColumnUiState] = []
+      self.rows: list[DataGridRowUiState] = []  # only the rows that have a specific state
       self.headers: list[DataGridHeaderFooterConf] = []
       self.footers: list[DataGridHeaderFooterConf] = []
       self.sorted: list = []
@@ -101,20 +104,6 @@ class DatagridSettings(DbObject):
       self.enable_edition: bool = True
 
 
-class DatagridStore(DbObject):
-  """
-  Store Dataframes
-  """
-  
-  def __init__(self, owner, save_state):
-    with self.initializing():
-      super().__init__(owner, name=f"{owner.get_id()}#df", save_state=save_state)
-      self.ne_df = None
-      self.ns_fast_access = None
-      self.ns_row_data = None
-      self.ns_total_rows = None
-
-
 class Commands(BaseCommands):
   def get_page(self, page_index: int):
     return Command("GetPage",
@@ -231,11 +220,15 @@ class DataGrid(MultipleInstance):
     name, namespace = (conf.name, conf.namespace) if conf else ("No name", "__default__")
     self._settings = DatagridSettings(self, save_state=save_state, name=name, namespace=namespace)
     self._state = DatagridState(self, save_state=self._settings.save_state)
-    self._df_store = DatagridStore(self, save_state=self._settings.save_state)
     self._formatting_engine = FormattingEngine()
     self._columns = None
     self.commands = Commands(self)
-    self.init_from_dataframe(self._df_store.ne_df, init_state=False)  # data comes from DatagridStore
+    
+    # Obtain DataService from DataServicesManager (no parent hierarchy)
+    data_services_manager = InstancesManager.get_by_type(self._session, DataServicesManager)
+    data_service_id = self.get_data_service_id_from_data_grid_id(self._id)
+    self._data_service = data_services_manager.restore_service(data_service_id)
+    self._init_columns()
     
     # add Panel
     self._panel = Panel(self, conf=PanelConf(show_display_right=False), _id="-panel")
@@ -253,15 +246,17 @@ class DataGrid(MultipleInstance):
     
     # add Selection Selector
     selection_types = {
+        "cell": mk.icon(grid, tooltip="Cell selection"),  # default
         "row": mk.icon(row, tooltip="Row selection"),
         "column": mk.icon(column, tooltip="Column selection"),
-        "cell": mk.icon(grid, tooltip="Cell selection")
     }
     self._selection_mode_selector = CycleStateControl(self,
                                                       controls=selection_types,
                                                       save_state=False,
                                                       _id="-cycle_state")
     self._selection_mode_selector.bind_command("CycleState", self.commands.change_selection_mode())
+    if self._state.selection.selection_mode is None:
+      self.change_selection_mode()
     
     # add columns manager
     self._columns_manager = DataGridColumnsManager(self)
@@ -269,7 +264,8 @@ class DataGrid(MultipleInstance):
     self._columns_manager.bind_command("SaveColumnDetails", self.commands.on_column_changed())
     
     if self._settings.enable_formatting:
-      completion_engine = FormattingCompletionEngine(self._parent, self.get_table_name())
+      provider = InstancesManager.get_by_type(self._session, DatagridMetadataProvider, None)
+      completion_engine = FormattingCompletionEngine(provider, self.get_table_name())
       editor_conf = DslEditorConf(engine_id=completion_engine.get_id())
       dsl = FormattingDSL()
       self._formatting_editor = DataGridFormattingEditor(self,
@@ -301,7 +297,13 @@ class DataGrid(MultipleInstance):
   
   @property
   def _df(self):
-    return self._df_store.ne_df
+    if self._data_service is None:
+      return None
+    return self._data_service.get_store().ne_df
+  
+  @property
+  def _fast_access(self):
+    return self._data_service.get_store().ns_fast_access
   
   def _apply_sort(self, df):
     if df is None:
@@ -344,7 +346,8 @@ class DataGrid(MultipleInstance):
     df = self._df.copy()
     df = self._apply_sort(df)  # need to keep the real type to sort
     df = self._apply_filter(df)
-    self._df_store.ns_total_rows = len(df)
+    if self._data_service is not None:
+      self._data_service.get_store().ns_total_rows = len(df)
     
     return df
   
@@ -375,36 +378,6 @@ class DataGrid(MultipleInstance):
     self._state.selection.selected = pos
     self._state.save()
   
-  def _register_existing_formulas(self) -> None:
-    """
-    Re-register all formula columns with the FormulaEngine.
-
-    Called after data reload to ensure the engine knows about all
-    formula columns and their expressions.
-    """
-    engine = self.get_formula_engine()
-    if engine is None:
-      return
-    table = self.get_table_name()
-    for col_def in self._state.columns:
-      if col_def.formula:
-        try:
-          engine.set_formula(table, col_def.col_id, col_def.formula)
-        except Exception as e:
-          logger.warning("Failed to register formula for %s.%s: %s", table, col_def.col_id, e)
-  
-  def _recalculate_formulas(self) -> None:
-    """
-    Recalculate dirty formula columns before rendering.
-
-    Called at the start of mk_body_content_page() to ensure formula
-    columns are up-to-date before cells are rendered.
-    """
-    engine = self.get_formula_engine()
-    if engine is None:
-      return
-    engine.recalculate_if_needed(self.get_table_name(), self._df_store)
-  
   def _get_format_rules(self, col_pos, row_index, col_def):
     """
     Get format rules for a cell, returning only the most specific level defined.
@@ -440,197 +413,55 @@ class DataGrid(MultipleInstance):
     if self._state.table_format:
       return self._state.table_format
     
-    # Get global tables formatting from manager
-    return self._parent.all_tables_formats
+    # Get global tables formatting from DatagridMetadataProvider
+    provider = InstancesManager.get_by_type(self._session, DatagridMetadataProvider, None)
+    return provider.all_tables_formats if provider is not None else []
   
   def _init_columns(self):
-    self._columns = self._state.columns.copy()
+    # Populate UI state from DataService columns when creating a new grid
+    columns_defs = self._data_service.columns
+    
+    columns = []
+    if self._state.columns:
+      # we need to make sure that we match the DataGridColumnUiState and the ColumnDefinition
+      for col_ui_state in self._state.columns:
+        col_def = [col_def for col_def in columns_defs if col_def.col_id == col_ui_state.col_id][0]
+        columns.append(DataGridColumnState(col_def, col_ui_state))
+      
+      self._columns = columns
+    
+    else:
+      # init the ui states
+      ui_states = [DataGridColumnUiState(col_def.col_id) for col_def in columns_defs]
+      self._state.columns = ui_states  # this saves the state of the columns
+      
+      self._columns = [DataGridColumnState(col_def, col_ui_state)
+                       for col_def, col_ui_state in zip(columns_defs, ui_states)]
+    
     if self._settings.enable_edition:
-      self._columns.insert(0, DataGridColumnState(ROW_SELECTION_ID,
-                                                  -1,
-                                                  "",
-                                                  ColumnType.RowSelection_))
-  
-  def init_from_dataframe(self, df, init_state=True):
-    
-    def _get_column_type(dtype):
-      if pd.api.types.is_integer_dtype(dtype):
-        return ColumnType.Number
-      elif pd.api.types.is_float_dtype(dtype):
-        return ColumnType.Number
-      elif pd.api.types.is_bool_dtype(dtype):
-        return ColumnType.Bool
-      elif pd.api.types.is_datetime64_any_dtype(dtype):
-        return ColumnType.Datetime
-      else:
-        return ColumnType.Text  # Default to Text if no match
-    
-    def _init_columns(_df):
-      columns = [DataGridColumnState(make_safe_id(col_id),
-                                     col_index,
-                                     col_id,
-                                     _get_column_type(self._df[make_safe_id(col_id)].dtype))
-                 for col_index, col_id in enumerate(_df.columns)]
-      
-      return columns
-    
-    def _init_fast_access(_df):
-      """
-      Generates a fast-access dictionary for a DataFrame.
-
-      This method converts the columns of the provided DataFrame into NumPy arrays
-      and stores them as values in a dictionary, using the column names as keys.
-      This allows for efficient access to the data stored in the DataFrame.
-
-      Args:
-          _df (DataFrame): The input pandas DataFrame whose columns are to be converted
-          into a dictionary of NumPy arrays.
-
-      Returns:
-          dict: A dictionary where the keys are the column names of the input DataFrame
-          and the values are the corresponding column values as NumPy arrays.
-      """
-      if _df is None:
-        return {}
-      
-      res = {col: _df[col].to_numpy() for col in _df.columns}
-      res[ROW_INDEX_ID] = _df.index.to_numpy()
-      return res
-    
-    def _init_row_data(_df):
-      """
-      Generates a list of row data dictionaries for column references in formatting conditions.
-
-      Each dict contains {col_id: value} for a single row, used by FormattingEngine
-      to evaluate conditions that reference other columns (e.g., {"col": "budget"}).
-
-      Args:
-          _df (DataFrame): The input pandas DataFrame.
-
-      Returns:
-          list[dict]: A list where each element is a dict of column values for that row.
-      """
-      if _df is None:
-        return []
-      
-      return _df.to_dict(orient='records')
-    
-    if df is not None:
-      self._df_store.ne_df = df
-      if init_state:
-        self._df.columns = self._df.columns.map(make_safe_id)  # make sure column names are trimmed
-        self._df_store.save()
-        self._state.rows = []  # sparse: only rows with non-default state are stored
-        self._state.columns = _init_columns(df)  # use df not self._df to keep the original title
-      
-      self._init_columns()
-      self._df_store.ns_fast_access = _init_fast_access(self._df)
-      self._df_store.ns_row_data = _init_row_data(self._df)
-      self._df_store.ns_total_rows = len(self._df) if self._df is not None else 0
-      self._register_existing_formulas()
-    
-    return self
+      self._columns.insert(0, DataGridRowSelectionColumnState())
   
   def add_new_column(self, col_def: DataGridColumnState) -> None:
-    """Add a new column to the DataGrid.
-
-    For Formula columns, only _state.columns is updated; the FormulaEngine
-    computes values on demand via recalculate_if_needed().
-
-    For other column types, also adds the column to the DataFrame and updates
-    ns_fast_access and ns_row_data incrementally with type-appropriate defaults.
+    """Add a new column, delegating data mutation to DataService.
 
     Args:
-        col_def: Column definition with title and type already set.
-                 col_id is derived from title via make_safe_id.
+        col_def: Column definition with title and type set. col_id will be
+            assigned by DataService.
     """
-    col_def.col_id = make_unique_safe_id(col_def.title, [c.col_id for c in self._state.columns])
-    
-    if col_def.type == ColumnType.Formula:
-      col_def.col_index = -1
-      self._state.columns.append(col_def)
-      return
-    
-    default_value = column_type_defaults.get(col_def.type, "")
-    col_def.col_index = len(self._df.columns) if self._df is not None else 0
-    self._state.columns.append(col_def)
-    
-    if self._df is not None:
-      self._df_store.ne_df[col_def.col_id] = default_value
-      self._df_store.ns_fast_access[col_def.col_id] = self._df_store.ne_df[col_def.col_id].to_numpy()
-      for row_dict in self._df_store.ns_row_data:
-        row_dict[col_def.col_id] = default_value
-      self._df_store.save()
-  
-  def add_new_row(self, row_data: dict = None) -> None:
-    """Add a new row to the DataGrid with incremental updates.
-
-    Creates default values based on column types and handles formula columns.
-    Updates ns_fast_access and ns_row_data incrementally for better performance.
-
-    Args:
-        row_data: Optional dict with initial values. If None, uses type defaults.
-    """
-    import numpy as np
-    
-    def _get_default_value(col_def: DataGridColumnState):
-      """Get default value for a column based on its type."""
-      if col_def.type == ColumnType.Formula:
-        return None  # Will be calculated by FormulaEngine
-      return column_type_defaults.get(col_def.type, "")
-    
-    if row_data is None:
-      # Create default values for all non-formula, non-selection columns
-      row_data = {}
-      for col in self._state.columns:
-        if col.type not in (ColumnType.Formula, ColumnType.RowSelection_):
-          row_data[col.col_id] = _get_default_value(col)
-    
-    # 1. Add row to DataFrame (only non-formula columns)
-    new_index = len(self._df)
-    self._df.loc[new_index] = row_data
-    
-    # 2. Incremental update of ns_fast_access
-    for col_id, value in row_data.items():
-      if col_id in self._df_store.ns_fast_access:
-        self._df_store.ns_fast_access[col_id] = np.append(
-          self._df_store.ns_fast_access[col_id],
-          value
-        )
-      else:
-        # First value for this column (rare case)
-        self._df_store.ns_fast_access[col_id] = np.array([value])
-    
-    # Update row index
-    if ROW_INDEX_ID in self._df_store.ns_fast_access:
-      self._df_store.ns_fast_access[ROW_INDEX_ID] = np.append(
-        self._df_store.ns_fast_access[ROW_INDEX_ID],
-        new_index
+    if self._data_service is not None:
+      data_col = ColumnDefinition(
+        col_id="",
+        col_index=col_def.col_index,
+        title=col_def.title,
+        type=col_def.type,
+        formula=col_def.formula,
       )
-    else:
-      self._df_store.ns_fast_access[ROW_INDEX_ID] = np.array([new_index])
-    
-    # 3. Incremental update of ns_row_data
-    self._df_store.ns_row_data.append(row_data.copy())
-    
-    # 4. Handle formula columns
-    engine = self.get_formula_engine()
-    if engine is not None:
-      table_name = self.get_table_name()
-      # Check if there are any formula columns
-      has_formulas = any(col.type == ColumnType.Formula for col in self._state.columns)
-      if has_formulas:
-        try:
-          # Recalculate all formulas (engine handles efficiency)
-          engine.recalculate_if_needed(table_name, self._df_store)
-        except Exception as e:
-          logger.warning(f"Failed to calculate formulas for new row: {e}")
-    
-    # 5. Update total row count
-    self._df_store.ns_total_rows = len(self._df)
-    
-    # Save changes
-    self._df_store.save()
+      self._data_service.add_column(data_col)
+      col_def.col_id = data_col.col_id
+      col_def.col_index = data_col.col_index
+    self._state.columns.append(col_def)
+    self._init_columns()
+    self._state.save()
   
   def move_column(self, source_col_id: str, target_col_id: str):
     """Move column to new position. Called via Command from JS."""
@@ -676,7 +507,7 @@ class DataGrid(MultipleInstance):
     Returns:
         Optimal width in pixels (between 50 and 500)
     """
-    col_def = next((c for c in self._state.columns if c.col_id == col_id), None)
+    col_def = next((c for c in self._columns if c.col_id == col_id), None)
     if not col_def:
       logger.warning(f"calculate_optimal_column_width: column not found {col_id=}")
       return 150  # default width
@@ -686,8 +517,8 @@ class DataGrid(MultipleInstance):
     
     # Max data length
     max_data_length = 0
-    if col_id in self._df_store.ns_fast_access:
-      col_array = self._df_store.ns_fast_access[col_id]
+    if col_id in self._fast_access:
+      col_array = self._fast_access[col_id]
       if col_array is not None and len(col_array) > 0:
         max_data_length = max(len(str(v)) for v in col_array)
     
@@ -774,7 +605,6 @@ class DataGrid(MultipleInstance):
   def change_selection_mode(self):
     logger.debug(f"change_selection_mode")
     new_state = self._selection_mode_selector.get_state()
-    logger.debug(f"  {new_state=}")
     self._state.selection.selection_mode = new_state
     self._state.save()
     return self.render_partial()
@@ -812,7 +642,8 @@ class DataGrid(MultipleInstance):
     self._state.save()
   
   def handle_add_row(self):
-    self.add_new_row()  # Add row to data
+    if self._data_service is not None:
+      self._data_service.add_row()
     row_index = len(self._df) - 1  # Index of the newly added row
     len_columns_1 = len(self._columns) - 1
     rows = [self.mk_row(row_index, None, len_columns_1)]
@@ -835,8 +666,17 @@ class DataGrid(MultipleInstance):
     return f"{self._settings.namespace}.{self._settings.name}" if self._settings.namespace else self._settings.name
   
   def get_formula_engine(self):
-    """Return the FormulaEngine from the DataGridsManager, if available."""
-    return self._parent.get_formula_engine()
+    """Return the shared FormulaEngine from DataServicesManager."""
+    manager = InstancesManager.get_by_type(self._session, DataServicesManager, None)
+    return manager.get_formula_engine() if manager is not None else None
+  
+  @staticmethod
+  def get_grid_id_from_data_service_id(data_service_id):
+    return data_service_id.replace(DataService.compute_prefix(), DataGrid.compute_prefix(), 1)
+  
+  @staticmethod
+  def get_data_service_id_from_data_grid_id(datagrid_id):
+    return datagrid_id.replace(DataGrid.compute_prefix(), DataService.compute_prefix(), 1)
   
   def mk_headers(self):
     resize_cmd = self.commands.set_column_width()
@@ -917,7 +757,7 @@ class DataGrid(MultipleInstance):
     # Formula column: safe read — ns_fast_access entry may not exist yet if the
     # engine hasn't run its first recalculation pass.
     if column_type == ColumnType.Formula:
-      col_array = self._df_store.ns_fast_access.get(col_def.col_id)
+      col_array = self._fast_access.get(col_def.col_id)
       if col_array is None or row_index >= len(col_array):
         return NotStr('<span class="dt2-cell-content-text truncate">—</span>')
       value = col_array[row_index]
@@ -932,7 +772,7 @@ class DataGrid(MultipleInstance):
       else:
         column_type = ColumnType.Text
     else:
-      value = self._df_store.ns_fast_access[col_def.col_id][row_index]
+      value = self._fast_access[col_def.col_id][row_index]
     
     # Boolean type - uses cached HTML (only 2 possible values)
     if column_type == ColumnType.Bool:
@@ -981,7 +821,7 @@ class DataGrid(MultipleInstance):
     if col_def.type == ColumnType.RowSelection_:
       return OptimizedDiv(cls="dt2-row-selection")
     
-    col_array = self._df_store.ns_fast_access.get(col_def.col_id)
+    col_array = self._fast_access.get(col_def.col_id)
     value = col_array[row_index] if col_array is not None and row_index < len(col_array) else None
     content = self.mk_body_cell_content(col_pos, row_index, col_def, filter_keyword_lower)
     
@@ -1013,7 +853,8 @@ class DataGrid(MultipleInstance):
     OPTIMIZED: Extract filter keyword once instead of 10,000 times.
     OPTIMIZED: Uses OptimizedDiv for rows instead of Div for faster rendering.
     """
-    self._recalculate_formulas()
+    if self._data_service is not None:
+      self._data_service.ensure_ready()
     df = self._get_filtered_df()
     if df is None:
       return []
@@ -1026,7 +867,8 @@ class DataGrid(MultipleInstance):
     len_columns_1 = len(self._columns) - 1
     
     rows = [self.mk_row(row_index, filter_keyword_lower, len_columns_1) for row_index in df.index[start:end]]
-    if self._df_store.ns_total_rows > end:
+    store = self._data_service.get_store() if self._data_service is not None else None
+    if store is not None and store.ns_total_rows > end:
       rows[-1].attrs.extend(self.commands.get_page(page_index + 1).get_htmx_params(escaped=True))
     
     self._mk_append_add_row_if_needed(rows)
@@ -1184,7 +1026,7 @@ class DataGrid(MultipleInstance):
                )
   
   def render(self):
-    if self._df_store.ne_df is None:
+    if self._df is None:
       return Div("No data to display !")
     
     return Div(
@@ -1201,8 +1043,10 @@ class DataGrid(MultipleInstance):
           cls="flex items-center justify-between mb-2"),
       self._panel.set_main(self.mk_table_wrapper()),
       Script(f"initDataGrid('{self._id}');"),
-      # Mouse(self, combinations=self._mouse_support, _id="-mouse"),
-      Keyboard(self, combinations=self._key_support, _id="-keyboard"),
+      Div(
+        Mouse(self, combinations=self._mouse_support, _id="-mouse"),
+        Keyboard(self, combinations=self._key_support, _id="-keyboard"),
+      ),
       id=self._id,
       cls="grid",
       style="height: 100%; grid-template-rows: auto 1fr;"

src/myfasthtml/controls/DataGridColumnsManager.py

@@ -159,24 +159,20 @@ class DataGridColumnsManager(MultipleInstance):
     return self.mk_column_details(col_def)
   
   def _register_formula(self, col_def) -> None:
-    """Register or remove a formula column with the FormulaEngine.
+    """Register or remove a formula column with the FormulaEngine via DataService.
 
     Registers only when col_def.type is Formula and the formula text is
     non-empty.  Removes the formula in all other cases so the engine stays
     consistent with the column definition.
     """
-    engine = self._parent.get_formula_engine()
-    if engine is None:
+    data_service = getattr(self._parent, "_data_service", None)
+    if data_service is None:
       return
-    table = self._parent.get_table_name()
     if col_def.type == ColumnType.Formula and col_def.formula:
-      try:
-        engine.set_formula(table, col_def.col_id, col_def.formula)
-        logger.debug("Registered formula for %s.%s", table, col_def.col_id)
-      except Exception as e:
-        logger.warning("Formula error for %s.%s: %s", table, col_def.col_id, e)
+      data_service.register_formula(col_def.col_id, col_def.formula)
+      logger.debug("Registered formula for col %s", col_def.col_id)
     else:
-      engine.remove_formula(table, col_def.col_id)
+      data_service.remove_formula(col_def.col_id)
   
   def mk_column_label(self, col_def: DataGridColumnState):
     return Div(

src/myfasthtml/controls/DataGridFormattingEditor.py

@@ -2,7 +2,7 @@ import logging
 from collections import defaultdict
 
 from myfasthtml.controls.DslEditor import DslEditor
-from myfasthtml.controls.datagrid_objects import DataGridRowState
+from myfasthtml.controls.datagrid_objects import DataGridRowUiState
 from myfasthtml.core.formatting.dsl import parse_dsl, DSLSyntaxError, ColumnScope, RowScope, CellScope, TableScope, TablesScope
 from myfasthtml.core.instances import InstancesManager
 
@@ -111,7 +111,7 @@ class DataGridFormattingEditor(DslEditor):
     for row_index, rules in rows_rules.items():
       row_state = next((r for r in state.rows if r.row_id == row_index), None)
       if row_state is None:
-        row_state = DataGridRowState(row_id=row_index)
+        row_state = DataGridRowUiState(row_id=row_index)
         state.rows.append(row_state)
       row_state.format = rules
     

src/myfasthtml/controls/DataGridsManager.py

@@ -13,10 +13,8 @@ from myfasthtml.controls.TreeView import TreeView, TreeNode, TreeViewConf
 from myfasthtml.controls.helpers import mk
 from myfasthtml.core.DataGridsRegistry import DataGridsRegistry
 from myfasthtml.core.commands import Command
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
 from myfasthtml.core.dbmanager import DbObject
-from myfasthtml.core.formatting.dsl.completion.provider import DatagridMetadataProvider
-from myfasthtml.core.formatting.presets import DEFAULT_STYLE_PRESETS, DEFAULT_FORMATTER_PRESETS
-from myfasthtml.core.formula.engine import FormulaEngine
 from myfasthtml.core.instances import InstancesManager, SingleInstance
 from myfasthtml.icons.fluent_p1 import table_add20_regular
 from myfasthtml.icons.fluent_p3 import folder_open20_regular
@@ -50,7 +48,7 @@ class Commands(BaseCommands):
     return Command("NewGrid",
                    "New grid",
                    self._owner,
-                   self._owner.new_grid).htmx(target=f"#{self._owner._tree.get_id()}")
+                   self._owner.handle_new_grid).htmx(target=f"#{self._owner._tree.get_id()}")
   
   def open_from_excel(self, tab_id, file_upload):
     return Command("OpenFromExcel",
@@ -81,11 +79,16 @@ class Commands(BaseCommands):
                    key="DeleteNode")
 
 
-class DataGridsManager(SingleInstance, DatagridMetadataProvider):
+class DataGridsManager(SingleInstance):
+  """UI manager for DataGrids.
+
+  Responsible for the visual organisation of DataGrids: TreeView, TabsManager,
+  and document lifecycle (create, open, delete). All data concerns are handled
+  by DataServicesManager and DataService.
+  """
   
   def __init__(self, parent, _id=None):
     if not getattr(self, "_is_new_instance", False):
-      # Skip __init__ if instance already existed
       return
     super().__init__(parent, _id=_id)
     self.commands = Commands(self)
@@ -96,53 +99,47 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
     self._tabs_manager = InstancesManager.get_by_type(self._session, TabsManager, None)
     self._registry = DataGridsRegistry(parent)
     
-    # Global presets shared across all DataGrids
-    self.style_presets: dict = DEFAULT_STYLE_PRESETS.copy()
-    self.formatter_presets: dict = DEFAULT_FORMATTER_PRESETS.copy()
-    self.all_tables_formats: list = []
+    # Data layer — session-scoped singletons
+    self._data_services_manager = DataServicesManager(self._parent)
   
-    # Formula engine shared across all DataGrids in this session
-    self._formula_engine = FormulaEngine(
-      registry_resolver=self._resolve_store_for_table
-    )
-  
-  def upload_from_source(self):
-    file_upload = FileUpload(self)
-    tab_id = self._tabs_manager.create_tab("Upload Datagrid", file_upload)
-    file_upload.set_on_ok(self.commands.open_from_excel(tab_id, file_upload))
-    return self._tabs_manager.show_tab(tab_id)
+  # ------------------------------------------------------------------
+  # Grid lifecycle
+  # ------------------------------------------------------------------
   
   def _create_and_register_grid(self, namespace: str, name: str, df: pd.DataFrame) -> DataGrid:
-    """
-    Create and register a DataGrid.
+    """Create a DataGrid and its companion DataService, then register both.
 
     Args:
-        namespace: Grid namespace
-        name: Grid name
-        df: DataFrame to initialize the grid with
+        namespace: Grid namespace.
+        name: Grid name.
+        df: DataFrame to initialise the grid with.
 
     Returns:
-        Created DataGrid instance
+        Created DataGrid instance.
     """
+    table_name = f"{namespace}.{name}" if namespace else name
+    data_service = self._data_services_manager.create_service(table_name, save_state=True)
+    data_service.load_dataframe(df)
+    
+    grid_id = DataGrid.get_grid_id_from_data_service_id(data_service.get_id())
     dg_conf = DatagridConf(namespace=namespace, name=name)
-    dg = DataGrid(self, conf=dg_conf, save_state=True)
-    dg.init_from_dataframe(df)
+    dg = DataGrid(self, conf=dg_conf, save_state=True, _id=grid_id)
+    
     self._registry.put(namespace, name, dg.get_id())
     return dg
   
   def _create_document(self, namespace: str, name: str, datagrid: DataGrid, tab_id: str = None) -> tuple[
     str, DocumentDefinition]:
-    """
-    Create a DocumentDefinition and its associated tab.
+    """Create a DocumentDefinition and its associated tab.
 
     Args:
-        namespace: Document namespace
-        name: Document name
-        datagrid: Associated DataGrid instance
-        tab_id: Optional existing tab ID. If None, creates a new tab
+        namespace: Document namespace.
+        name: Document name.
+        datagrid: Associated DataGrid instance.
+        tab_id: Optional existing tab ID. If None, creates a new tab.
 
     Returns:
-        Tuple of (tab_id, document)
+        Tuple of (tab_id, document).
     """
     if tab_id is None:
       tab_id = self._tabs_manager.create_tab(name, datagrid)
@@ -159,15 +156,14 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
     return tab_id, document
   
   def _add_document_to_tree(self, document: DocumentDefinition, parent_id: str) -> TreeNode:
-    """
-    Add a document to the tree view.
+    """Add a document node to the tree view.
 
     Args:
-        document: Document to add
-        parent_id: Parent node ID in the tree
+        document: Document to add.
+        parent_id: Parent node ID in the tree.
 
     Returns:
-        Created TreeNode
+        Created TreeNode.
     """
     tree_node = TreeNode(
       id=document.document_id,
@@ -179,8 +175,17 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
     self._tree.add_node(tree_node, parent_id=parent_id)
     return tree_node
   
-  def new_grid(self):
-    # Determine parent folder
+  # ------------------------------------------------------------------
+  # Commands handlers
+  # ------------------------------------------------------------------
+  
+  def upload_from_source(self):
+    file_upload = FileUpload(self)
+    tab_id = self._tabs_manager.create_tab("Upload Datagrid", file_upload)
+    file_upload.set_on_ok(self.commands.open_from_excel(tab_id, file_upload))
+    return self._tabs_manager.show_tab(tab_id)
+  
+  def handle_new_grid(self):
     selected_id = self._tree.get_selected_id()
     if selected_id is None:
       parent_id = self._tree.ensure_path("Untitled")
@@ -188,52 +193,40 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
       node = self._tree._state.items[selected_id]
       if node.type == "folder":
         parent_id = selected_id
-      else:  # leaf
+      else:
         parent_id = node.parent
     
-    # Get namespace and generate unique name
-    namespace = self._tree._state.items[parent_id].label
+    namespace = self._tree.get_state().items[parent_id].label
     name = self._generate_unique_sheet_name(parent_id)
     
-    # Create and register DataGrid
     dg = self._create_and_register_grid(namespace, name, pd.DataFrame())
-    
-    # Create document and tab
     tab_id, document = self._create_document(namespace, name, dg)
-    
-    # Add to tree
     self._add_document_to_tree(document, parent_id)
     
-    # UI-specific handling: open parent, select node, start rename
-    if parent_id not in self._tree._state.opened:
-      self._tree._state.opened.append(parent_id)
-    self._tree._state.selected = document.document_id
+    if parent_id not in self._tree.get_state().opened:
+      self._tree.get_state().opened.append(parent_id)
+    self._tree.get_state().selected = document.document_id
     self._tree._start_rename(document.document_id)
     
     return self._tree, self._tabs_manager.show_tab(tab_id)
   
   def _generate_unique_sheet_name(self, parent_id: str) -> str:
-    children = self._tree._state.items[parent_id].children
-    existing_labels = {self._tree._state.items[c].label for c in children}
+    children = self._tree.get_state().items[parent_id].children
+    existing_labels = {self._tree.get_state().items[c].label for c in children}
     n = 1
     while f"Sheet{n}" in existing_labels:
       n += 1
     return f"Sheet{n}"
   
   def open_from_excel(self, tab_id, file_upload: FileUpload):
-    # Read Excel data
     excel_content = file_upload.get_content()
     df = pd.read_excel(BytesIO(excel_content), file_upload.get_sheet_name())
     namespace = file_upload.get_file_basename()
     name = file_upload.get_sheet_name()
     
-    # Create and register DataGrid
     dg = self._create_and_register_grid(namespace, name, df)
-    
-    # Create document with existing tab
     tab_id, document = self._create_document(namespace, name, dg, tab_id=tab_id)
     
-    # Add to tree
     parent_id = self._tree.ensure_path(document.namespace)
     self._add_document_to_tree(document, parent_id)
     
@@ -243,80 +236,61 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
     document_id = self._tree.get_bag(node_id)
     try:
       document = next(filter(lambda x: x.document_id == document_id, self._state.elements))
-      dg = DataGrid(self, _id=document.datagrid_id)  # reload the state & settings
+      dg = DataGrid(self, _id=document.datagrid_id)
       return self._tabs_manager.show_or_create_tab(document.tab_id, document.name, dg)
     except StopIteration:
-      # the selected node is not a document (it's a folder)
       return None
   
   def delete_grid(self, node_id):
-    """
-    Delete a grid and all its associated resources.
+    """Delete a grid and all its associated resources.
 
-    This method is called BEFORE TreeView._delete_node() to ensure we can
-    access the node's bag to retrieve the document_id.
+    Called BEFORE TreeView._delete_node() so we can access the node bag.
 
     Args:
-        node_id: ID of the TreeView node to delete
+        node_id: ID of the TreeView node to delete.
 
     Returns:
-        None (TreeView will handle the node removal)
+        List of UI updates, or None.
     """
     document_id = self._tree.get_bag(node_id)
     if document_id is None:
-      # Node is a folder, not a document - nothing to clean up
       return None
     
     res = []
     try:
-      # Find the document
       document = next(filter(lambda x: x.document_id == document_id, self._state.elements))
-      
-      # Get the DataGrid instance
       dg = DataGrid(self, _id=document.datagrid_id)
       
-      # Close the tab
       close_tab_res = self._tabs_manager.close_tab(document.tab_id)
       res.append(close_tab_res)
       
-      # Remove from registry
       self._registry.remove(document.datagrid_id)
+      self._data_services_manager.remove_service(document.datagrid_id)
       
-      # Clean up DataGrid (delete DBEngine entries)
       dg.delete()
-      
-      # Remove from InstancesManager
       InstancesManager.remove(self._session, document.datagrid_id)
       
-      # Remove DocumentDefinition from state
       self._state.elements = [d for d in self._state.elements if d.document_id != document_id]
       self._state.save()
     
     except StopIteration:
-      # Document not found - already deleted or invalid state
       pass
     
     return res
   
   def create_tab_content(self, tab_id):
-    """
-    Recreate the content for a tab managed by this DataGridsManager.
-    Called by TabsManager when the content is not in cache (e.g., after restart).
+    """Recreate tab content after restart.
 
     Args:
-        tab_id: ID of the tab to recreate content for
+        tab_id: ID of the tab to recreate.
 
     Returns:
-        The recreated component (Panel with DataGrid)
+        DataGrid instance for the tab.
     """
-    # Find the document associated with this tab
     document = next((d for d in self._state.elements if d.tab_id == tab_id), None)
-    
     if document is None:
       raise ValueError(f"No document found for tab {tab_id}")
-    
-    # Recreate the DataGrid with its saved state
-    dg = DataGrid(self, _id=document.datagrid_id)  # reload the state & settings
+    dg = DataGrid(self, _id=document.datagrid_id)
     return dg
   
   def clear_tree(self):
@@ -324,106 +298,22 @@ class DataGridsManager(SingleInstance, DatagridMetadataProvider):
     self._tree.clear()
     return self._tree
   
-  # === DatagridMetadataProvider ===
-  
-  def list_tables(self):
-    return self._registry.get_all_tables()
-  
-  def list_columns(self, table_name):
-    return self._registry.get_columns(table_name)
-  
-  def list_column_values(self, table_name, column_name):
-    return self._registry.get_column_values(table_name, column_name)
-  
-  def get_row_count(self, table_name):
-    return self._registry.get_row_count(table_name)
-  
-  def get_column_type(self, table_name, column_name):
-    return self._registry.get_column_type(table_name, column_name)
-  
-  def list_style_presets(self) -> list[str]:
-    return list(self.style_presets.keys())
-  
-  def list_format_presets(self) -> list[str]:
-    return list(self.formatter_presets.keys())
-  
-  def _resolve_store_for_table(self, table_name: str):
-    """
-    Resolve the DatagridStore for a given table name.
-
-    Used by FormulaEngine as the registry_resolver callback.
-
-    Args:
-        table_name: Full table name in ``"namespace.name"`` format.
-
-    Returns:
-        DatagridStore instance or None if not found.
-    """
-    try:
-      as_fullname_dict = self._registry._get_entries_as_full_name_dict()
-      grid_id = as_fullname_dict.get(table_name)
-      if grid_id is None:
-        return None
-      datagrid = InstancesManager.get(self._session, grid_id, None)
-      if datagrid is None:
-        return None
-      return datagrid._df_store
-    except Exception:
-      return None
-  
-  def get_style_presets(self) -> dict:
-    """Get the global style presets."""
-    return self.style_presets
-  
-  def get_formatter_presets(self) -> dict:
-    """Get the global formatter presets."""
-    return self.formatter_presets
-  
-  def get_formula_engine(self) -> FormulaEngine:
-    """The FormulaEngine shared across all DataGrids in this session."""
-    return self._formula_engine
-  
-  def add_style_preset(self, name: str, preset: dict):
-    """
-    Add or update a style preset.
-
-    Args:
-        name: Preset name (e.g., "custom_highlight")
-        preset: Dict with CSS properties (e.g., {"background-color": "yellow", "color": "black"})
-    """
-    self.style_presets[name] = preset
-  
-  def add_formatter_preset(self, name: str, preset: dict):
-    """
-    Add or update a formatter preset.
-
-    Args:
-        name: Preset name (e.g., "custom_currency")
-        preset: Dict with formatter config (e.g., {"type": "number", "prefix": "CHF ", "precision": 2})
-    """
-    self.formatter_presets[name] = preset
-  
-  def remove_style_preset(self, name: str):
-    """Remove a style preset."""
-    if name in self.style_presets:
-      del self.style_presets[name]
-  
-  def remove_formatter_preset(self, name: str):
-    """Remove a formatter preset."""
-    if name in self.formatter_presets:
-      del self.formatter_presets[name]
-  
-  # === UI ===
+  # ------------------------------------------------------------------
+  # UI
+  # ------------------------------------------------------------------
   
   def mk_main_icons(self):
     return Div(
-      mk.icon(folder_open20_regular, tooltip="Upload from source", command=self.commands.upload_from_source()),
-      mk.icon(table_add20_regular, tooltip="New grid", command=self.commands.new_grid()),
+      mk.icon(folder_open20_regular, tooltip="Upload from source",
+              command=self.commands.upload_from_source()),
+      mk.icon(table_add20_regular, tooltip="New grid",
+              command=self.commands.new_grid()),
       cls="flex"
     )
   
   def _mk_tree(self):
-    conf = TreeViewConf(add_leaf=False, icons={"folder": "database20_regular", "excel": "table20_regular"})
+    conf = TreeViewConf(add_leaf=False,
+                        icons={"folder": "database20_regular", "excel": "table20_regular"})
     tree = TreeView(self, conf=conf, _id="-treeview")
     for element in self._state.elements:
       parent_id = tree.ensure_path(element.namespace, node_type="folder")

src/myfasthtml/controls/TreeView.py

@@ -294,6 +294,9 @@ class TreeView(MultipleInstance):
     except KeyError:
       return None
     
+  def get_state(self) -> TreeViewState:
+    return self._state
+  
   def _toggle_node(self, node_id: str):
     """Toggle expand/collapse state of a node."""
     if node_id in self._state.opened:

src/myfasthtml/controls/datagrid_objects.py

@@ -1,10 +1,11 @@
 from dataclasses import dataclass, field
 
-from myfasthtml.core.constants import ColumnType, DATAGRID_DEFAULT_COLUMN_WIDTH, ViewType
+from myfasthtml.core.constants import DATAGRID_DEFAULT_COLUMN_WIDTH, ViewType, ROW_SELECTION_ID, ColumnType
+from myfasthtml.core.data.ColumnDefinition import ColumnDefinition
 
 
 @dataclass
-class DataGridRowState:
+class DataGridRowUiState:
   row_id: int
   visible: bool = True
   height: int | None = None
@@ -12,19 +13,32 @@ class DataGridRowState:
 
 
 @dataclass
-class DataGridColumnState:
-  col_id: str  # name of the column: cannot be changed
-  col_index: int  # index of the column in the dataframe: cannot be changed
-  title: str = None
-  type: ColumnType = ColumnType.Text
+class DataGridColumnUiState:
+  """UI presentation state of a DataGrid column.
+
+  Holds only the visual properties of a column. Keyed by col_id to match
+  the corresponding ColumnDefinition held by DataService.
+
+  Attributes:
+      col_id: Column identifier. Must match the col_id in ColumnDefinition.
+      width: Column width in pixels.
+      visible: Whether the column is displayed.
+      format: List of format rules applied during rendering.
+  """
+  
+  col_id: str
   visible: bool = True
   width: int = DATAGRID_DEFAULT_COLUMN_WIDTH
-  format: list = field(default_factory=list)  #
-  formula: str = ""  # formula expression for ColumnType.Formula columns
+  format: list = field(default_factory=list)
   
-  def copy(self):
-    props = {k: v for k, v in self.__dict__.items() if not k.startswith("_")}
-    return DataGridColumnState(**props)
+  def copy(self) -> "DataGridColumnUiState":
+    """Return a shallow copy of this state."""
+    return DataGridColumnUiState(
+      col_id=self.col_id,
+      width=self.width,
+      visible=self.visible,
+      format=list(self.format),
+    )
 
 
 @dataclass
@@ -56,4 +70,71 @@ class DataGridHeaderFooterConf:
 class DatagridView:
   name: str
   type: ViewType = ViewType.Table
-  columns: list[DataGridColumnState] = None
+  columns: list[DataGridColumnUiState] = None
+
+
+class DataGridColumnState:
+  def __init__(self, col_def: ColumnDefinition, col_ui_state: DataGridColumnUiState):
+    self._col_def = col_def
+    self._col_ui_state = col_ui_state
+  
+  @property
+  def col_id(self):
+    return self._col_def.col_id
+  
+  @property
+  def col_index(self):
+    return self._col_def.col_index
+  
+  @property
+  def title(self):
+    return self._col_def.title
+  
+  @property
+  def type(self):
+    return self._col_def.type
+  
+  @property
+  def visible(self):
+    return self._col_ui_state.visible
+  
+  @property
+  def width(self):
+    return self._col_ui_state.width
+  
+  @property
+  def format(self):
+    return self._col_ui_state.format
+
+
+class DataGridRowSelectionColumnState(DataGridColumnState):
+  def __init__(self):
+    super().__init__(None, None)
+  
+  @property
+  def col_id(self):
+    return ROW_SELECTION_ID
+  
+  @property
+  def col_index(self):
+    return -1
+  
+  @property
+  def title(self):
+    return ""
+  
+  @property
+  def type(self):
+    return ColumnType.RowSelection_
+  
+  @property
+  def visible(self):
+    return True  # not used
+  
+  @property
+  def width(self):
+    return 24  # not used
+  
+  @property
+  def format(self):
+    return ""

src/myfasthtml/core/DataGridsRegistry.py

@@ -37,74 +37,9 @@ class DataGridsRegistry(SingleInstance):
       del all_entries[datagrid_id]
       self._db_manager.save(DATAGRIDS_REGISTRY_ENTRY_KEY, all_entries)
   
-  def get_all_tables(self):
-    all_entries = self._get_all_entries()
-    return [f"{namespace}.{name}" for (namespace, name) in all_entries.values()]
-  
-  def get_columns(self, table_name):
-    try:
-      as_fullname_dict = self._get_entries_as_full_name_dict()
-      grid_id = as_fullname_dict[table_name]
-      
-      # load datagrid state
-      state_id = f"{grid_id}#state"
-      state = self._db_manager.load(state_id)
-      return [c.col_id for c in state["columns"]] if state else []
-    except KeyError:
-      return []
-  
-  def get_column_values(self, table_name, column_name):
-    try:
-      as_fullname_dict = self._get_entries_as_full_name_dict()
-      grid_id = as_fullname_dict[table_name]
-      
-      # load dataframe from dedicated store
-      store = self._db_manager.load(f"{grid_id}#df")
-      df = store["ne_df"] if store else None
-      return df[column_name].tolist() if df is not None else []
-    
-    except KeyError:
-      return []
-  
-  def get_row_count(self, table_name):
-    try:
-      as_fullname_dict = self._get_entries_as_full_name_dict()
-      grid_id = as_fullname_dict[table_name]
-      
-      # load dataframe from dedicated store
-      store = self._db_manager.load(f"{grid_id}#df")
-      df = store["ne_df"] if store else None
-      return len(df) if df is not None else 0
-    
-    except KeyError:
-      return 0
-  
-  def get_column_type(self, table_name, column_name):
-    """
-    Get the type of a column.
-
-    Args:
-        table_name: The DataGrid name
-        column_name: The column name
-
-    Returns:
-        ColumnType enum value or None if not found
-    """
-    try:
-      as_fullname_dict = self._get_entries_as_full_name_dict()
-      grid_id = as_fullname_dict[table_name]
-      
-      # load datagrid state
-      state_id = f"{grid_id}#state"
-      state = self._db_manager.load(state_id)
-      
-      if state and "columns" in state:
-        for col in state["columns"]:
-          if col.col_id == column_name:
-            return col.type
-      return None
-    except KeyError:
-      return None
+  def get_all_entries(self) -> dict:
+    """Return all registry entries as {datagrid_id: (namespace, name)}."""
+    return self._get_all_entries()
 
   def _get_all_entries(self):
     return {k: v for k, v in self._db_manager.load(DATAGRIDS_REGISTRY_ENTRY_KEY).items()

src/myfasthtml/core/data/ColumnDefinition.py

@@ -0,0 +1,37 @@
+from dataclasses import dataclass, field
+
+from myfasthtml.core.constants import ColumnType
+
+
+@dataclass
+class ColumnDefinition:
+    """Data semantics of a DataGrid column.
+
+    Holds the structural and computational properties of a column.
+    Does not contain any UI-related attributes (width, visibility, formatting).
+    Those are stored in ColumnUiState within DatagridState.
+
+    Attributes:
+        col_id: Unique identifier for the column. Cannot be changed after creation.
+        col_index: Index of the column in the DataFrame. -1 for virtual columns
+            (Formula, RowIndex).
+        title: Display title of the column.
+        type: Column data type, determines rendering and mutation behaviour.
+        formula: DSL expression for ColumnType.Formula columns. Empty string otherwise.
+    """
+
+    col_id: str
+    col_index: int
+    title: str = None
+    type: ColumnType = ColumnType.Text
+    formula: str = ""
+
+    def copy(self) -> "ColumnDefinition":
+        """Return a shallow copy of this definition."""
+        return ColumnDefinition(
+            col_id=self.col_id,
+            col_index=self.col_index,
+            title=self.title,
+            type=self.type,
+            formula=self.formula,
+        )

src/myfasthtml/core/data/DataService.py

@@ -0,0 +1,386 @@
+import logging
+from typing import Optional
+
+import numpy as np
+import pandas as pd
+
+from myfasthtml.core.constants import ColumnType, ROW_INDEX_ID
+from myfasthtml.core.data.ColumnDefinition import ColumnDefinition
+from myfasthtml.core.dbmanager import DbObject
+from myfasthtml.core.instances import MultipleInstance
+from myfasthtml.core.utils import make_safe_id, make_unique_safe_id
+
+logger = logging.getLogger(__name__)
+
+_COLUMN_TYPE_DEFAULTS = {
+    ColumnType.Number: 0,
+    ColumnType.Text: "",
+    ColumnType.Bool: False,
+    ColumnType.Datetime: pd.NaT,
+}
+
+
+class DataStore(DbObject):
+  """Persistent storage for a DataGrid's tabular data.
+
+  Holds the DataFrame and its derived caches used for rendering and formula
+  evaluation. Contains no business logic — all mutations are performed by
+  DataService.
+
+  Attributes:
+      ne_df: The pandas DataFrame. Source of truth for non-formula columns.
+      ns_fast_access: Dict mapping col_id to a numpy array. O(1) column
+          lookup used by FormulaEngine and rendering.
+      ns_row_data: List of row dicts built from ns_fast_access. Used by
+          FormattingEngine for rule evaluation.
+      ns_total_rows: Cached total row count after filtering.
+  """
+  
+  def __init__(self, owner, save_state: bool = True):
+    with self.initializing():
+      super().__init__(owner, name=f"{owner.get_id()}#store", save_state=save_state)
+      self.ne_df = None
+      self.ns_fast_access = None
+      self.ns_row_data = None
+      self.ns_total_rows = None
+
+
+class DataServiceState(DbObject):
+  """Persistent state for DataService.
+
+  Stores the column definitions and the table name associated with the
+  DataGrid. Persists across sessions via DbObject.
+
+  Attributes:
+      columns: Ordered list of column data definitions.
+      table_name: Fully qualified table name used by FormulaEngine
+          (format: "namespace.name" or "name").
+  """
+  
+  def __init__(self, owner, save_state: bool = True):
+    with self.initializing():
+      super().__init__(owner, name="#state", save_state=save_state)
+      self.columns: list[ColumnDefinition] = []
+      self.table_name: str = ""
+
+
+class DataService(MultipleInstance):
+  """Data companion to DataGrid.
+
+  Owns the DataStore and the list of ColumnDefinition objects for one
+  DataGrid. All data mutations go through this class. Holds a reference to
+  DataServicesManager to access the shared FormulaEngine.
+
+  This class can exist and operate independently of any rendering component.
+
+  Attributes:
+      _state: Persistent state (columns, table_name).
+      _store: Persistent storage (DataFrame, caches).
+  """
+  
+  def __init__(self, parent, _id: Optional[str] = None, save_state: bool = True):
+    super().__init__(parent, _id=_id)
+    self._state = DataServiceState(self, save_state=save_state)
+    self._store = DataStore(self, save_state=save_state)
+    self._init_store()
+  
+  @property
+  def columns(self) -> list[ColumnDefinition]:
+    """Return the list of column definitions."""
+    return self._state.columns
+  
+  @property
+  def table_name(self) -> str:
+    """Return the fully qualified table name used by FormulaEngine."""
+    return self._state.table_name
+  
+  def set_table_name(self, table_name: str) -> None:
+    """Update the table name (e.g. after a rename)."""
+    self._state.table_name = table_name
+  
+  def get_store(self) -> DataStore:
+    """Return the underlying DataStore."""
+    return self._store
+  
+  def get_formula_engine(self):
+    """Return the shared FormulaEngine from DataServicesManager."""
+    return self._parent.get_formula_engine()
+  
+  # ------------------------------------------------------------------
+  # Data initialisation
+  # ------------------------------------------------------------------
+  
+  def load_dataframe(self, df: pd.DataFrame, init_columns: bool = True) -> None:
+    """Load a DataFrame into the store and initialise caches.
+
+    Args:
+        df: Source DataFrame. Column names are normalised to safe IDs.
+        init_columns: When True, build ColumnDefinition list from the
+            DataFrame columns and register any existing formula columns
+            with the FormulaEngine.
+    """
+    if df is None:
+      return
+    
+    df.columns = df.columns.map(make_safe_id)
+    self._store.ne_df = df
+    
+    if init_columns:
+      self._state.columns = self._build_column_definitions(df)
+      self._state.save()
+    
+    self._init_store()
+  
+  # ------------------------------------------------------------------
+  # Mutations
+  # ------------------------------------------------------------------
+  
+  def add_column(self, col_def: ColumnDefinition) -> None:
+    """Add a new column to the DataGrid data layer.
+
+    Assigns a unique safe col_id from the title. For Formula and RowIndex
+    columns, no DataFrame column is created. For all other types, a column
+    with a type-appropriate default value is added to the DataFrame.
+
+    Args:
+        col_def: Column definition. col_id will be set by this method.
+    """
+    col_def.col_id = make_unique_safe_id(
+      col_def.title, [c.col_id for c in self._state.columns]
+    )
+    
+    if col_def.type == ColumnType.Formula:
+      col_def.col_index = -1
+      self._state.columns.append(col_def)
+      self._state.save()
+      return
+    
+    if col_def.type == ColumnType.RowIndex:
+      col_def.col_index = -1
+      self._state.columns.append(col_def)
+      if self._store.ne_df is not None:
+        self._store.ns_fast_access[col_def.col_id] = (
+            self._store.ne_df.index.to_numpy()
+        )
+      self._state.save()
+      self._store.save()
+      return
+    
+    default_value = _COLUMN_TYPE_DEFAULTS.get(col_def.type, "")
+    col_def.col_index = (
+        len(self._store.ne_df.columns) if self._store.ne_df is not None else 0
+    )
+    self._state.columns.append(col_def)
+    
+    if self._store.ne_df is not None:
+      self._store.ne_df[col_def.col_id] = default_value
+      self._store.ns_fast_access[col_def.col_id] = (
+          self._store.ne_df[col_def.col_id].to_numpy()
+      )
+      for row_dict in self._store.ns_row_data:
+        row_dict[col_def.col_id] = default_value
+    
+    self._state.save()
+    self._store.save()
+    self._mark_changed(col_def.col_id)
+  
+  def add_row(self, row_data: Optional[dict] = None) -> None:
+    """Append a new row with incremental cache updates.
+
+    Creates default values for all non-virtual columns when row_data is
+    not provided. Marks formula columns dirty so ensure_ready() will
+    recalculate them on the next render.
+
+    Args:
+        row_data: Optional dict of {col_id: value}. Defaults to
+            type-appropriate values for each column.
+    """
+    if self._store.ne_df is None:
+      return
+    
+    new_index = len(self._store.ne_df)
+    
+    if row_data is None:
+      row_data = {}
+      for col in self._state.columns:
+        if col.type not in (ColumnType.Formula, ColumnType.RowSelection_):
+          value = (
+              new_index
+              if col.type == ColumnType.RowIndex
+              else _COLUMN_TYPE_DEFAULTS.get(col.type, "")
+          )
+          row_data[col.col_id] = value
+    
+    self._store.ne_df.loc[new_index] = row_data
+    
+    for col_id, value in row_data.items():
+      if col_id in self._store.ns_fast_access:
+        self._store.ns_fast_access[col_id] = np.append(
+          self._store.ns_fast_access[col_id], value
+        )
+      else:
+        self._store.ns_fast_access[col_id] = np.array([value])
+    
+    self._store.ns_row_data.append(row_data.copy())
+    self._store.ns_total_rows = len(self._store.ne_df)
+    self._store.save()
+    
+    self._mark_all_formula_columns_dirty()
+  
+  def set_data(self, col_id: str, row_index: int, value) -> None:
+    """Update a single cell value.
+
+    Updates the DataFrame, fast-access cache, and row data dict, then
+    marks dependent formula columns dirty.
+
+    Args:
+        col_id: Column identifier.
+        row_index: Zero-based row index.
+        value: New cell value.
+    """
+    if self._store.ne_df is None:
+      return
+    
+    self._store.ne_df.at[row_index, col_id] = value
+    
+    if self._store.ns_fast_access and col_id in self._store.ns_fast_access:
+      self._store.ns_fast_access[col_id][row_index] = value
+    
+    if self._store.ns_row_data and row_index < len(self._store.ns_row_data):
+      self._store.ns_row_data[row_index][col_id] = value
+    
+    self._store.save()
+    self._mark_changed(col_id, rows=[row_index])
+  
+  # ------------------------------------------------------------------
+  # Formula management
+  # ------------------------------------------------------------------
+  
+  def register_formula(self, col_id: str, formula_text: str) -> None:
+    """Register or update a formula for a column with the FormulaEngine.
+
+    Args:
+        col_id: Column identifier.
+        formula_text: DSL formula expression.
+    """
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    try:
+      engine.set_formula(self._state.table_name, col_id, formula_text)
+    except Exception as e:
+      logger.warning("Failed to register formula for %s.%s: %s",
+                     self._state.table_name, col_id, e)
+  
+  def remove_formula(self, col_id: str) -> None:
+    """Remove a formula for a column from the FormulaEngine.
+
+    Args:
+        col_id: Column identifier.
+    """
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    engine.remove_formula(self._state.table_name, col_id)
+  
+  def ensure_ready(self) -> None:
+    """Recalculate dirty formula columns before rendering.
+
+    Called by DataGrid.mk_body_content_page() to ensure formula columns
+    are up-to-date. No-op when no columns are dirty.
+    """
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    engine.recalculate_if_needed(self._state.table_name, self._store)
+  
+  # ------------------------------------------------------------------
+  # Private helpers
+  # ------------------------------------------------------------------
+  
+  def _init_store(self):
+    df = self._store.ne_df
+    if df is None:
+      return
+    self._store.ns_fast_access = self._build_fast_access(df)
+    self._store.ns_row_data = df.to_dict(orient="records")
+    self._store.ns_total_rows = len(df)
+    self._store.save()
+    self._register_existing_formulas()
+  
+  def _build_column_definitions(self, df: pd.DataFrame) -> list[ColumnDefinition]:
+    """Build ColumnDefinition objects from DataFrame columns.
+
+    Args:
+        df: Source DataFrame with normalised column names.
+
+    Returns:
+        Ordered list of ColumnDefinition objects.
+    """
+    return [
+        ColumnDefinition(
+          col_id=make_safe_id(col_id),
+          col_index=col_index,
+          title=col_id,
+          type=self._infer_column_type(df[make_safe_id(col_id)].dtype),
+        )
+        for col_index, col_id in enumerate(df.columns)
+    ]
+  
+  @staticmethod
+  def _infer_column_type(dtype) -> ColumnType:
+    """Infer ColumnType from a pandas dtype."""
+    if pd.api.types.is_integer_dtype(dtype):
+      return ColumnType.Number
+    if pd.api.types.is_float_dtype(dtype):
+      return ColumnType.Number
+    if pd.api.types.is_bool_dtype(dtype):
+      return ColumnType.Bool
+    if pd.api.types.is_datetime64_any_dtype(dtype):
+      return ColumnType.Datetime
+    return ColumnType.Text
+  
+  @staticmethod
+  def _build_fast_access(df: pd.DataFrame) -> dict:
+    """Build ns_fast_access from a DataFrame.
+
+    Args:
+        df: Source DataFrame.
+
+    Returns:
+        Dict mapping col_id to numpy array, plus ROW_INDEX_ID.
+    """
+    result = {col: df[col].to_numpy() for col in df.columns}
+    result[ROW_INDEX_ID] = df.index.to_numpy()
+    return result
+  
+  def _register_existing_formulas(self) -> None:
+    """Re-register all formula columns with the FormulaEngine."""
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    for col_def in self._state.columns:
+      if col_def.formula:
+        self.register_formula(col_def.col_id, col_def.formula)
+  
+  def _mark_changed(self, col_id: str, rows: Optional[list[int]] = None) -> None:
+    """Notify FormulaEngine that a column's data has changed.
+
+    Args:
+        col_id: Changed column identifier.
+        rows: Optional list of changed row indices. None means all rows.
+    """
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    engine.mark_data_changed(self._state.table_name, col_id, rows)
+  
+  def _mark_all_formula_columns_dirty(self) -> None:
+    """Mark all formula columns dirty after a structural change (e.g. add_row)."""
+    engine = self.get_formula_engine()
+    if engine is None:
+      return
+    table = self._state.table_name
+    for col in self._state.columns:
+      if col.type == ColumnType.Formula and col.formula:
+        engine.mark_data_changed(table, col.col_id)

src/myfasthtml/core/data/DataServicesManager.py

@@ -0,0 +1,121 @@
+import logging
+from typing import Optional
+
+from myfasthtml.core.data.DataService import DataService
+from myfasthtml.core.formula.engine import FormulaEngine
+from myfasthtml.core.instances import SingleInstance
+
+logger = logging.getLogger(__name__)
+
+
+class DataServicesManager(SingleInstance):
+  """Session-scoped manager for all DataService instances.
+
+  Owns the shared FormulaEngine and acts as the single entry point for
+  creating and retrieving DataService instances. Provides the resolver
+  callback that allows the FormulaEngine to access any table's DataStore
+  by table name.
+
+  Access pattern (from any component):
+      manager = InstancesManager.get_by_type(session, DataServicesManager)
+      service = manager.get_service(grid_id)
+  """
+  
+  def __init__(self, parent=None, _id: Optional[str] = None):
+    if not getattr(self, "_is_new_instance", False):
+      return
+    super().__init__(parent, _id)
+    self._services: dict[str, DataService] = {}
+    self._formula_engine = FormulaEngine(registry_resolver=self._resolve_store_for_table)
+  
+  # ------------------------------------------------------------------
+  # Service lifecycle
+  # ------------------------------------------------------------------
+  
+  def create_service(self, table_name: str, _id=None, save_state: bool = True) -> DataService:
+    """Create and register a new DataService for a DataGrid.
+
+    Called by DataGridsManager when a new grid is created.
+
+    Args:
+        table_name: Fully qualified table name ("namespace.name" or "name").
+        save_state: Whether to persist the DataService state to DB.
+        _id: Unique identifier of the DataGrid.
+
+    Returns:
+        The newly created DataService instance.
+    """
+    service = DataService(self, _id=_id, save_state=save_state)
+    service.set_table_name(table_name)
+    self._services[service.get_id()] = service
+    logger.debug(f"DataService created for '{table_name}' (grid_id={service.get_id()})")
+    return service
+  
+  def get_service(self, grid_id: str) -> Optional[DataService]:
+    """Return the DataService for a given grid_id.
+
+    Args:
+        grid_id: Unique identifier of the DataGrid.
+
+    Returns:
+        DataService instance, or None if not found.
+    """
+    return self._services.get(grid_id)
+  
+  def restore_service(self, grid_id: str) -> Optional[DataService]:
+    """Restore a DataService from persisted state on session restart.
+
+    Called by DataGrid on restart to re-attach its DataService.
+    The DataService state (columns, table_name) and DataStore (DataFrame)
+    are loaded from DB automatically via DbObject.
+
+    Args:
+        grid_id: Unique identifier of the DataGrid.
+
+    Returns:
+        The restored DataService instance.
+    """
+    if grid_id in self._services:
+      return self._services[grid_id]
+    
+    service = DataService(self, _id=grid_id)
+    self._services[grid_id] = service
+    logger.debug("DataService restored for grid_id=%s", grid_id)
+    return service
+  
+  def remove_service(self, grid_id: str) -> None:
+    """Unregister and discard a DataService.
+
+    Called by DataGridsManager when a grid is deleted.
+
+    Args:
+        grid_id: Unique identifier of the DataGrid.
+    """
+    self._services.pop(grid_id, None)
+    logger.debug("DataService removed for grid_id=%s", grid_id)
+  
+  # ------------------------------------------------------------------
+  # FormulaEngine
+  # ------------------------------------------------------------------
+  
+  def get_formula_engine(self) -> FormulaEngine:
+    """Return the shared FormulaEngine for this session."""
+    return self._formula_engine
+  
+  def _resolve_store_for_table(self, table_name: str):
+    """Resolve the DataStore for a given table name.
+
+    Used by FormulaEngine as the registry_resolver callback for
+    cross-table formula evaluation.
+
+    Args:
+        table_name: Fully qualified table name ("namespace.name").
+
+    Returns:
+        DataStore instance, or None if the table is not found.
+    """
+    for service in self._services.values():
+      if service.table_name == table_name:
+        return service.get_store()
+    logger.warning(f"DataServicesManager: table '{table_name}' not found")
+    return None

src/myfasthtml/core/formatting/dsl/completion/provider.py

@@ -1,89 +1,203 @@
 """
 Metadata provider for DataGrid formatting DSL autocompletion.
 
-Provides access to DataGrid metadata (columns, values, row counts)
-for context-aware autocompletion.
+Provides access to DataGrid metadata (columns, values, row counts, presets)
+for context-aware autocompletion. Delegates live data queries to
+DataServicesManager and holds global formatting presets.
 """
 
-from typing import Any
+import logging
+from typing import Any, Optional
 
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
 from myfasthtml.core.dsl.base_provider import BaseMetadataProvider
+from myfasthtml.core.formatting.presets import DEFAULT_FORMATTER_PRESETS, DEFAULT_STYLE_PRESETS
+from myfasthtml.core.instances import SingleInstance, InstancesManager
+
+logger = logging.getLogger(__name__)
 
 
-class DatagridMetadataProvider(BaseMetadataProvider):
-  """
-  Protocol for providing DataGrid metadata to the autocompletion engine.
+class DatagridMetadataProvider(SingleInstance, BaseMetadataProvider):
+    """Concrete session-scoped metadata provider for DataGrid DSL engines.
 
-  Implementations must provide access to:
-  - Available DataGrids (tables)
-  - Column names for each DataGrid
-  - Distinct values for each column
-  - Row count for each DataGrid
-  - Style and format presets
+    Implements BaseMetadataProvider by delegating live data queries to
+    DataServicesManager. Also holds the global formatting presets and the
+    all_tables_formats rule applied to every table.
 
-  DataGrid names follow the pattern namespace.name (multi-level namespaces).
-  """
+    Access pattern (from any component):
+        provider = InstancesManager.get_by_type(session, DatagridMetadataProvider)
 
-  def list_tables(self) -> list[str]:
+    Attributes:
+        style_presets: Dict of named style presets available in the DSL.
+        formatter_presets: Dict of named formatter presets available in the DSL.
+        all_tables_formats: Global format rules applied to all tables.
     """
-    Return the list of available DataGrid names.
 
-    Returns:
-        List of DataGrid names (e.g., ["app.orders", "app.customers"])
-    """
-    ...
+    def __init__(self, parent=None, session: Optional[dict] = None,
+                 _id: Optional[str] = None):
+        super().__init__(parent, session, _id)
+        with self.initializing():
+            self.style_presets: dict = DEFAULT_STYLE_PRESETS.copy()
+            self.formatter_presets: dict = DEFAULT_FORMATTER_PRESETS.copy()
+            self.all_tables_formats: list = []
 
-  def list_columns(self, table_name: str) -> list[str]:
-    """
-    Return the column names for a specific DataGrid.
+    # ------------------------------------------------------------------
+    # Table and column metadata — delegated to DataServicesManager
+    # ------------------------------------------------------------------
 
-    Args:
-        table_name: The DataGrid name
+    def list_tables(self) -> list[str]:
+        """Return the list of all registered table names.
 
-    Returns:
-        List of column names (e.g., ["id", "amount", "status"])
-    """
-    ...
+        Returns:
+            List of table names in "namespace.name" format.
+        """
+        manager = self._get_data_services_manager()
+        if manager is None:
+            return []
+        return [s.table_name for s in manager._services.values() if s.table_name]
 
-  def list_column_values(self, table_name, column_name: str) -> list[Any]:
-    """
-    Return the distinct values for a column in the current DataGrid.
+    def list_columns(self, table_name: str) -> list[str]:
+        """Return the column identifiers for a table.
 
-    This is used to suggest values in conditions like `value == |`.
+        Args:
+            table_name: Fully qualified table name.
 
-    Args:
-        column_name: The column name
+        Returns:
+            List of col_id strings.
+        """
+        service = self._get_service(table_name)
+        if service is None:
+            return []
+        return [c.col_id for c in service.columns]
 
-    Returns:
-        List of distinct values in the column
-    """
-    ...
+    def list_column_values(self, table_name: str, column_name: str) -> list[Any]:
+        """Return the distinct values present in a column.
 
-  def get_row_count(self, table_name: str) -> int:
-    """
-    Return the number of rows in a DataGrid.
+        Args:
+            table_name: Fully qualified table name.
+            column_name: Column identifier.
 
-    Used to suggest row indices for row scope and cell scope.
+        Returns:
+            List of distinct values, empty list if not found.
+        """
+        service = self._get_service(table_name)
+        if service is None:
+            return []
+        store = service.get_store()
+        if store.ne_df is None or column_name not in store.ne_df.columns:
+            return []
+        return store.ne_df[column_name].dropna().unique().tolist()
 
-    Args:
-        table_name: The DataGrid name
+    def get_row_count(self, table_name: str) -> int:
+        """Return the number of rows in a table.
 
-    Returns:
-        Number of rows
-    """
-    ...
+        Args:
+            table_name: Fully qualified table name.
 
-  def get_column_type(self, table_name: str, column_name: str):
-    """
-    Return the type of a column.
+        Returns:
+            Row count, or 0 if not found.
+        """
+        service = self._get_service(table_name)
+        if service is None:
+            return 0
+        store = service.get_store()
+        return store.ns_total_rows or 0
 
-    Used to filter suggestions based on column type.
+    def get_column_type(self, table_name: str, column_name: str):
+        """Return the ColumnType for a column.
 
-    Args:
-        table_name: The DataGrid name
-        column_name: The column name
+        Args:
+            table_name: Fully qualified table name.
+            column_name: Column identifier.
 
-    Returns:
-        ColumnType enum value or None if not found
-    """
-    ...
+        Returns:
+            ColumnType enum value, or None if not found.
+        """
+        service = self._get_service(table_name)
+        if service is None:
+            return None
+        for col in service.columns:
+            if col.col_id == column_name:
+                return col.type
+        return None
+
+    # ------------------------------------------------------------------
+    # Preset metadata — held locally
+    # ------------------------------------------------------------------
+
+    def list_style_presets(self) -> list[str]:
+        """Return the names of all registered style presets."""
+        return list(self.style_presets.keys())
+
+    def list_format_presets(self) -> list[str]:
+        """Return the names of all registered formatter presets."""
+        return list(self.formatter_presets.keys())
+
+    def get_style_presets(self) -> dict:
+        """Return the full style presets dict."""
+        return self.style_presets
+
+    def get_formatter_presets(self) -> dict:
+        """Return the full formatter presets dict."""
+        return self.formatter_presets
+
+    def add_style_preset(self, name: str, preset: dict) -> None:
+        """Add or update a named style preset.
+
+        Args:
+            name: Preset name.
+            preset: Style definition dict.
+        """
+        self.style_presets[name] = preset
+
+    def add_formatter_preset(self, name: str, preset: dict) -> None:
+        """Add or update a named formatter preset.
+
+        Args:
+            name: Preset name.
+            preset: Formatter definition dict.
+        """
+        self.formatter_presets[name] = preset
+
+    def remove_style_preset(self, name: str) -> None:
+        """Remove a style preset by name.
+
+        Args:
+            name: Preset name to remove.
+        """
+        self.style_presets.pop(name, None)
+
+    def remove_formatter_preset(self, name: str) -> None:
+        """Remove a formatter preset by name.
+
+        Args:
+            name: Preset name to remove.
+        """
+        self.formatter_presets.pop(name, None)
+
+    # ------------------------------------------------------------------
+    # Private helpers
+    # ------------------------------------------------------------------
+
+    def _get_data_services_manager(self) -> Optional[DataServicesManager]:
+        """Return the DataServicesManager for this session."""
+        return InstancesManager.get_by_type(
+            self._session, DataServicesManager, default=None
+        )
+
+    def _get_service(self, table_name: str):
+        """Return the DataService for a given table name.
+
+        Args:
+            table_name: Fully qualified table name.
+
+        Returns:
+            DataService instance, or None if not found.
+        """
+        manager = self._get_data_services_manager()
+        if manager is None:
+            return None
+        for service in manager._services.values():
+            if service.table_name == table_name:
+                return service
+        return None

src/myfasthtml/core/formula/dsl/grammar.py

@@ -75,7 +75,7 @@ FORMULA_GRAMMAR = r"""
     where_clause: TABLE_COL_REF "=" COL_NAME
 
     // TABLE_COL_REF matches "TableName.ColumnName" (dot-separated, no spaces)
-    TABLE_COL_REF: /[A-Za-z_][A-Za-z0-9_]*\.[A-Za-z_][A-Za-z0-9_]*/
+    TABLE_COL_REF: /[A-Za-z_][A-Za-z0-9_]*(\.[A-Za-z_][A-Za-z0-9_]*)+/
     COL_NAME: /[A-Za-z_][A-Za-z0-9_ ]*/
 
     // ==================== Functions ====================

src/myfasthtml/core/formula/dsl/transformer.py

@@ -171,16 +171,16 @@ class FormulaTransformer(Transformer):
   # ==================== References ====================
   
   def cross_ref_simple(self, items):
-    """{ Table.Column }"""
+    """{ Table.Column } or { namespace.Table.Column }"""
     table_col = str(items[0])
-    table, column = table_col.split(".", 1)
+    table, column = table_col.rsplit(".", 1)
     return CrossTableRef(table=table, column=column)
 
   def cross_ref_where(self, items):
     """{ Table.Column WHERE remote_table.remote_col = local_col }"""
     table_col = str(items[0])
     where = items[1]
-    table, column = table_col.split(".", 1)
+    table, column = table_col.rsplit(".", 1)
     return CrossTableRef(table=table, column=column, where_clause=where)
   
   def column_ref(self, items):
@@ -192,7 +192,7 @@ class FormulaTransformer(Transformer):
     """TABLE_COL_REF = COL_NAME"""
     remote_table_col = str(items[0])
     local_col = str(items[1]).strip()
-    remote_table, remote_col = remote_table_col.split(".", 1)
+    remote_table, remote_col = remote_table_col.rsplit(".", 1)
     return WhereClause(
       remote_table=remote_table,
       remote_column=remote_col,

src/myfasthtml/core/formula/engine.py

@@ -12,6 +12,7 @@ from typing import Any, Callable, Optional
 
 import numpy as np
 
+from myfasthtml.core.dsl.exceptions import DSLSyntaxError
 from .dataclasses import FormulaDefinition, WhereClause
 from .dependency_graph import DependencyGraph
 from .dsl.parser import get_parser
@@ -43,7 +44,7 @@ def parse_formula(text: str) -> FormulaDefinition | None:
   parser = get_parser()
   tree = parser.parse(text)
   if tree is None:
-    return None
+    raise DSLSyntaxError(message=f"Formula could not be parsed: '{text}'")
 
   transformer = FormulaTransformer()
   formula = transformer.transform(tree)

tests/controls/test_datagrid.py

@@ -0,0 +1,955 @@
+import pandas as pd
+import pytest
+from fasthtml.components import Div, Script
+
+from myfasthtml.controls.DataGrid import DataGrid, DatagridConf
+from myfasthtml.controls.DataGridsManager import DataGridsManager
+from myfasthtml.controls.TabsManager import TabsManager
+from myfasthtml.core.constants import ColumnType
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
+from myfasthtml.core.instances import InstancesManager
+from myfasthtml.test.matcher import AnyValue, Contains, NoChildren, find, find_one, matches, TestLabel, TestObject
+
+
+@pytest.fixture
+def datagrids_manager(root_instance):
+  """Create a DataGridsManager instance for testing."""
+  InstancesManager.reset()
+  TabsManager(root_instance)  # just define it
+  return DataGridsManager(root_instance)
+
+
+@pytest.fixture
+def datagrid(datagrids_manager):
+  """Create an empty DataGrid for testing."""
+  return DataGrid(datagrids_manager)
+
+
+@pytest.fixture
+def datagrid_with_data(datagrids_manager):
+  """Create a DataGrid loaded with a sample DataFrame (name, age, active columns)."""
+  session = datagrids_manager._session
+  dsm = InstancesManager.get_by_type(session, DataServicesManager)
+  
+  df = pd.DataFrame({
+      "name": ["Alice", "Bob", "Charlie"],
+      "age": [25, 30, 35],
+      "active": [True, False, True],
+  })
+  
+  data_service = dsm.create_service("test.grid1", save_state=False)
+  data_service.load_dataframe(df)
+  
+  grid_id = DataGrid.get_grid_id_from_data_service_id(data_service.get_id())
+  conf = DatagridConf(namespace="test", name="grid1")
+  return DataGrid(datagrids_manager, conf=conf, save_state=False, _id=grid_id)
+
+
+@pytest.fixture
+def datagrid_with_full_data(datagrids_manager):
+  """Create a DataGrid covering all basic column types: Text, Number and Bool.
+
+  Designed to be extended with additional column types (Datetime, Enum, Formula)
+  as they are added to the test suite.
+  """
+  session = datagrids_manager._session
+  dsm = InstancesManager.get_by_type(session, DataServicesManager)
+  
+  df = pd.DataFrame({
+      "name": ["Alice", "Bob", "Charlie"],  # Text
+      "age": [25, 30, 35],  # Number
+      "active": [True, False, True],  # Bool
+  })
+  
+  data_service = dsm.create_service("test.full_types", save_state=False)
+  data_service.load_dataframe(df)
+  
+  grid_id = DataGrid.get_grid_id_from_data_service_id(data_service.get_id())
+  conf = DatagridConf(namespace="test", name="full_types")
+  dg = DataGrid(datagrids_manager, conf=conf, save_state=False, _id=grid_id)
+  
+  # Assign distinct widths (name=150, age=80, active=120) so tests that rely
+  # on column widths are not masked by a uniform default value of 100px.
+  for col_ui_state, width in zip(dg._state.columns, [150, 80, 120]):
+    col_ui_state.width = width
+  dg._init_columns()
+  
+  return dg
+
+
+@pytest.fixture
+def datagrid_no_edition(datagrid_with_data):
+  """DataGrid with edition disabled (no RowSelection column, no add-column button)."""
+  dg = datagrid_with_data
+  dg._settings.enable_edition = False
+  dg._init_columns()
+  return dg
+
+
+class TestDataGridBehaviour:
+  def test_i_can_create_empty_datagrid(self, datagrids_manager):
+    dg = DataGrid(datagrids_manager)
+    assert dg is not None
+  
+  # ------------------------------------------------------------------
+  # Table Name
+  # ------------------------------------------------------------------
+  
+  def test_i_can_get_table_name_with_namespace(self, datagrids_manager):
+    """Test that get_table_name() returns 'namespace.name' when a namespace is set.
+
+    The dot-separated format is required by FormulaEngine for cross-table
+    reference resolution.
+    """
+    conf = DatagridConf(namespace="reports", name="sales")
+    dg = DataGrid(datagrids_manager, conf=conf)
+    assert dg.get_table_name() == "reports.sales"
+  
+  def test_i_can_get_table_name_without_namespace(self, datagrids_manager):
+    """Test that get_table_name() returns just the name when namespace is absent.
+
+    Grids without a namespace use the plain name, which is valid for
+    single-namespace applications.
+    """
+    conf = DatagridConf(name="employees")
+    dg = DataGrid(datagrids_manager, conf=conf)
+    assert dg.get_table_name() == "employees"
+  
+  # ------------------------------------------------------------------
+  # Element ID Generation
+  # ------------------------------------------------------------------
+  
+  @pytest.mark.parametrize("mode, pos, expected_template", [
+      ("row", (2, 5), "trow_{id}-5"),
+      ("column", (2, 5), "tcol_{id}-2"),
+      ("cell", (2, 5), "tcell_{id}-2-5"),
+  ])
+  def test_i_can_get_element_id_from_pos(self, datagrid, mode, pos, expected_template):
+    """Test that _get_element_id_from_pos generates the correct element ID per selection mode.
+
+    Why these cases matter:
+    - 'row' mode: ID references the row index (pos[1]) to target HTMX row updates.
+    - 'column' mode: ID references the column index (pos[0]) for column highlighting.
+    - 'cell' mode: ID includes both indices for precise cell targeting and navigation.
+    """
+    result = datagrid._get_element_id_from_pos(mode, pos)
+    expected = expected_template.format(id=datagrid._id)
+    assert result == expected
+  
+  def test_i_can_get_element_id_when_pos_is_none(self, datagrid):
+    """Test that _get_element_id_from_pos returns None when position is None.
+
+    None position means no cell is selected; the caller must receive None
+    to know that no DOM element should be targeted.
+    """
+    assert datagrid._get_element_id_from_pos("cell", None) is None
+  
+  # ------------------------------------------------------------------
+  # Element ID Parsing
+  # ------------------------------------------------------------------
+  
+  def test_i_can_get_pos_from_cell_element_id(self, datagrid):
+    """Test that _get_pos_from_element_id correctly parses (col, row) from a cell ID.
+
+    The position tuple (col, row) is used for cell navigation and selection
+    state tracking. Correct parsing is required for keyboard navigation and
+    mouse selection to target the right cell.
+    """
+    element_id = f"tcell_{datagrid._id}-3-7"
+    assert datagrid._get_pos_from_element_id(element_id) == (3, 7)
+  
+  def test_i_can_get_pos_returns_none_for_non_cell_id(self, datagrid):
+    """Test that _get_pos_from_element_id returns None for row IDs and None input.
+
+    Row and column IDs don't carry a (col, row) position. Returning None
+    signals that no cell-level position can be derived.
+    """
+    assert datagrid._get_pos_from_element_id(f"trow_{datagrid._id}-5") is None
+    assert datagrid._get_pos_from_element_id(None) is None
+  
+  # ------------------------------------------------------------------
+  # Static ID Conversions
+  # ------------------------------------------------------------------
+  
+  def test_i_can_convert_grid_id_to_data_service_id_and_back(self, datagrid):
+    """Test that the grid↔data-service ID conversion is a perfect round-trip.
+
+    DataGrid and DataService share the same UUID but use different prefixes.
+    The round-trip property ensures neither conversion loses information, so
+    DataGrid can always locate its companion DataService from its own ID.
+    """
+    grid_id = datagrid.get_id()
+    data_service_id = DataGrid.get_data_service_id_from_data_grid_id(grid_id)
+    assert data_service_id != grid_id
+    assert DataGrid.get_grid_id_from_data_service_id(data_service_id) == grid_id
+  
+  # ------------------------------------------------------------------
+  # Column Management
+  # ------------------------------------------------------------------
+  
+  def test_i_can_move_column(self, datagrid_with_data):
+    """Test that move_column reorders _state.columns correctly.
+
+    Column order in _state.columns drives both header and body rendering.
+    Moving the last column to the first position verifies the pop-then-insert
+    logic for the source-after-target case.
+    """
+    dg = datagrid_with_data
+    initial_order = [c.col_id for c in dg._state.columns]
+    last_col_id = initial_order[-1]
+    first_col_id = initial_order[0]
+    
+    dg.move_column(last_col_id, first_col_id)
+    
+    new_order = [c.col_id for c in dg._state.columns]
+    assert new_order[0] == last_col_id
+    
+    # init_columns keeps the order
+    dg._init_columns()
+    assert [c.col_id for c in dg._columns][1:] == new_order
+  
+  def test_i_cannot_move_column_when_column_not_found(self, datagrid_with_data):
+    """Test that move_column does not alter state when a col_id is unknown.
+
+    An unknown column ID must be silently ignored (logged as a warning) so
+    that stale JS events cannot corrupt the column order.
+    """
+    dg = datagrid_with_data
+    initial_order = [c.col_id for c in dg._state.columns]
+    
+    dg.move_column("nonexistent_col", initial_order[0])
+    
+    assert [c.col_id for c in dg._state.columns] == initial_order
+  
+  def test_i_can_move_column_to_same_position(self, datagrid_with_data):
+    """Test that move_column is a no-op when source and target are the same column.
+
+    Dropping a column header back onto itself must not change the order,
+    preventing unnecessary state saves and re-renders.
+    """
+    dg = datagrid_with_data
+    initial_order = [c.col_id for c in dg._state.columns]
+    first_col_id = initial_order[0]
+    
+    dg.move_column(first_col_id, first_col_id)
+    
+    assert [c.col_id for c in dg._state.columns] == initial_order
+  
+  def test_i_can_set_column_width(self, datagrid_with_data):
+    """Test that handle_set_column_width persists the new pixel width in state.
+
+    Column width in _state.columns is the source of truth for header and
+    body cell sizing. A correct update ensures the resized width survives
+    the next render cycle.
+    """
+    dg = datagrid_with_data
+    col_id = dg._state.columns[0].col_id
+    
+    dg.handle_set_column_width(col_id, "300")
+    
+    col = next(c for c in dg._state.columns if c.col_id == col_id)
+    assert col.width == 300
+  
+  # ------------------------------------------------------------------
+  # Column Width Calculation
+  # ------------------------------------------------------------------
+  
+  def test_i_can_calculate_optimal_column_width_returns_default_for_unknown_column(self, datagrid):
+    """Test that calculate_optimal_column_width returns 150 for an unknown column.
+
+    A safe default prevents layout breakage when a col_id from JS does not
+    match any known column (e.g. after a column was deleted mid-session).
+    """
+    result = datagrid.calculate_optimal_column_width("nonexistent_col")
+    assert result == 150
+  
+  @pytest.mark.parametrize("col_title, expected_width", [
+      ("name", 86),  # data dominates: max("name"=4, "Charlie"=7) → 7*8+30=86
+      ("age", 54),  # title dominates: max("age"=3, "35"=2) → 3*8+30=54
+      ("active", 78),  # title dominates: max("active"=6, "False"=5) → 6*8+30=78
+  ])
+  def test_i_can_calculate_optimal_column_width_with_data(self, datagrid_with_data, col_title, expected_width):
+    """Test that calculate_optimal_column_width returns the correct pixel width based on content.
+
+    The formula max(title_length, max_data_length) * 8 + 30 must produce the
+    right width for each column so that the auto-size feature fits all visible
+    content without truncation.
+
+    Why these cases matter:
+    - 'name' (data dominates): "Charlie" (7 chars) > "name" (4 chars) → 86px
+    - 'age' (title dominates): "age" (3 chars) > "35" (2 chars) → 54px
+    - 'active' (title dominates): "active" (6 chars) > "False" (5 chars) → 78px
+    """
+    dg = datagrid_with_data
+    col = next(c for c in dg._columns if c.title == col_title)
+    
+    result = dg.calculate_optimal_column_width(col.col_id)
+    
+    assert result == expected_width
+  
+  # ------------------------------------------------------------------
+  # Selection and Interaction
+  # ------------------------------------------------------------------
+  
+  def test_i_can_on_key_pressed_esc_clears_selection(self, datagrid):
+    """Test that pressing ESC resets both the focused cell and extra selections.
+
+    ESC is the standard 'deselect all' shortcut. Both selected and
+    extra_selected must be cleared so the grid visually deselects everything
+    and subsequent navigation starts from a clean state.
+    """
+    dg = datagrid
+    dg._state.selection.selected = (1, 2)
+    dg._state.selection.extra_selected.append(("range", (0, 0, 2, 2)))
+    
+    dg.on_key_pressed("esc", has_focus=True, is_inside=True)
+    
+    assert dg._state.selection.selected is None
+    assert dg._state.selection.extra_selected == []
+  
+  def test_i_can_on_click_outside_does_not_update_position(self, datagrid):
+    """Test that a click outside the grid does not change the selected position.
+
+    Clicks on surrounding UI elements must not accidentally move the grid
+    cursor, so is_inside=False must be a complete no-op for selection state.
+    """
+    dg = datagrid
+    dg._state.selection.selected = (1, 2)
+    
+    dg.on_click("click", is_inside=False, cell_id=f"tcell_{dg._id}-1-2")
+    
+    assert dg._state.selection.selected == (1, 2)
+  
+  def test_i_can_on_click_on_cell_updates_position(self, datagrid):
+    """Test that clicking a cell sets selection.selected to the cell's (col, row) position.
+
+    The selected position drives the focus highlight rendered by
+    mk_selection_manager. Correct parsing and storage ensures the right cell
+    is visually highlighted after each click.
+    """
+    dg = datagrid
+    cell_id = f"tcell_{dg._id}-2-5"
+    
+    dg.on_click("click", is_inside=True, cell_id=cell_id)
+    
+    assert dg._state.selection.selected == (2, 5)
+  
+  def test_i_can_change_selection_mode(self, datagrid):
+    """Test that change_selection_mode updates selection_mode from the CycleStateControl.
+
+    The selection mode (row / column / cell) determines how clicks and
+    keyboard events interpret the selected position. The grid must store
+    the mode chosen by the cycle selector so rendering and JS handlers
+    receive a consistent value.
+    """
+    dg = datagrid
+    assert dg._state.selection.selection_mode == "cell"
+    
+    dg._selection_mode_selector.cycle_state()
+    dg.change_selection_mode()
+    assert dg._state.selection.selection_mode == "row"
+    
+    dg._selection_mode_selector.cycle_state()
+    dg.change_selection_mode()
+    assert dg._state.selection.selection_mode == "column"
+    
+    dg._selection_mode_selector.cycle_state()
+    dg.change_selection_mode()
+    assert dg._state.selection.selection_mode == "cell"
+
+
+class TestDataGridRender:
+  
+  # ------------------------------------------------------------------
+  # Global structure (UTR-11.1)
+  # ------------------------------------------------------------------
+  
+  def test_no_data_layout_is_rendered(self, datagrid):
+    """Test that DataGrid renders a placeholder when no DataFrame is loaded.
+
+    Why these elements matter:
+    - Div tag: render must return a valid element, not raise
+    - "No data to display !": gives clear feedback and prevents a crash
+      when the DataService has no loaded DataFrame
+    """
+    html = datagrid.render()
+    assert matches(html, Div("No data to display !"))
+  
+  def test_layout_is_rendered(self, datagrid_with_data):
+    """Test that DataGrid renders all 4 main structural sections when data is loaded.
+
+    Why these elements matter:
+    - id=dg._id: root ID required for JS init (initDataGrid) and HTMX targeting
+    - cls Contains 'grid': CSS grid layout controls header/body row sizing
+    - child[0] Div: filter bar + toolbar icons
+    - child[1] DoNotCheck: Panel containing the scrollable table
+    - child[2] Script: initDataGrid call that activates JS behaviour
+    - child[3] DoNotCheck: Keyboard handler for in-grid shortcuts
+    """
+    dg = datagrid_with_data
+    html = dg.render()
+    expected = Div(
+      Div(
+        TestObject("Query"),  # filter bar
+        Div(),  # toolbar icons
+        cls=Contains("flex"),
+      ),
+      TestObject("Panel"),  # Panel containing the table
+      Script(Contains("initDataGrid")),  # initDataGrid script
+      Div(),  # Mouse and Keyboard handler
+      id=dg._id,
+      cls=Contains("grid"),
+    )
+    assert matches(html, expected)
+  
+  # ------------------------------------------------------------------
+  # Selection Manager
+  # ------------------------------------------------------------------
+  
+  def test_i_can_render_selection_manager_with_no_selection(self, datagrid):
+    """Test that the selection manager renders with no children when nothing is selected.
+
+    Why these elements matter:
+    - id=tsm_{id}: required by updateDatagridSelection JS to locate the manager
+    - NoChildren: an empty manager signals JS that nothing is selected,
+      preventing stale highlight artefacts between renders
+    """
+    dg = datagrid
+    html = dg.mk_selection_manager()
+    expected = Div(
+      NoChildren(),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  def test_i_can_render_selection_manager_with_selected_cell(self, datagrid):
+    """Test that the selection manager renders one focus element when a cell is selected.
+
+    Why these elements matter:
+    - Div child: each selected element becomes a child directive read by JS
+    - selection_type='focus': tells JS to apply the focus highlight style
+    - element_id: the cell DOM ID that JS will scroll to and highlight
+    """
+    dg = datagrid
+    dg._state.selection.selected = (2, 5)
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      Div(selection_type="focus", element_id=f"tcell_{dg._id}-2-5"),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  @pytest.mark.parametrize("mode", ["row", "column", "range", None])
+  def test_i_can_render_selection_manager_selection_mode_attribute(self, datagrid, mode):
+    """Test that the selection_mode attribute on the manager Div reflects the current mode.
+
+    Why this element matters:
+    - selection_mode attribute: read by updateDatagridSelection JS to know which
+      highlight strategy to apply (row stripe, column stripe, range rectangle, or
+      single cell focus). All four valid values must round-trip correctly.
+    """
+    dg = datagrid
+    dg._state.selection.selection_mode = mode
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      id=f"tsm_{dg._id}",
+      selection_mode=f"{mode}",
+    )
+    assert matches(html, expected)
+  
+  def test_i_can_render_extra_selected_row(self, datagrid):
+    """Test that a row extra-selection entry renders as a Div with selection_type='row'.
+
+    Why these elements matter:
+    - selection_type='row': JS applies the row-stripe highlight to the entire row
+    - element_id: the DOM ID of the row element that JS will highlight
+    """
+    dg = datagrid
+    row_element_id = f"trow_{dg._id}-3"
+    dg._state.selection.extra_selected.append(("row", row_element_id))
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      Div(selection_type="row", element_id=row_element_id),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  def test_i_can_render_extra_selected_column(self, datagrid):
+    """Test that a column extra-selection entry renders as a Div with selection_type='column'.
+
+    Why these elements matter:
+    - selection_type='column': JS applies the column-stripe highlight to the entire column
+    - element_id: the DOM ID of the column header element that JS will highlight
+    """
+    dg = datagrid
+    col_element_id = f"tcol_{dg._id}-2"
+    dg._state.selection.extra_selected.append(("column", col_element_id))
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      Div(selection_type="column", element_id=col_element_id),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  def test_i_can_render_extra_selected_range(self, datagrid):
+    """Test that a range extra-selection entry renders with the tuple stringified as element_id.
+
+    Why these elements matter:
+    - selection_type='range': JS draws a rectangular highlight over the cell region
+    - element_id=str(tuple): the range bounds (min_col, min_row, max_col, max_row)
+      are passed as a string; JS parses this to locate all cells in the rectangle
+    """
+    dg = datagrid
+    range_bounds = (0, 0, 2, 2)
+    dg._state.selection.extra_selected.append(("range", range_bounds))
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      Div(selection_type="range", element_id=f"{range_bounds}"),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  def test_i_can_render_selection_manager_with_multiple_extra_selected(self, datagrid):
+    """Test that all extra_selected entries are rendered as individual children.
+
+    Why these elements matter:
+    - 3 children: each extra selection must produce exactly one Div child;
+      missing entries would cause JS to miss highlights, extra entries would
+      cause phantom highlights on unselected cells
+    """
+    dg = datagrid
+    dg._state.selection.extra_selected.append(("row", f"trow_{dg._id}-1"))
+    dg._state.selection.extra_selected.append(("column", f"tcol_{dg._id}-0"))
+    dg._state.selection.extra_selected.append(("range", (1, 1, 3, 3)))
+    
+    html = dg.mk_selection_manager()
+    
+    children = find(html, Div(selection_type=AnyValue()))
+    assert len(children) == 3, "Each extra_selected entry must produce exactly one child Div"
+  
+  def test_i_can_render_selection_manager_with_focus_and_extra_selected(self, datagrid):
+    """Test that a focused cell and an extra selection are both rendered as children.
+
+    Why these elements matter:
+    - focus child: the primary selection must always be present when selected is set
+    - extra child: secondary selections must not replace the focus entry; both must
+      coexist so JS can apply focus highlight and secondary highlights simultaneously
+    """
+    dg = datagrid
+    dg._state.selection.selected = (2, 5)
+    dg._state.selection.extra_selected.append(("row", f"trow_{dg._id}-5"))
+    
+    html = dg.mk_selection_manager()
+    
+    expected = Div(
+      Div(selection_type="focus", element_id=f"tcell_{dg._id}-2-5"),
+      Div(selection_type="row", element_id=f"trow_{dg._id}-5"),
+      id=f"tsm_{dg._id}",
+    )
+    assert matches(html, expected)
+  
+  # ------------------------------------------------------------------
+  # Headers
+  # ------------------------------------------------------------------
+  
+  def test_i_can_render_headers(self, datagrid_with_data):
+    """Test that the header row renders with the correct ID and one cell per visible data column.
+
+    Why these elements matter:
+    - id=th_{id}: targeted by reset_column_width to swap the header after auto-size
+    - cls Contains 'dt2-header': CSS hook for header styling and sticky positioning
+    - 3 data cells: one resizable cell per data column; verifies all DataFrame
+      columns are represented (RowSelection and add-button columns excluded)
+    """
+    dg = datagrid_with_data
+    html = dg.mk_headers()
+    
+    # Step 1: Validate global header structure
+    expected = Div(
+      id=f"th_{dg._id}",
+      cls=Contains("dt2-header"),
+    )
+    assert matches(html, expected)
+    
+    # Step 2: Count the visible data column headers
+    col_headers = find(html, Div(cls=Contains("dt2-cell", "dt2-resizable")))
+    assert len(col_headers) == 3, "Should have one resizable header cell per visible data column"
+  
+  def test_i_can_render_row_selection_header_in_edition_mode(self, datagrid_with_data):
+    """Test that a RowSelection header cell is rendered when edition mode is enabled.
+
+    Why these elements matter:
+    - dt2-row-selection: the selection checkbox column is only meaningful in edition
+      mode where rows can be individually selected for bulk operations; JS uses this
+      cell to anchor the row-selection toggle handler
+    - exactly 1 cell: a second dt2-row-selection would double the checkbox column
+    """
+    dg = datagrid_with_data
+    html = dg.mk_headers()
+    
+    row_sel_cells = find(html, Div(cls=Contains("dt2-row-selection")))
+    assert len(row_sel_cells) == 1, "Edition mode must render exactly one row-selection header cell"
+  
+  def test_i_cannot_render_row_selection_header_without_edition_mode(self, datagrid_no_edition):
+    """Test that no RowSelection header cell is rendered when edition mode is disabled.
+
+    Why this matters:
+    - Without edition, there is no row selection column in _columns; rendering one
+      would create an orphan cell misaligned with the body rows
+    """
+    dg = datagrid_no_edition
+    html = dg.mk_headers()
+    
+    row_sel_cells = find(html, Div(cls=Contains("dt2-row-selection")))
+    assert len(row_sel_cells) == 0, "Without edition mode, no row-selection header cell should be rendered"
+  
+  def test_i_can_render_add_column_button_in_edition_mode(self, datagrid_with_data):
+    """Test that the add-column button is appended to the header in edition mode.
+
+    Why this element matters:
+    - dt2-add-column: the '+' icon at the end of the header lets users add new
+      columns interactively; it must be present in edition mode and absent otherwise
+      to avoid exposing mutation UI in read-only grids
+    """
+    dg = datagrid_with_data
+    html = dg.mk_headers()
+    
+    add_col_cells = find(html, Div(cls=Contains("dt2-add-column")))
+    assert len(add_col_cells) == 1, "Edition mode must render exactly one add-column button"
+  
+  def test_i_cannot_render_add_column_button_without_edition_mode(self, datagrid_no_edition):
+    """Test that no add-column button is rendered when edition mode is disabled.
+
+    Why this matters:
+    - Read-only grids must not expose mutation controls; the absence of dt2-add-column
+      guarantees that the JS handler for toggling the column editor is never reachable
+    """
+    dg = datagrid_no_edition
+    html = dg.mk_headers()
+    
+    add_col_cells = find(html, Div(cls=Contains("dt2-add-column")))
+    assert len(add_col_cells) == 0, "Without edition mode, no add-column button should be rendered"
+  
+  def test_i_can_render_headers_in_column_order(self, datagrid_with_data):
+    """Test that resizable header cells appear in the same order as self._columns.
+
+    Why this matters:
+    - The header order drives column alignment with body rows; a mismatch between
+      header and body column order produces visually broken tables where header
+      labels do not correspond to the values beneath them
+    - data-col: JS drag-and-drop uses this attribute to identify the column being
+      moved; it must match the col_id in _columns for move_column to work correctly
+    """
+    dg = datagrid_with_data
+    html = dg.mk_headers()
+    
+    expected_order = [c.col_id for c in dg._columns if c.type != ColumnType.RowSelection_]
+    rendered_cells = find(html, Div(cls=Contains("dt2-cell", "dt2-resizable")))
+    rendered_order = [cell.attrs.get("data-col") for cell in rendered_cells]
+    
+    assert rendered_order == expected_order, (
+        f"Header column order {rendered_order} does not match _columns order {expected_order}"
+    )
+  
+  def test_i_can_render_hidden_column_is_not_rendered(self, datagrid_with_data):
+    """Test that a hidden column produces no header cell.
+
+    Why this matters:
+    - Columns with visible=False must be completely absent from the header so that
+      body cell alignment is preserved; a hidden header cell would shift all
+      subsequent column headers one position to the right
+    """
+    dg = datagrid_with_data
+    dg._state.columns[0].visible = False
+    dg._init_columns()
+    
+    html = dg.mk_headers()
+    
+    col_headers = find(html, Div(cls=Contains("dt2-cell", "dt2-resizable")))
+    assert len(col_headers) == 2, "Hiding one column must reduce the resizable header count from 3 to 2"
+  
+  @pytest.mark.parametrize("col_title, expected_icon", [
+      ("name", "text_field20_regular"),  # Text column
+      ("age", "number_row20_regular"),  # Number column
+      ("active", "checkbox_checked20_filled"),  # Bool column
+  ])
+  def test_i_can_render_header_cell_with_correct_icon_for_column_type(
+      self, datagrid_with_full_data, col_title, expected_icon):
+    """Test that each column header renders the correct type icon in its label.
+
+    Why these elements matter:
+    - TestLabel(col_title, icon): the label inside _mk_header_name must carry the
+      right icon so the user can visually identify the column type at a glance
+    - data-col: used to locate the cell in the tree; ensures we test the right column
+    - Parametrization covers the three fundamental types (Text, Number, Bool) so
+      that adding a new type requires only a new row in the parametrize decorator
+    """
+    dg = datagrid_with_full_data
+    col = next(c for c in dg._columns if c.title == col_title)
+    
+    html = dg.mk_headers()
+    
+    # Step 1: Find the header cell for this column
+    cell = find_one(html, Div(data_col=col.col_id, cls=Contains("dt2-cell", "dt2-resizable")))
+    
+    # Step 2: Verify the header name section contains the correct icon
+    expected = Div(
+      Div(
+        TestLabel(col_title, icon=expected_icon),
+      ),
+    )
+    assert matches(cell, expected)
+  
+  @pytest.mark.parametrize("col_title, expected_width", [
+      ("name", 150),
+      ("age", 80),
+      ("active", 120),
+  ])
+  def test_i_can_render_header_cell_width_matches_state(
+      self, datagrid_with_full_data, col_title, expected_width):
+    """Test that the style attribute of each header cell reflects its width in _columns.
+
+    Why these elements matter:
+    - style Contains 'width:{n}px': the inline style is the sole mechanism that
+      sizes each column; a mismatch between _columns state and the rendered style
+      would produce misaligned headers and body cells after a resize operation
+    - Distinct widths (150 / 80 / 120) in the fixture guarantee that a correct
+      match cannot happen by accident with a uniform default value
+    """
+    dg = datagrid_with_full_data
+    col = next(c for c in dg._columns if c.title == col_title)
+    
+    html = dg.mk_headers()
+    
+    cell = find_one(html, Div(data_col=col.col_id, cls=Contains("dt2-cell", "dt2-resizable")))
+    
+    assert matches(cell, Div(style=Contains(f"width:{expected_width}px")))
+  
+  def test_i_can_render_header_resize_handle_has_correct_commands(self, datagrid_with_full_data):
+    """Test that every resizable header cell contains a resize handle with the correct command IDs.
+
+    Why these elements matter:
+    - dt2-resize-handle: the DOM target for JS drag-to-resize; its absence completely
+      disables column resizing for the affected column
+    - data-command-id: JS fires this command on mouseup to persist the new width to
+      the server; an incorrect ID would silently discard every resize action
+    - data-reset-command-id: JS fires this command on double-click to auto-size the
+      column; an incorrect ID would break the double-click reset feature
+    - 3 handles: one per data column; a missing handle disables resize for that column
+    """
+    dg = datagrid_with_full_data
+    resize_cmd_id = dg.commands.set_column_width().id
+    reset_cmd_id = dg.commands.reset_column_width().id
+    
+    html = dg.mk_headers()
+    
+    handles = find(html, Div(
+      cls=Contains("dt2-resize-handle"),
+      data_command_id=resize_cmd_id,
+      data_reset_command_id=reset_cmd_id,
+    ))
+    assert len(handles) == 3, "Each data column must have exactly one resize handle with correct command IDs"
+  
+  # ------------------------------------------------------------------
+  # Body
+  # ------------------------------------------------------------------
+  
+  def test_i_can_render_body_wrapper(self, datagrid_with_data):
+    """Test that the body wrapper renders with the correct ID and CSS class.
+
+    Why these elements matter:
+    - id=tb_{id}: targeted by get_page to append new rows during lazy loading
+      and by render_partial('body') to swap the whole body on filter/sort
+    - cls Contains 'dt2-body-container': CSS hook that enables the custom
+      JS scrollbar and overflow handling
+    """
+    dg = datagrid_with_data
+    html = dg.mk_body_wrapper()
+    expected = Div(
+      id=f"tb_{dg._id}",
+      cls=Contains("dt2-body-container"),
+    )
+    assert matches(html, expected)
+  
+  # ------------------------------------------------------------------
+  # Body rows
+  # ------------------------------------------------------------------
+  
+  def test_i_can_render_body_row_count(self, datagrid_with_full_data):
+    """Test that mk_body_content_page returns one row per DataFrame row plus the add-row button.
+
+    Why these elements matter:
+    - 3 rows: one per DataFrame row (Alice, Bob, Charlie); a missing row means data
+      was silently dropped during rendering
+    - 1 add-row button: required for edition mode; its absence disables row insertion
+    - total 4: the exact count prevents both missing and phantom rows
+    """
+    dg = datagrid_with_full_data
+    
+    rows = dg.mk_body_content_page(0)
+    
+    assert len(rows) == 4, "Expected 3 data rows and 1 add-row button in edition mode"
+  
+  def test_i_can_render_row_structure(self, datagrid_with_full_data):
+    """Test that a rendered row carries the correct ID, CSS class and data-row attribute.
+
+    Why these elements matter:
+    - id=tr_{id}-{row_index}: targeted by handle_add_row (outerHTML swap) and lazy
+      loading; an incorrect ID breaks row-level HTMX updates
+    - cls Contains 'dt2-row': CSS hook for row styling and hover effects
+    - data_row: read by JS click handlers to identify which row was interacted with
+    """
+    dg = datagrid_with_full_data
+    len_columns_1 = len(dg._columns) - 1
+    
+    row = dg.mk_row(0, None, len_columns_1)
+    
+    expected = Div(
+      id=f"tr_{dg._id}-0",
+      cls=Contains("dt2-row"),
+      data_row="0",
+    )
+    assert matches(row, expected)
+  
+  def test_i_can_render_row_has_correct_cell_count(self, datagrid_with_full_data):
+    """Test that each row renders exactly one cell per column in _columns.
+
+    Why this matters:
+    - cell count must equal column count: a mismatch shifts body cells out of
+      alignment with header cells, producing a visually broken table
+    - includes RowSelection_: the selection column must be present so the body
+      grid matches the header grid column by column
+    """
+    dg = datagrid_with_full_data
+    len_columns_1 = len(dg._columns) - 1
+    
+    row = dg.mk_row(0, None, len_columns_1)
+    
+    assert len(row.children) == len(dg._columns), (
+        f"Row must have {len(dg._columns)} cells (one per column), got {len(row.children)}"
+    )
+  
+  # ------------------------------------------------------------------
+  # Body cells
+  # ------------------------------------------------------------------
+  
+  def test_i_can_render_body_cell_structure(self, datagrid_with_full_data):
+    """Test that a data cell carries the correct ID, class, style and data-col attribute.
+
+    Why these elements matter:
+    - id=tcell_{id}-{col_pos}-{row_index}: parsed by _get_pos_from_element_id for
+      click/keyboard selection; an incorrect ID breaks cell-level navigation
+    - cls Contains 'dt2-cell': base CSS class for cell layout and borders
+    - style Contains 'width:{n}px': inline style that aligns the cell with its header;
+      a mismatch causes the column to appear misaligned after rendering
+    - data_col: read by JS resize drag to identify which column is being resized
+    """
+    dg = datagrid_with_full_data
+    name_col = next(c for c in dg._columns if c.title == "name")
+    col_pos = dg._columns.index(name_col)
+    
+    cell = dg.mk_body_cell(col_pos, 0, name_col, None, is_last=False)
+    
+    expected = Div(
+      id=f"tcell_{dg._id}-{col_pos}-0",
+      cls=Contains("dt2-cell"),
+      style=Contains(f"width:{name_col.width}px"),
+      data_col=name_col.col_id,
+    )
+    assert matches(cell, expected)
+  
+  def test_i_can_render_last_body_cell_has_dt2_last_cell_class(self, datagrid_with_full_data):
+    """Test that the last cell in a row carries the dt2-last-cell class.
+
+    Why this element matters:
+    - dt2-last-cell: CSS hook that removes the right border on the final column
+      so the table edge looks clean; applied only to the last column to avoid
+      removing borders on intermediate cells
+    """
+    dg = datagrid_with_full_data
+    last_col_pos = len(dg._columns) - 1
+    last_col_def = dg._columns[last_col_pos]
+    
+    cell = dg.mk_body_cell(last_col_pos, 0, last_col_def, None, is_last=True)
+    
+    assert matches(cell, Div(cls=Contains("dt2-cell", "dt2-last-cell")))
+  
+  def test_i_can_render_body_cell_row_selection(self, datagrid_with_full_data):
+    """Test that the RowSelection_ column renders a selection cell, not a data cell.
+
+    Why this element matters:
+    - dt2-row-selection: the DOM hook used by JS to toggle per-row checkboxes in
+      edition mode; it must not carry dt2-cell markup which would give it a data
+      cell appearance and break the selection column layout
+    """
+    dg = datagrid_with_full_data
+    row_sel_col = dg._columns[0]  # RowSelection_ is always inserted first
+    
+    cell = dg.mk_body_cell(0, 0, row_sel_col, None, is_last=False)
+    
+    assert matches(cell, Div(cls=Contains("dt2-row-selection")))
+  
+  def test_i_cannot_render_body_cell_when_hidden(self, datagrid_with_full_data):
+    """Test that mk_body_cell returns None for a column marked as not visible.
+
+    Why this matters:
+    - None return: the row renderer filters out None values so the hidden column
+      produces no DOM element; any non-None return would insert a phantom cell
+      and shift all subsequent cells one position to the right
+    """
+    dg = datagrid_with_full_data
+    dg._state.columns[0].visible = False  # hide "name" (first data column)
+    dg._init_columns()
+    
+    hidden_col = next(c for c in dg._columns if c.title == "name")
+    col_pos = dg._columns.index(hidden_col)
+    
+    cell = dg.mk_body_cell(col_pos, 0, hidden_col, None, is_last=False)
+    
+    assert cell is None
+  
+  # ------------------------------------------------------------------
+  # Body cell content
+  # ------------------------------------------------------------------
+  
+  @pytest.mark.parametrize("col_title, expected_css_class, expected_value", [
+      ("name", "dt2-cell-content-text", "Alice"),
+      ("age", "dt2-cell-content-number", "25"),
+      ("active", "dt2-cell-content-checkbox", None),
+  ])
+  def test_i_can_render_body_cell_content_for_column_type(
+      self, datagrid_with_full_data, col_title, expected_css_class, expected_value):
+    """Test that cell content carries the correct CSS class and value for each column type.
+
+    Why these elements matter:
+    - dt2-cell-content-text / number / checkbox: type-specific CSS classes that
+      control text alignment, font weight and boolean display; an incorrect class
+      makes numbers left-aligned or text right-aligned
+    - expected_value ('Alice', '25'): the actual DataFrame value must appear in the
+      rendered content so the cell is not empty or showing a wrong row
+    - Bool uses None for expected_value: the value is an icon, not a text string,
+      so only the wrapper class is verified
+    """
+    dg = datagrid_with_full_data
+    col_def = next(c for c in dg._columns if c.title == col_title)
+    col_pos = dg._columns.index(col_def)
+    
+    content = dg.mk_body_cell_content(col_pos, 0, col_def, None)
+    
+    assert expected_css_class in str(content), (
+        f"Expected CSS class '{expected_css_class}' in cell content for column '{col_title}'"
+    )
+    if expected_value is not None:
+      assert expected_value in str(content), (
+          f"Expected value '{expected_value}' in cell content for column '{col_title}'"
+      )

tests/controls/test_datagrid_formatting.py

@@ -9,7 +9,7 @@ import pytest
 from myfasthtml.controls.DataGrid import DataGrid
 from myfasthtml.controls.DataGridFormattingEditor import DataGridFormattingEditor
 from myfasthtml.controls.DataGridsManager import DataGridsManager
-from myfasthtml.controls.datagrid_objects import DataGridColumnState, DataGridRowState
+from myfasthtml.controls.datagrid_objects import DataGridColumnState, DataGridRowUiState
 from myfasthtml.core.constants import ColumnType
 from myfasthtml.core.formatting.dataclasses import FormatRule, Style
 from myfasthtml.core.formatting.dsl.definition import FormattingDSL
@@ -38,8 +38,8 @@ def datagrid(manager):
   
   # Add some rows
   grid._state.rows = [
-      DataGridRowState(0),
-      DataGridRowState(1),
+      DataGridRowUiState(0),
+      DataGridRowUiState(1),
   ]
   
   yield grid

tests/controls/test_datagridsmanager.py

@@ -3,9 +3,11 @@ import shutil
 
 import pytest
 
+from myfasthtml.controls.DataGrid import DataGrid
 from myfasthtml.controls.DataGridsManager import DataGridsManager
 from myfasthtml.controls.TabsManager import TabsManager
 from myfasthtml.controls.TreeView import TreeNode
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
 from myfasthtml.core.instances import InstancesManager
 from .conftest import root_instance
 
@@ -16,7 +18,7 @@ def cleanup_db():
 
 
 @pytest.fixture
-def datagrid_manager(root_instance):
+def datagrids_manager(root_instance):
   """Create a DataGridsManager instance for testing."""
   InstancesManager.reset()
   TabsManager(root_instance)  # just define it
@@ -26,7 +28,7 @@ def datagrid_manager(root_instance):
 class TestDataGridsManagerBehaviour:
   """Tests for DataGridsManager behavior and logic."""
   
-  def test_i_can_create_new_grid_with_nothing_selected(self, datagrid_manager):
+  def test_i_can_create_new_grid_with_nothing_selected(self, datagrids_manager):
     """Test creating a new grid when no node is selected.
 
     Verifies that:
@@ -35,10 +37,10 @@ class TestDataGridsManagerBehaviour:
     - Node is selected and in edit mode
     - Document definition is created
     """
-    result = datagrid_manager.new_grid()
+    datagrids_manager.handle_new_grid()
     
     # Verify tree structure
-    tree = datagrid_manager._tree
+    tree = datagrids_manager._tree
     assert len(tree._state.items) == 2, "Should have Untitled folder + Sheet1 node"
     
     # Find the Untitled folder and Sheet1 node
@@ -55,13 +57,13 @@ class TestDataGridsManagerBehaviour:
     assert tree._state.editing == sheet.id, "Sheet1 should be in edit mode"
     
     # Verify document definition
-    assert len(datagrid_manager._state.elements) == 1, "Should have one document"
-    doc = datagrid_manager._state.elements[0]
+    assert len(datagrids_manager._state.elements) == 1, "Should have one document"
+    doc = datagrids_manager._state.elements[0]
     assert doc.namespace == "Untitled"
     assert doc.name == "Sheet1"
     assert doc.type == "excel"
   
-  def test_i_can_create_new_grid_under_selected_folder(self, datagrid_manager):
+  def test_i_can_create_new_grid_under_selected_folder(self, datagrids_manager):
     """Test creating a new grid when a folder is selected.
 
     Verifies that:
@@ -69,24 +71,24 @@ class TestDataGridsManagerBehaviour:
     - Namespace matches folder name
     """
     # Create a folder and select it
-    folder_id = datagrid_manager._tree.ensure_path("MyFolder")
-    datagrid_manager._tree._select_node(folder_id)
+    folder_id = datagrids_manager._tree.ensure_path("MyFolder")
+    datagrids_manager._tree._select_node(folder_id)
     
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
     # Verify the new grid is under MyFolder
-    tree = datagrid_manager._tree
+    tree = datagrids_manager._tree
     nodes = list(tree._state.items.values())
     sheet = [n for n in nodes if n.label == "Sheet1"][0]
     
     assert sheet.parent == folder_id, "Sheet1 should be under MyFolder"
     
     # Verify document definition
-    doc = datagrid_manager._state.elements[0]
+    doc = datagrids_manager._state.elements[0]
     assert doc.namespace == "MyFolder"
     assert doc.name == "Sheet1"
   
-  def test_i_can_create_new_grid_under_selected_leaf_parent(self, datagrid_manager):
+  def test_i_can_create_new_grid_under_selected_leaf_parent(self, datagrids_manager):
     """Test creating a new grid when a leaf node is selected.
 
     Verifies that:
@@ -94,62 +96,62 @@ class TestDataGridsManagerBehaviour:
     - Not under the leaf itself
     """
     # Create a folder with a leaf
-    folder_id = datagrid_manager._tree.ensure_path("MyFolder")
+    folder_id = datagrids_manager._tree.ensure_path("MyFolder")
     leaf = TreeNode(label="ExistingSheet", type="excel", parent=folder_id)
-    datagrid_manager._tree.add_node(leaf, parent_id=folder_id)
+    datagrids_manager._tree.add_node(leaf, parent_id=folder_id)
     
     # Select the leaf
-    datagrid_manager._tree._select_node(leaf.id)
+    datagrids_manager._tree._select_node(leaf.id)
     
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
     # Verify the new grid is under MyFolder (not under ExistingSheet)
-    tree = datagrid_manager._tree
+    tree = datagrids_manager._tree
     nodes = list(tree._state.items.values())
     new_sheet = [n for n in nodes if n.label == "Sheet1"][0]
     
     assert new_sheet.parent == folder_id, "Sheet1 should be under MyFolder (leaf's parent)"
     assert new_sheet.parent != leaf.id, "Sheet1 should not be under the leaf"
   
-  def test_new_grid_generates_unique_sheet_names(self, datagrid_manager):
+  def test_new_grid_generates_unique_sheet_names(self, datagrids_manager):
     """Test that new_grid generates unique sequential sheet names.
 
     Verifies Sheet1, Sheet2, Sheet3... generation.
     """
     # Create first grid
-    datagrid_manager.new_grid()
-    assert datagrid_manager._state.elements[0].name == "Sheet1"
+    datagrids_manager.handle_new_grid()
+    assert datagrids_manager._state.elements[0].name == "Sheet1"
     
     # Create second grid
-    datagrid_manager.new_grid()
-    assert datagrid_manager._state.elements[1].name == "Sheet2"
+    datagrids_manager.handle_new_grid()
+    assert datagrids_manager._state.elements[1].name == "Sheet2"
     
     # Create third grid
-    datagrid_manager.new_grid()
-    assert datagrid_manager._state.elements[2].name == "Sheet3"
+    datagrids_manager.handle_new_grid()
+    assert datagrids_manager._state.elements[2].name == "Sheet3"
   
-  def test_new_grid_expands_parent_folder(self, datagrid_manager):
+  def test_new_grid_expands_parent_folder(self, datagrids_manager):
     """Test that creating a new grid automatically expands the parent folder.
 
     Verifies parent is added to tree._state.opened.
     """
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
-    tree = datagrid_manager._tree
+    tree = datagrids_manager._tree
     nodes = list(tree._state.items.values())
     untitled = [n for n in nodes if n.label == "Untitled"][0]
     
     # Verify parent is expanded
     assert untitled.id in tree._state.opened, "Parent folder should be expanded"
   
-  def test_new_grid_selects_and_edits_new_node(self, datagrid_manager):
+  def test_new_grid_selects_and_edits_new_node(self, datagrids_manager):
     """Test that new grid node is both selected and in edit mode.
 
     Verifies _state.selected and _state.editing are set to new node.
     """
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
-    tree = datagrid_manager._tree
+    tree = datagrids_manager._tree
     nodes = list(tree._state.items.values())
     sheet = [n for n in nodes if n.label == "Sheet1"][0]
     
@@ -159,17 +161,17 @@ class TestDataGridsManagerBehaviour:
     # Verify edit mode
     assert tree._state.editing == sheet.id, "New node should be in edit mode"
   
-  def test_new_grid_creates_document_definition(self, datagrid_manager):
+  def test_new_grid_creates_document_definition(self, datagrids_manager):
     """Test that new_grid creates a DocumentDefinition with correct fields.
 
     Verifies document_id, namespace, name, type, tab_id, datagrid_id.
     """
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
     # Verify document was created
-    assert len(datagrid_manager._state.elements) == 1, "Should have one document"
+    assert len(datagrids_manager._state.elements) == 1, "Should have one document"
     
-    doc = datagrid_manager._state.elements[0]
+    doc = datagrids_manager._state.elements[0]
     
     # Verify all fields
     assert doc.document_id is not None, "Should have document_id"
@@ -180,33 +182,34 @@ class TestDataGridsManagerBehaviour:
     assert doc.tab_id is not None, "Should have tab_id"
     assert doc.datagrid_id is not None, "Should have datagrid_id"
   
-  def test_new_grid_creates_datagrid_and_registers(self, datagrid_manager):
+  def test_new_grid_creates_datagrid_and_registers(self, datagrids_manager):
     """Test that new_grid creates a DataGrid and registers it.
 
     Verifies DataGrid exists and is in registry with namespace.name.
     """
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
-    doc = datagrid_manager._state.elements[0]
+    doc = datagrids_manager._state.elements[0]
     
     # Verify DataGrid is registered
-    tables = datagrid_manager._registry.get_all_tables()
-    assert "Untitled.Sheet1" in tables, "DataGrid should be registered as Untitled.Sheet1"
+    entries = datagrids_manager._registry.get_all_entries()
+    assert doc.datagrid_id in entries, "DataGrid should be registered by grid_id"
+    assert entries[doc.datagrid_id] == ("Untitled", "Sheet1"), "Registry entry should match namespace and name"
     
     # Verify DataGrid exists in InstancesManager
     from myfasthtml.core.instances import InstancesManager
-    datagrid = InstancesManager.get(datagrid_manager._session, doc.datagrid_id, None)
+    datagrid = InstancesManager.get(datagrids_manager._session, doc.datagrid_id, None)
     assert datagrid is not None, "DataGrid instance should exist"
   
-  def test_new_grid_creates_tab_with_datagrid(self, datagrid_manager):
+  def test_new_grid_creates_tab_with_datagrid(self, datagrids_manager):
     """Test that new_grid creates a tab with correct label and content.
 
     Verifies tab is created via TabsManager with DataGrid as content.
     """
-    result = datagrid_manager.new_grid()
+    result = datagrids_manager.handle_new_grid()
     
-    doc = datagrid_manager._state.elements[0]
-    tabs_manager = datagrid_manager._tabs_manager
+    doc = datagrids_manager._state.elements[0]
+    tabs_manager = datagrids_manager._tabs_manager
     
     # Verify tab exists in TabsManager
     assert doc.tab_id in tabs_manager._state.tabs, "Tab should exist in TabsManager"
@@ -215,47 +218,82 @@ class TestDataGridsManagerBehaviour:
     tab_metadata = tabs_manager._state.tabs[doc.tab_id]
     assert tab_metadata['label'] == "Sheet1", "Tab label should be Sheet1"
   
-  def test_generate_unique_sheet_name_with_no_children(self, datagrid_manager):
+  def test_i_can_access_data_service_after_new_grid(self, datagrids_manager):
+    """Test that a DataService is accessible via datagrid_id after new_grid().
+
+    Why this matters:
+    - DataService is now created first and its id becomes the datagrid_id
+    - Verifies the DataService is properly registered in DataServicesManager
+    - Ensures the link between datagrid_id and DataService is established
+    """
+    datagrids_manager.handle_new_grid()
+    
+    doc = datagrids_manager._state.elements[0]
+    datagrid_id = doc.datagrid_id
+    dataservice_id = DataGrid.get_data_service_id_from_data_grid_id(datagrid_id)
+    
+    dataservice_manager = InstancesManager.get_by_type(datagrids_manager._session, DataServicesManager)
+    service = dataservice_manager.get_service(dataservice_id)
+    
+    assert service is not None, "DataService should be accessible via dataservice_id (taken from datagrid_id)"
+  
+  def test_new_grid_data_service_has_correct_table_name(self, datagrids_manager):
+    """Test that the DataService receives the correct table_name after new_grid().
+
+    Why this matters:
+    - table_name is now passed to create_service() before the DataGrid exists
+    - Verifies the namespace.name format is correctly built and assigned
+    """
+    datagrids_manager.handle_new_grid()
+    
+    doc = datagrids_manager._state.elements[0]
+    dataservice_manager = InstancesManager.get_by_type(datagrids_manager._session, DataServicesManager)
+    dataservice_id = DataGrid.get_data_service_id_from_data_grid_id(doc.datagrid_id)
+    data_service = dataservice_manager.get_service(dataservice_id)
+    
+    assert data_service.table_name == "Untitled.Sheet1", "DataService table_name should be 'Untitled.Sheet1'"
+  
+  def test_generate_unique_sheet_name_with_no_children(self, datagrids_manager):
     """Test _generate_unique_sheet_name on an empty folder.
 
     Verifies it returns "Sheet1" when no children exist.
     """
-    folder_id = datagrid_manager._tree.ensure_path("EmptyFolder")
+    folder_id = datagrids_manager._tree.ensure_path("EmptyFolder")
     
-    name = datagrid_manager._generate_unique_sheet_name(folder_id)
+    name = datagrids_manager._generate_unique_sheet_name(folder_id)
     
     assert name == "Sheet1", "Should generate Sheet1 for empty folder"
   
-  def test_generate_unique_sheet_name_with_existing_sheets(self, datagrid_manager):
+  def test_generate_unique_sheet_name_with_existing_sheets(self, datagrids_manager):
     """Test _generate_unique_sheet_name with existing sheets.
 
     Verifies it generates the next sequential number.
     """
-    folder_id = datagrid_manager._tree.ensure_path("MyFolder")
+    folder_id = datagrids_manager._tree.ensure_path("MyFolder")
     
     # Add Sheet1 and Sheet2 manually
     sheet1 = TreeNode(label="Sheet1", type="excel", parent=folder_id)
     sheet2 = TreeNode(label="Sheet2", type="excel", parent=folder_id)
-    datagrid_manager._tree.add_node(sheet1, parent_id=folder_id)
-    datagrid_manager._tree.add_node(sheet2, parent_id=folder_id)
+    datagrids_manager._tree.add_node(sheet1, parent_id=folder_id)
+    datagrids_manager._tree.add_node(sheet2, parent_id=folder_id)
     
-    name = datagrid_manager._generate_unique_sheet_name(folder_id)
+    name = datagrids_manager._generate_unique_sheet_name(folder_id)
     
     assert name == "Sheet3", "Should generate Sheet3 when Sheet1 and Sheet2 exist"
   
-  def test_generate_unique_sheet_name_skips_gaps(self, datagrid_manager):
+  def test_generate_unique_sheet_name_skips_gaps(self, datagrids_manager):
     """Test _generate_unique_sheet_name fills gaps in sequence.
 
     Verifies it generates Sheet2 when Sheet1 and Sheet3 exist (missing Sheet2).
     """
-    folder_id = datagrid_manager._tree.ensure_path("MyFolder")
+    folder_id = datagrids_manager._tree.ensure_path("MyFolder")
     
     # Add Sheet1 and Sheet3 (skip Sheet2)
     sheet1 = TreeNode(label="Sheet1", type="excel", parent=folder_id)
     sheet3 = TreeNode(label="Sheet3", type="excel", parent=folder_id)
-    datagrid_manager._tree.add_node(sheet1, parent_id=folder_id)
-    datagrid_manager._tree.add_node(sheet3, parent_id=folder_id)
+    datagrids_manager._tree.add_node(sheet1, parent_id=folder_id)
+    datagrids_manager._tree.add_node(sheet3, parent_id=folder_id)
     
-    name = datagrid_manager._generate_unique_sheet_name(folder_id)
+    name = datagrids_manager._generate_unique_sheet_name(folder_id)
     
     assert name == "Sheet2", "Should generate Sheet2 to fill the gap"

tests/core/data/conftest.py

@@ -0,0 +1,44 @@
+import shutil
+
+import pytest
+from dbengine.handlers import handlers
+
+from myfasthtml.core.data.DataServicesManager import DataServicesManager
+from myfasthtml.core.dbengine_utils import DataFrameHandler
+from myfasthtml.core.dbmanager import DbManager
+from myfasthtml.core.instances import SingleInstance, InstancesManager
+
+
+@pytest.fixture(scope="session")
+def session():
+  handlers.register_handler(DataFrameHandler())
+  return {
+      "user_info": {
+          "id": "test_tenant_id",
+          "email": "test@email.com",
+          "username": "test user",
+          "role": [],
+      }
+  }
+
+
+@pytest.fixture
+def parent(session):
+  instance = SingleInstance(session=session, _id="test_parent_id")
+  return instance
+
+
+@pytest.fixture
+def db_manager(parent):
+  shutil.rmtree("TestDb", ignore_errors=True)
+  db_manager_instance = DbManager(parent, root="TestDb", auto_register=True)
+
+  yield db_manager_instance
+
+  shutil.rmtree("TestDb", ignore_errors=True)
+  InstancesManager.reset()
+
+
+@pytest.fixture
+def dsm(parent, db_manager):
+  return DataServicesManager(parent, parent._session)

tests/core/data/test_dataservice.py

@@ -0,0 +1,213 @@
+"""Unit tests for DataService."""
+import pandas as pd
+import pytest
+
+from myfasthtml.core.constants import ColumnType
+from myfasthtml.core.data.ColumnDefinition import ColumnDefinition
+
+
+class TestDataInitialisation:
+  """Tests for the Data initialisation section of DataService."""
+  
+  @pytest.fixture
+  def service(self, dsm):
+    return dsm.create_service("ns.tbl", save_state=False)
+  
+  def test_i_can_load_a_dataframe(self, service):
+    """load_dataframe() populates the store and column definitions."""
+    df = pd.DataFrame({"name": ["Alice", "Bob"], "age": [30, 25]})
+    
+    service.load_dataframe(df)
+    
+    assert service.get_store().ne_df is not None
+    assert service.get_store().ns_total_rows == 2
+    assert len(service.columns) == 2
+  
+  def test_i_can_load_an_empty_dataframe(self, service):
+    """load_dataframe() with empty DataFrame sets total_rows to 0."""
+    service.load_dataframe(pd.DataFrame())
+    
+    assert service.get_store().ns_total_rows == 0
+    assert service.columns == []
+  
+  def test_i_can_load_dataframe_without_reinitializing_columns(self, service):
+    """load_dataframe(init_columns=False) preserves existing column definitions."""
+    df = pd.DataFrame({"a": [1]})
+    service.load_dataframe(df)
+    original_columns = list(service.columns)
+    
+    df2 = pd.DataFrame({"a": [1, 2], "b": [3, 4]})
+    service.load_dataframe(df2, init_columns=False)
+    
+    assert service.columns == original_columns
+  
+  def test_i_can_load_none_dataframe_without_error(self, service):
+    """load_dataframe(None) is a no-op and does not raise.
+
+    Why this matters:
+    - Early return on None protects against uninitialized callers.
+    - ne_df must remain None (no side effects on the store).
+    """
+    service.load_dataframe(None)
+    
+    assert service.get_store().ne_df is None
+  
+  def test_i_can_load_dataframe_with_column_name_normalization(self, service):
+    """load_dataframe() normalizes column names to safe IDs via make_safe_id.
+
+    Why this matters:
+    - Columns with spaces or special characters must be accessible as safe IDs.
+    - make_safe_id lowercases and replaces non-safe characters with underscores.
+    """
+    df = pd.DataFrame({"First Name": ["Alice"], "Last Name": ["Smith"]})
+    
+    service.load_dataframe(df)
+    
+    col_ids = [c.col_id for c in service.columns]
+    assert col_ids == ["first_name", "last_name"]
+
+
+class TestMutations:
+  """Tests for the Mutations section of DataService."""
+  
+  @pytest.fixture
+  def service(self, dsm):
+    svc = dsm.create_service("ns.mutations", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"value": [1, 2, 3]}))
+    return svc
+  
+  def test_i_can_add_a_row(self, service):
+    """add_row() appends a row with default values and updates the caches."""
+    service.add_row()
+    
+    assert service.get_store().ns_total_rows == 4
+    assert len(service.get_store().ne_df) == 4
+  
+  def test_i_can_add_a_row_with_custom_data(self, service):
+    """add_row() with explicit data stores the provided values."""
+    service.add_row(row_data={"value": 99})
+    
+    assert service.get_store().ne_df.iloc[-1]["value"] == 99
+  
+  def test_i_can_set_data(self, service):
+    """set_data() updates the cell in the DataFrame, fast-access cache, and row data."""
+    service.set_data("value", 1, 99)
+    
+    assert service.get_store().ne_df.at[1, "value"] == 99
+    assert service.get_store().ns_fast_access["value"][1] == 99
+    assert service.get_store().ns_row_data[1]["value"] == 99
+  
+  @pytest.mark.parametrize("col_type, expected_default", [
+      (ColumnType.Text, ""),
+      (ColumnType.Number, 0),
+      (ColumnType.Bool, False),
+      (ColumnType.Datetime, pd.NaT),
+      (ColumnType.Choice, ""),
+      (ColumnType.Enum, ""),
+      (ColumnType.RowSelection_, ""),
+  ])
+  def test_i_can_add_column_with_correct_default_value(self, service, col_type, expected_default):
+    """add_column() creates a DataFrame column with the type-appropriate default value.
+
+    Why these assertions matter:
+    - col_id in ne_df.columns: Confirms the column is materialized in the DataFrame.
+    - len(columns) == 2: Confirms the column is registered in the metadata.
+    - default value: Each type has a specific sentinel value; wrong defaults corrupt data.
+    - pd.isna() for Datetime: pd.NaT does not support equality comparison.
+    """
+    col_def = ColumnDefinition(col_id="__new__", col_index=-1, title="New Col", type=col_type)
+    service.add_column(col_def)
+    
+    assert col_def.col_id in service.get_store().ne_df.columns
+    assert len(service.columns) == 2
+    actual = service.get_store().ne_df[col_def.col_id].iloc[0]
+    if pd.isna(expected_default):
+      assert pd.isna(actual)
+    else:
+      assert actual == expected_default
+  
+  @pytest.mark.parametrize("col_type", [ColumnType.Formula, ColumnType.RowIndex])
+  def test_i_can_add_virtual_column_without_dataframe_column(self, service, col_type):
+    """add_column() with virtual types does not create a DataFrame column.
+
+    Why these assertions matter:
+    - col_id not in ne_df.columns: Virtual columns are computed, not stored in the DataFrame.
+    - col_index == -1: Sentinel value marking virtual columns.
+    - len(columns) == 2: Column is registered in the state metadata despite being virtual.
+    """
+    col_def = ColumnDefinition(col_id="__new__", col_index=-1, title="Virtual", type=col_type)
+    service.add_column(col_def)
+    
+    assert col_def.col_id not in service.get_store().ne_df.columns
+    assert col_def.col_index == -1
+    assert len(service.columns) == 2
+  
+  def test_i_can_add_row_without_loaded_dataframe_without_error(self, dsm):
+    """add_row() is a no-op and does not raise when no DataFrame is loaded."""
+    service = dsm.create_service("ns.nodf_row", save_state=False)
+    
+    service.add_row()
+    
+    assert service.get_store().ne_df is None
+  
+  def test_i_can_set_data_without_loaded_dataframe_without_error(self, dsm):
+    """set_data() is a no-op and does not raise when no DataFrame is loaded."""
+    service = dsm.create_service("ns.nodf_set", save_state=False)
+    
+    service.set_data("x", 0, 42)
+    
+    assert service.get_store().ne_df is None
+
+
+class TestFormulaManagement:
+  """Tests for the Formula management section of DataService."""
+  
+  @pytest.fixture
+  def service(self, dsm):
+    svc = dsm.create_service("ns.formula", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"a": [1, 2, 3]}))
+    return svc
+  
+  def test_i_can_get_table_name(self, service):
+    """table_name property returns the value set at creation."""
+    assert service.table_name == "ns.formula"
+  
+  def test_i_can_update_table_name(self, service):
+    """set_table_name() updates the table name."""
+    service.set_table_name("ns.new_name")
+    
+    assert service.table_name == "ns.new_name"
+  
+  def test_i_can_register_formula(self, service):
+    """register_formula() registers a formula in the shared FormulaEngine.
+
+    Why these assertions matter:
+    - has_formula: Confirms the formula was registered in the engine's DAG.
+    - get_formula_text: Confirms the source expression is stored as-is.
+    """
+    service.register_formula("computed", "{a} + 1")
+    
+    engine = service.get_formula_engine()
+    assert engine.has_formula("ns.formula", "computed")
+    assert engine.get_formula_text("ns.formula", "computed") == "{a} + 1"
+  
+  def test_i_can_remove_formula(self, service):
+    """remove_formula() unregisters a formula from the FormulaEngine."""
+    service.register_formula("computed", "{a} + 1")
+    service.remove_formula("computed")
+
+    engine = service.get_formula_engine()
+    assert not engine.has_formula("ns.formula", "computed")
+
+  def test_i_cannot_register_invalid_formula(self, service):
+    """register_formula() with invalid DSL syntax does not register the formula.
+
+    Why this matters:
+    - parse_formula() raises DSLSyntaxError when it cannot parse the expression.
+    - register_formula() catches the exception to protect the caller, but the
+      formula must remain absent from the engine — not silently removed.
+    """
+    service.register_formula("computed", "invalid syntax without braces")
+
+    engine = service.get_formula_engine()
+    assert not engine.has_formula("ns.formula", "computed")

tests/core/data/test_dataservice_formula_integration.py

@@ -0,0 +1,210 @@
+"""Integration tests for DataService formula evaluation through DataServicesManager.
+
+These tests exercise the full stack: DataServicesManager owns the FormulaEngine
+and provides the registry_resolver that enables cross-table formula resolution.
+Each test uses real DataService instances and real DataStore objects — no fakes.
+"""
+import pytest
+import pandas as pd
+
+from myfasthtml.core.constants import ColumnType
+from myfasthtml.core.data.ColumnDefinition import ColumnDefinition
+
+
+class TestIntraTableFormula:
+  """Single-table formula evaluation through the DSM/DataService stack."""
+
+  @pytest.fixture
+  def service(self, dsm):
+    svc = dsm.create_service("ns.sales", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"price": [10, 20, 30], "qty": [2, 3, 4]}))
+    return svc
+
+  def test_i_can_evaluate_formula_on_single_table(self, service):
+    """register_formula() + ensure_ready() computes the column for all rows.
+
+    Why these assertions matter:
+    - ns_fast_access["total"]: ensure_ready() writes results to the cache used by rendering.
+    - All three rows: verifies the formula is applied to every row, not just the first.
+    """
+    service.register_formula("total", "{price} * {qty}")
+    service.ensure_ready()
+
+    result = service.get_store().ns_fast_access["total"]
+    assert result[0] == 20
+    assert result[1] == 60
+    assert result[2] == 120
+
+  def test_i_can_reevaluate_formula_after_data_change(self, service):
+    """set_data() marks dependent formula columns dirty; ensure_ready() recomputes them.
+
+    Why these assertions matter:
+    - result[0] updated: confirms the dirty flag propagated and row 0 was recomputed.
+    - result[1] unchanged: confirms only affected rows are recomputed (no unnecessary work).
+    """
+    service.register_formula("total", "{price} * {qty}")
+    service.ensure_ready()
+
+    service.set_data("price", 0, 100)
+    service.ensure_ready()
+
+    result = service.get_store().ns_fast_access["total"]
+    assert result[0] == 200
+    assert result[1] == 60
+
+
+class TestCrossTableFormula:
+  """Cross-table formula resolution via DataServicesManager.registry_resolver.
+
+  Table names use namespace notation (e.g. "ns.products"). The DSL grammar
+  now supports multiple dots in TABLE_COL_REF; the transformer splits on the
+  last dot to separate the table name from the column name.
+  """
+
+  @pytest.fixture
+  def orders_service(self, dsm):
+    svc = dsm.create_service("ns.orders", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"qty": [2, 3]}))
+    return svc
+
+  @pytest.fixture
+  def products_service(self, dsm):
+    svc = dsm.create_service("ns.products", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"price": [10, 20]}))
+    return svc
+
+  def test_i_can_evaluate_cross_table_formula(self, orders_service, products_service):
+    """A formula in one service can reference a namespaced table from another service.
+
+    Why these assertions matter:
+    - result[0] and result[1]: confirms the registry_resolver resolved "ns.products"
+      correctly and combined its data with the orders data row by row.
+    """
+    orders_service.register_formula("total", "{ns.products.price} * {qty}")
+    orders_service.ensure_ready()
+
+    result = orders_service.get_store().ns_fast_access["total"]
+    assert result[0] == 20
+    assert result[1] == 60
+
+  def test_i_cannot_resolve_cross_table_formula_for_unknown_table(self, orders_service):
+    """A formula referencing an unregistered table resolves to None without raising.
+
+    Why this matters:
+    - result is None: confirms the engine degrades gracefully when the resolver
+      returns None, instead of raising or producing corrupt values.
+    """
+    orders_service.register_formula("total", "{ns.unknown_table.price} * {qty}")
+    orders_service.ensure_ready()
+
+    result = orders_service.get_store().ns_fast_access["total"]
+    assert result[0] is None
+    assert result[1] is None
+
+
+class TestCrossTableFormulaWhere:
+  """Cross-table formula resolution using an explicit WHERE clause.
+
+  The WHERE clause scans the remote table for a row where remote_column == local_value,
+  enabling correct lookups regardless of row ordering between tables.
+  """
+
+  @pytest.fixture
+  def orders_service(self, dsm):
+    svc = dsm.create_service("ns.orders", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"product_id": [2, 1], "qty": [3, 5]}))
+    return svc
+
+  @pytest.fixture
+  def products_service(self, dsm):
+    svc = dsm.create_service("ns.products", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"product_id": [1, 2], "price": [10, 20]}))
+    return svc
+
+  def test_i_can_lookup_value_with_where_clause_non_sequential(self, orders_service, products_service):
+    """WHERE resolves the correct remote row even when tables are not aligned by position.
+
+    Why these assertions matter:
+    - result[0] == 60: order row 0 has product_id=2, products row 1 has price=20 → 20*3=60.
+    - result[1] == 50: order row 1 has product_id=1, products row 0 has price=10 → 10*5=50.
+    Row-index fallback would return 10*3=30 and 20*5=100 — both wrong.
+    """
+    orders_service.register_formula(
+      "total",
+      "{ns.products.price where ns.products.product_id = product_id} * {qty}"
+    )
+    orders_service.ensure_ready()
+
+    result = orders_service.get_store().ns_fast_access["total"]
+    assert result[0] == 60
+    assert result[1] == 50
+    
+
+  def test_i_can_lookup_returns_none_when_no_match(self, orders_service, products_service):
+    """WHERE returns None when the local value has no matching row in the remote table.
+
+    Why this matters:
+    - result[0] is None: product_id=2 exists in products, but product_id=99 does not.
+    - No exception is raised: the engine must degrade gracefully on missing lookups.
+    """
+    orders_service_no_match = orders_service
+    orders_service_no_match.get_store().ns_fast_access["product_id"][0] = 99
+
+    orders_service_no_match.register_formula(
+      "total",
+      "{ns.products.price where ns.products.product_id = product_id} * {qty}"
+    )
+    orders_service_no_match.ensure_ready()
+
+    result = orders_service_no_match.get_store().ns_fast_access["total"]
+    assert result[0] is None
+
+
+class TestFormulaLifecycle:
+  """End-to-end formula lifecycle: column creation, registration, and evaluation."""
+
+  @pytest.fixture
+  def service(self, dsm):
+    svc = dsm.create_service("ns.lifecycle", save_state=False)
+    svc.load_dataframe(pd.DataFrame({"a": [1, 2, 3]}))
+    return svc
+
+  def test_i_can_add_formula_column_and_evaluate(self, service):
+    """add_column(Formula) + register_formula() + ensure_ready() produces computed values.
+
+    Why these assertions matter:
+    - col_id in ns_fast_access: ensure_ready() must write the formula column into the cache.
+    - Values [2, 4, 6]: validates the formula expression is correctly applied to all rows.
+    """
+    col_def = ColumnDefinition(col_id="__new__", col_index=-1,
+                               title="Doubled", type=ColumnType.Formula,
+                               formula="{a} * 2")
+    service.add_column(col_def)
+    service.register_formula(col_def.col_id, col_def.formula)
+    service.ensure_ready()
+
+    result = service.get_store().ns_fast_access[col_def.col_id]
+    assert list(result) == [2, 4, 6]
+
+  def test_i_can_evaluate_formula_after_adding_row(self, service):
+    """add_row() marks formula columns dirty; ensure_ready() computes the new row.
+
+    Why these assertions matter:
+    - len(result) == 4: confirms the new row was appended and the cache extended.
+    - result[3] == 20: confirms the formula was recalculated for the new row (a=10, * 2).
+    - result[0] == 2: confirms existing rows are not corrupted by the recalculation.
+    """
+    col_def = ColumnDefinition(col_id="__new__", col_index=-1,
+                               title="Doubled", type=ColumnType.Formula,
+                               formula="{a} * 2")
+    service.add_column(col_def)
+    service.register_formula(col_def.col_id, col_def.formula)
+    service.ensure_ready()
+
+    service.add_row(row_data={"a": 10})
+    service.ensure_ready()
+
+    result = service.get_store().ns_fast_access[col_def.col_id]
+    assert len(result) == 4
+    assert result[0] == 2
+    assert result[3] == 20

tests/core/data/test_dataservicesmanager.py

@@ -0,0 +1,121 @@
+"""Unit tests for DataServicesManager."""
+import pandas as pd
+
+
+class TestDataServicesManagerServiceLifecycle:
+  def test_i_can_create_a_service(self, dsm):
+    """create_service() returns a DataService accessible by grid_id."""
+    service = dsm.create_service("ns.tbl", save_state=False)
+    
+    assert service is not None
+    assert service.get_id() is not None
+    assert dsm.get_service(service.get_id()) is service
+  
+  def test_i_can_create_service_with_correct_table_name(self, dsm):
+    """create_service() sets the table_name on the returned DataService.
+  
+    create_service() calls service.set_table_name() internally.
+    This test verifies the side effect is applied before returning the service.
+    """
+    service = dsm.create_service("ns.my_table", save_state=False)
+    
+    assert service.table_name == "ns.my_table"
+  
+  def test_i_can_create_service_forcing_the_id(self, dsm):
+    """create_service() sets the table_name on the returned DataService.
+  
+    create_service() calls service.set_table_name() internally.
+    This test verifies the side effect is applied before returning the service.
+    """
+    service = dsm.create_service("ns.my_table", _id="grid_id", save_state=False)
+    
+    assert service.get_id() == "grid_id"
+  
+  def test_i_can_get_a_service_by_grid_id(self, dsm):
+    """get_service() returns the correct service."""
+    svc1 = dsm.create_service("ns.t1", _id="g1", save_state=False)
+    svc2 = dsm.create_service("ns.t2", _id="g2", save_state=False)
+    
+    assert dsm.get_service("g1") is svc1
+    assert dsm.get_service("g2") is svc2
+  
+  def test_i_cannot_get_a_nonexistent_service(self, dsm):
+    """get_service() returns None for unknown grid_id."""
+    assert dsm.get_service("does_not_exist") is None
+  
+  def test_i_can_remove_a_service(self, dsm):
+    """remove_service() unregisters the service."""
+    service = dsm.create_service("ns.rm", save_state=False)
+    dsm.remove_service(service.get_id())
+    
+    assert dsm.get_service(service.get_id()) is None
+  
+  def test_i_can_remove_a_nonexistent_service_without_error(self, dsm):
+    """remove_service() on unknown grid_id does not raise."""
+    dsm.remove_service("ghost")  # should not raise
+  
+  def test_i_can_restore_a_service(self, dsm):
+    """restore_service() creates and registers a service if not already present."""
+    service = dsm.restore_service("grid_restore")
+    
+    assert service is not None
+    assert dsm.get_service("grid_restore") is service
+    assert service.get_id() == "grid_restore"
+  
+  def test_i_can_restore_existing_service(self, dsm):
+    """restore_service() returns the existing service when already registered."""
+    original = dsm.create_service("ns.e", _id="grid_exist", save_state=False)
+    restored = dsm.restore_service("grid_exist")
+    
+    assert restored is original
+
+
+class TestDataServicesManagerFormulaEngine:
+  
+  def test_i_can_get_formula_engine(self, dsm):
+    """get_formula_engine() returns the shared FormulaEngine instance."""
+    engine = dsm.get_formula_engine()
+    assert engine is not None
+  
+  def test_i_can_verify_shared_formula_engine(self, dsm):
+    """All services share the same FormulaEngine from DataServicesManager."""
+    svc1 = dsm.create_service("ns.fe1", save_state=False)
+    svc2 = dsm.create_service("ns.fe2", save_state=False)
+    
+    assert svc1.get_formula_engine() is svc2.get_formula_engine()
+    assert svc1.get_formula_engine() is dsm.get_formula_engine()
+  
+  def test_i_can_resolve_store_by_table_name(self, dsm):
+    """FormulaEngine resolver finds the DataStore for a given table name."""
+    service = dsm.create_service("ns.resolver", save_state=False)
+    df = pd.DataFrame({"a": [1, 2]})
+    service.load_dataframe(df)
+    
+    store = dsm._resolve_store_for_table("ns.resolver")
+    
+    assert store is service.get_store()
+  
+  def test_i_can_resolve_correct_store_among_multiple_services(self, dsm):
+    """_resolve_store_for_table() identifies the right store when multiple services are registered.
+  
+    The resolver iterates over all registered services and must return the store
+    whose service has a matching table_name, not another service's store.
+    """
+    svc_a = dsm.create_service("ns.table_a", save_state=False)
+    svc_b = dsm.create_service("ns.table_b", save_state=False)
+    
+    df = pd.DataFrame({"x": [10, 20]})
+    svc_a.load_dataframe(df)
+    svc_b.load_dataframe(df.copy())
+    
+    store_a = dsm._resolve_store_for_table("ns.table_a")
+    store_b = dsm._resolve_store_for_table("ns.table_b")
+    
+    assert store_a is svc_a.get_store()
+    assert store_b is svc_b.get_store()
+    assert store_a is not store_b
+  
+  def test_i_cannot_resolve_unknown_table(self, dsm):
+    """FormulaEngine resolver returns None for an unknown table name."""
+    result = dsm._resolve_store_for_table("unknown.table")
+    assert result is None

tests/core/test_datagrid_registry.py

@@ -2,9 +2,7 @@ import shutil
 
 import pytest
 from dbengine.handlers import handlers
-from pandas import DataFrame
 
-from myfasthtml.controls.DataGrid import DataGrid, DatagridConf
 from myfasthtml.core.DataGridsRegistry import DataGridsRegistry, DATAGRIDS_REGISTRY_ENTRY_KEY
 from myfasthtml.core.dbengine_utils import DataFrameHandler
 from myfasthtml.core.dbmanager import DbManager
@@ -31,7 +29,6 @@ def session():
 @pytest.fixture
 def parent(session):
   instance = SingleInstance(session=session, _id="test_parent_id")
-  instance.get_formula_engine = lambda: None
   return instance
 
 
@@ -46,69 +43,53 @@ def db_manager(parent):
   InstancesManager.reset()
 
 
-@pytest.fixture
-def dg(parent):
-  # the table must be created
-  data = {"name": ["john", "jane"], "id": [1, 2]}
-  df = DataFrame(data)
-  dgc = DatagridConf("namespace", "table_name")
-  datagrid = DataGrid(parent, conf=dgc, save_state=True)
-  datagrid.init_from_dataframe(df, init_state=True)
-  yield datagrid
-  
-  datagrid.dispose()
-
-
 @pytest.fixture
 def dgr(parent, db_manager):
   return DataGridsRegistry(parent)
 
 
-def test_entry_is_created_at_startup(db_manager, dgr, ):
+def test_i_can_create_registry_with_empty_state(db_manager, dgr):
+  """Registry is initialised with an empty dict in DB."""
   assert db_manager.exists_entry(DATAGRIDS_REGISTRY_ENTRY_KEY)
   assert clean_db_object(db_manager.load(DATAGRIDS_REGISTRY_ENTRY_KEY)) == {}
 
 
-def test_i_can_put_a_table_in_registry(dgr):
+def test_i_can_put_and_retrieve_entries(dgr):
+  """put() persists entries retrievable via get_all_entries()."""
   dgr.put("namespace", "name", "datagrid_id")
   dgr.put("namespace2", "name2", "datagrid_id2")
-  assert dgr.get_all_tables() == ["namespace.name", "namespace2.name2"]
+
+  entries = dgr.get_all_entries()
+  assert "datagrid_id" in entries
+  assert "datagrid_id2" in entries
+  assert entries["datagrid_id"] == ("namespace", "name")
+  assert entries["datagrid_id2"] == ("namespace2", "name2")
 
 
-def test_i_can_columns_names_for_a_table(dgr, dg):
-  expected = ["__row_index__", "name", "id"] if dg.get_state().row_index else ["name", "id"]
-  namespace, name = dg.get_settings().namespace, dg.get_settings().name
-  dgr.put(namespace, name, dg.get_id())
+def test_i_can_remove_an_entry(dgr):
+  """remove() deletes the entry from the registry."""
+  dgr.put("ns", "tbl", "grid_1")
+  assert "grid_1" in dgr.get_all_entries()
 
-  table_full_name = f"{namespace}.{name}"
-  assert dgr.get_columns(table_full_name) == expected
+  dgr.remove("grid_1")
+  assert "grid_1" not in dgr.get_all_entries()
 
 
-def test_i_can_get_columns_values(dgr, dg):
-  namespace, name = dg.get_settings().namespace, dg.get_settings().name
-  dgr.put(namespace, name, dg.get_id())
-  
-  table_full_name = f"{namespace}.{name}"
-  assert dgr.get_column_values(table_full_name, "name") == ["john", "jane"]
+def test_i_can_remove_nonexistent_entry_without_error(dgr):
+  """remove() on a missing id does not raise."""
+  dgr.remove("does_not_exist")  # should not raise
 
 
-def test_i_can_get_row_count(dgr, dg):
-  namespace, name = dg.get_settings().namespace, dg.get_settings().name
-  dgr.put(namespace, name, dg.get_id())
+def test_i_can_put_multiple_entries_and_get_all(dgr):
+  """get_all_entries() returns all registered grids."""
+  dgr.put("ns1", "t1", "id1")
+  dgr.put("ns2", "t2", "id2")
+  dgr.put("ns3", "t3", "id3")
 
-  table_full_name = f"{namespace}.{name}"
-  assert dgr.get_row_count(table_full_name) == 2
+  entries = dgr.get_all_entries()
+  assert len(entries) == 3
 
 
-def test_i_can_manage_when_table_name_does_not_exist(dgr):
-  assert dgr.get_columns("namespace.name") == []
-  assert dgr.get_row_count("namespace.name") == 0
-
-
-def test_i_can_manage_when_column_does_not_exist(dgr, dg):
-  namespace, name = dg.get_settings().namespace, dg.get_settings().name
-  dgr.put(namespace, name, dg.get_id())
-  
-  table_full_name = f"{namespace}.{name}"
-  assert len(dgr.get_columns(table_full_name)) > 0
-  assert dgr.get_column_values("namespace.name", "") == []
+def test_i_can_get_empty_entries_when_registry_is_empty(dgr):
+  """get_all_entries() returns empty dict when nothing registered."""
+  assert dgr.get_all_entries() == {}