subsurface-terra 2025.1.0rc15__py3-none-any.whl → 2025.1.0rc16__py3-none-any.whl

subsurface/modules/reader/volume/read_volume.py

@@ -1,230 +1,327 @@
-import os
-from io import BytesIO
-from typing import Union
-
-from subsurface.core.structs import StructuredData
-
-from .... import optional_requirements
-from ....core.structs import UnstructuredData
-from subsurface.core.reader_helpers.readers_data import GenericReaderFilesHelper
-import numpy as np
-import pandas as pd
-
-
-def read_VTK_structured_grid(file_or_buffer: Union[str, BytesIO], active_scalars: str) -> StructuredData:
-    pv = optional_requirements.require_pyvista()
-
-    if isinstance(file_or_buffer, BytesIO):
-        # If file_or_buffer is a BytesIO, write it to a temporary file
-        from tempfile import NamedTemporaryFile
-        with NamedTemporaryFile('wb', suffix='.vtk', delete=False) as temp_file:
-            # Write the BytesIO content to the temporary file
-            getvalue: bytes = file_or_buffer.getvalue()
-            temp_file.write(getvalue)
-            temp_file.flush()  # Make sure all data is written
-            temp_file_name = temp_file.name  # Store the temporary file name
-        try:
-            # Use pyvista.read() to read from the temporary file
-            pyvista_obj = pv.read(temp_file_name)
-        finally:
-            # Ensure the temporary file is deleted after reading
-            os.remove(temp_file_name)
-    else:
-        # If it's a file path, read directly
-        pyvista_obj = pv.read(file_or_buffer)
-    try:
-        pyvista_struct: pv.ExplicitStructuredGrid = pv_cast_to_explicit_structured_grid(pyvista_obj)
-    except Exception as e:
-        raise f"The file is not a structured grid: {e}"
-
-    if PLOT := False:
-        pyvista_struct.set_active_scalars(active_scalars)
-        pyvista_struct.plot()
-
-    struct: StructuredData = StructuredData.from_pyvista_structured_grid(
-        grid=pyvista_struct,
-        data_array_name=active_scalars
-    )
-
-    return struct
-
-
-def read_volumetric_mesh_to_subsurface(reader_helper_coord: GenericReaderFilesHelper,
-                                       reader_helper_attr: GenericReaderFilesHelper) -> UnstructuredData:
-    df_coord = read_volumetric_mesh_coord_file(reader_helper_coord)
-    if len(df_coord.columns) == 1:
-        raise ValueError(
-            "The attributes file has only one column, probably the columns are not being separated correctly. Use 'sep' in Additional Reader Arguments"
-        )
-
-    df_attr = read_volumetric_mesh_attr_file(reader_helper_attr)
-    # Check if there are more than one column and if it is only one raise an error that probably the columns have not been properly separated. Use "sep" in Additional Reader Arguments
-    if len(df_attr.columns) == 1:
-        raise ValueError(
-            "The attributes file has only one column, probably the columns are not being separated correctly. Use 'sep' in Additional Reader Arguments"
-        )
-    
-    combined_df = df_coord.merge(df_attr, left_index=True, right_index=True)
-    ud = UnstructuredData.from_array(
-        vertex=combined_df[['x', 'y', 'z']], cells="points",
-        attributes=combined_df[['pres', 'temp', 'sg', 'xco2']]
-    )
-    return ud
-
-
-def read_volumetric_mesh_coord_file(reader_helper: GenericReaderFilesHelper) -> pd.DataFrame:
-    df = pd.read_csv(
-        filepath_or_buffer=reader_helper.file_or_buffer,
-        **reader_helper.pandas_reader_kwargs
-    )
-    if reader_helper.columns_map is not None:
-        df.rename(
-            mapper=reader_helper.columns_map,
-            axis="columns",
-            inplace=True
-        )
-
-    df.dropna(axis=0, inplace=True)
-
-    df.x = df.x.astype(float)
-    df.y = df.y.astype(float)
-    df.z = df.z.astype(float)
-    # Throw error if empty
-    if df.empty:
-        raise ValueError("The file is empty")
-    
-    return df
-
-
-def read_volumetric_mesh_attr_file(reader_helper: GenericReaderFilesHelper) -> pd.DataFrame:
-    df = pd.read_table(reader_helper.file_or_buffer, **reader_helper.pandas_reader_kwargs)
-    df.columns = df.columns.astype(str).str.strip()
-    return df
-
-
-def pv_cast_to_explicit_structured_grid(pyvista_object: 'pv.DataSet') -> 'pv.ExplicitStructuredGrid':
-    pv = optional_requirements.require_pyvista()
-
-    match pyvista_object:
-        case pv.RectilinearGrid() as rectl_grid:
-            return __pv_convert_rectilinear_to_explicit(rectl_grid)
-        case pv.UnstructuredGrid() as unstr_grid:
-            return __pv_convert_unstructured_to_explicit(unstr_grid)
-        case _:
-            return pyvista_object.cast_to_explicit_structured_grid()
-
-
-def __pv_convert_unstructured_to_explicit(unstr_grid):
-    """
-    Convert a PyVista UnstructuredGrid to an ExplicitStructuredGrid if possible.
-    """
-    pv = optional_requirements.require_pyvista()
-
-    # First check if the grid has the necessary attributes to be treated as structured
-    if not hasattr(unstr_grid, 'n_cells') or unstr_grid.n_cells == 0:
-        raise ValueError("The unstructured grid has no cells.")
-    
-    # Try to detect if the grid has a structured topology
-    # Check if the grid has cell type 11 (VTK_VOXEL) or 12 (VTK_HEXAHEDRON)
-    cell_types = unstr_grid.celltypes
-    
-    # Voxels (11) and hexahedra (12) are the cell types used in structured grids
-    if not all(ct in [11, 12] for ct in cell_types):
-        raise ValueError("The unstructured grid contains non-hexahedral cells and cannot be converted to explicit structured.")
-    
-    # Try to infer dimensions from the grid
-    try:
-        # Method 1: Try PyVista's built-in conversion if available
-        return unstr_grid.cast_to_explicit_structured_grid()
-    except (AttributeError, TypeError):
-        pass
-    
-    try:
-        # Method 2: If the grid has dimensions stored as field data
-        if "dimensions" in unstr_grid.field_data:
-            dims = unstr_grid.field_data["dimensions"]
-            if len(dims) == 3:
-                nx, ny, nz = dims
-                # Verify that dimensions match the number of cells
-                if (nx-1)*(ny-1)*(nz-1) != unstr_grid.n_cells:
-                    raise ValueError("Stored dimensions do not match the number of cells.")
-                
-                # Extract points and reorder if needed
-                points = unstr_grid.points.reshape((nx, ny, nz, 3))
-                
-                # Create explicit structured grid
-                explicit_grid = pv.ExplicitStructuredGrid((nx, ny, nz), points.reshape((-1, 3)))
-                explicit_grid.compute_connectivity()
-                
-                # Transfer data arrays
-                for name, array in unstr_grid.cell_data.items():
-                    explicit_grid.cell_data[name] = array.copy()
-                for name, array in unstr_grid.point_data.items():
-                    explicit_grid.point_data[name] = array.copy()
-                for name, array in unstr_grid.field_data.items():
-                    if name != "dimensions":  # Skip dimensions field
-                        explicit_grid.field_data[name] = array.copy()
-                
-                return explicit_grid
-    except (ValueError, KeyError):
-        pass
-    
-    # If none of the above methods work, use PyVista's extract_cells function
-    # to reconstruct the structured grid if possible
-    try:
-        # This is a best-effort approach that tries multiple strategies
-        return pv.core.filters.convert_unstructured_to_structured_grid(unstr_grid)
-    except Exception as e:
-        raise ValueError(f"Failed to convert unstructured grid to explicit structured grid: {e}")
-    
-def __pv_convert_rectilinear_to_explicit(rectl_grid):
-
-    pv = optional_requirements.require_pyvista()
-
-    # Extract the coordinate arrays from the input RectilinearGrid.
-    x = np.asarray(rectl_grid.x)
-    y = np.asarray(rectl_grid.y)
-    z = np.asarray(rectl_grid.z)
-
-    # Helper function: "double" the coordinates to produce an expanded set
-    # that, when processed internally via np.unique, returns the original nodal values.
-    def doubled_coords(arr):
-        return np.repeat(arr, 2)[1:-1]
-
-    # Double the coordinate arrays.
-    xcorn = doubled_coords(x)
-    ycorn = doubled_coords(y)
-    zcorn = doubled_coords(z)
-
-    # Build a complete grid of corner points via meshgrid. Fortran ('F') order ensures
-    # the connectivity ordering aligns with VTK's expectations.
-    xx, yy, zz = np.meshgrid(xcorn, ycorn, zcorn, indexing='ij')
-    corners = np.column_stack((xx.ravel(order='F'),
-                               yy.ravel(order='F'),
-                               zz.ravel(order='F')))
-
-    # The dimensions to pass to the ExplicitStructuredGrid constructor should be
-    # the counts of unique coordinates in each direction.
-    dims = (len(np.unique(xcorn)),
-            len(np.unique(ycorn)),
-            len(np.unique(zcorn)))
-
-    # Create the ExplicitStructuredGrid.
-    explicit_grid = pv.ExplicitStructuredGrid(dims, corners)
-    explicit_grid.compute_connectivity()
-
-    # --- Copy associated data arrays ---
-
-    # Transfer all cell data arrays.
-    for name, array in rectl_grid.cell_data.items():
-        explicit_grid.cell_data[name] = array.copy()
-
-    # Transfer all point data arrays.
-    for name, array in rectl_grid.point_data.items():
-        explicit_grid.point_data[name] = array.copy()
-
-    # (Optional) Transfer field data as well.
-    for name, array in rectl_grid.field_data.items():
-        explicit_grid.field_data[name] = array.copy()
-
-    return explicit_grid
+import os
+from io import BytesIO
+from typing import Union
+
+from subsurface.core.structs import StructuredData
+
+from .... import optional_requirements
+from ....core.structs import UnstructuredData
+from subsurface.core.reader_helpers.readers_data import GenericReaderFilesHelper
+import numpy as np
+import pandas as pd
+
+
+def read_VTK_structured_grid(file_or_buffer: Union[str, BytesIO], active_scalars: str) -> StructuredData:
+    pv = optional_requirements.require_pyvista()
+
+    if isinstance(file_or_buffer, BytesIO):
+        # If file_or_buffer is a BytesIO, write it to a temporary file
+        from tempfile import NamedTemporaryFile
+        with NamedTemporaryFile('wb', suffix='.vtk', delete=False) as temp_file:
+            # Write the BytesIO content to the temporary file
+            getvalue: bytes = file_or_buffer.getvalue()
+            temp_file.write(getvalue)
+            temp_file.flush()  # Make sure all data is written
+            temp_file_name = temp_file.name  # Store the temporary file name
+        try:
+            # Use pyvista.read() to read from the temporary file
+            pyvista_obj = pv.read(temp_file_name)
+        finally:
+            # Ensure the temporary file is deleted after reading
+            os.remove(temp_file_name)
+    else:
+        # If it's a file path, read directly
+        pyvista_obj = pv.read(file_or_buffer)
+    try:
+        pyvista_struct: pv.ExplicitStructuredGrid = pv_cast_to_explicit_structured_grid(pyvista_obj)
+    except Exception as e:
+        raise f"The file is not a structured grid: {e}"
+
+    if PLOT := False:
+        pyvista_struct.set_active_scalars(active_scalars)
+        pyvista_struct.plot()
+
+    struct: StructuredData = StructuredData.from_pyvista_structured_grid(
+        grid=pyvista_struct,
+        data_array_name=active_scalars
+    )
+
+    return struct
+
+
+def read_volumetric_mesh_to_subsurface(reader_helper_coord: GenericReaderFilesHelper,
+                                       reader_helper_attr: GenericReaderFilesHelper) -> UnstructuredData:
+    df_coord = read_volumetric_mesh_coord_file(reader_helper_coord)
+    if len(df_coord.columns) == 1:
+        raise ValueError(
+            "The attributes file has only one column, probably the columns are not being separated correctly. Use 'sep' in Additional Reader Arguments"
+        )
+
+    df_attr = read_volumetric_mesh_attr_file(reader_helper_attr)
+    # Check if there are more than one column and if it is only one raise an error that probably the columns have not been properly separated. Use "sep" in Additional Reader Arguments
+    if len(df_attr.columns) == 1:
+        raise ValueError(
+            "The attributes file has only one column, probably the columns are not being separated correctly. Use 'sep' in Additional Reader Arguments"
+        )
+    
+    combined_df = df_coord.merge(df_attr, left_index=True, right_index=True)
+    ud = UnstructuredData.from_array(
+        vertex=combined_df[['x', 'y', 'z']], cells="points",
+        attributes=combined_df[['pres', 'temp', 'sg', 'xco2']]
+    )
+    return ud
+
+
+def read_volumetric_mesh_coord_file(reader_helper: GenericReaderFilesHelper) -> pd.DataFrame:
+    df = pd.read_csv(
+        filepath_or_buffer=reader_helper.file_or_buffer,
+        **reader_helper.pandas_reader_kwargs
+    )
+    if reader_helper.columns_map is not None:
+        df.rename(
+            mapper=reader_helper.columns_map,
+            axis="columns",
+            inplace=True
+        )
+
+    df.dropna(axis=0, inplace=True)
+
+    df.x = df.x.astype(float)
+    df.y = df.y.astype(float)
+    df.z = df.z.astype(float)
+    # Throw error if empty
+    if df.empty:
+        raise ValueError("The file is empty")
+    
+    return df
+
+
+def read_volumetric_mesh_attr_file(reader_helper: GenericReaderFilesHelper) -> pd.DataFrame:
+    df = pd.read_table(reader_helper.file_or_buffer, **reader_helper.pandas_reader_kwargs)
+    df.columns = df.columns.astype(str).str.strip()
+    return df
+
+
+def pv_cast_to_explicit_structured_grid(pyvista_object: 'pv.DataSet') -> 'pv.ExplicitStructuredGrid':
+    pv = optional_requirements.require_pyvista()
+
+    match pyvista_object:
+        case pv.RectilinearGrid() as rectl_grid:
+            return __pv_convert_rectilinear_to_explicit(rectl_grid)
+        case pv.UnstructuredGrid() as unstr_grid:
+            return __pv_convert_unstructured_to_explicit(unstr_grid)
+        case _:
+            return pyvista_object.cast_to_explicit_structured_grid()
+
+
+def __pv_convert_unstructured_to_explicit(unstr_grid):
+    """
+    Convert a PyVista UnstructuredGrid to an ExplicitStructuredGrid if possible.
+    """
+    pv = optional_requirements.require_pyvista()
+
+    # First check if the grid has the necessary attributes to be treated as structured
+    if not hasattr(unstr_grid, 'n_cells') or unstr_grid.n_cells == 0:
+        raise ValueError("The unstructured grid has no cells.")
+    
+    # Try to detect if the grid has a structured topology
+    # Check if the grid has cell type 11 (VTK_VOXEL) or 12 (VTK_HEXAHEDRON)
+    cell_types = unstr_grid.celltypes
+    
+    # Voxels (11) and hexahedra (12) are the cell types used in structured grids
+    if not all(ct in [11, 12] for ct in cell_types):
+        raise ValueError("The unstructured grid contains non-hexahedral cells and cannot be converted to explicit structured.")
+    
+    # Try to infer dimensions from the grid
+    try:
+        # Method 1: Try PyVista's built-in conversion if available
+        return unstr_grid.cast_to_explicit_structured_grid()
+    except (AttributeError, TypeError):
+        pass
+    
+    try:
+        # Method 2: If the grid has dimensions stored as field data
+        if "dimensions" in unstr_grid.field_data:
+            dims = unstr_grid.field_data["dimensions"]
+            if len(dims) == 3:
+                nx, ny, nz = dims
+                # Verify that dimensions match the number of cells
+                if (nx-1)*(ny-1)*(nz-1) != unstr_grid.n_cells:
+                    raise ValueError("Stored dimensions do not match the number of cells.")
+                
+                # Extract points and reorder if needed
+                points = unstr_grid.points.reshape((nx, ny, nz, 3))
+                
+                # Create explicit structured grid
+                explicit_grid = pv.ExplicitStructuredGrid((nx, ny, nz), points.reshape((-1, 3)))
+                explicit_grid.compute_connectivity()
+                
+                # Transfer data arrays
+                for name, array in unstr_grid.cell_data.items():
+                    explicit_grid.cell_data[name] = array.copy()
+                for name, array in unstr_grid.point_data.items():
+                    explicit_grid.point_data[name] = array.copy()
+                for name, array in unstr_grid.field_data.items():
+                    if name != "dimensions":  # Skip dimensions field
+                        explicit_grid.field_data[name] = array.copy()
+                
+                return explicit_grid
+    except (ValueError, KeyError):
+        pass
+    
+    # If none of the above methods work, use PyVista's extract_cells function
+    # to reconstruct the structured grid if possible
+    try:
+        # This is a best-effort approach that tries multiple strategies
+        return pv.core.filters.convert_unstructured_to_structured_grid(unstr_grid)
+    except Exception as e:
+        raise ValueError(f"Failed to convert unstructured grid to explicit structured grid: {e}")
+
+
+def __pv_convert_rectilinear_to_explicit(rectl_grid, *, temp_dtype=None):
+    """
+    Convert a PyVista RectilinearGrid to an ExplicitStructuredGrid with low peak memory.
+
+    Behavior:
+      - Output points are in world coordinates, dtype matches rectl_grid.points.dtype.
+      - Data arrays are shallow-transferred (no deep copies).
+      - temp_dtype controls the large temporary `corners` buffer dtype.
+        * If temp_dtype is None (default), use float32 when the output dtype is wider (e.g., float64),
+          else use the output dtype. This reduces peak memory automatically.
+        * If temp_dtype < output dtype, coordinates are recentred for precision and origin is added back after.
+
+    Parameters
+    ----------
+    rectl_grid : pv.RectilinearGrid
+    temp_dtype : numpy dtype or None
+        Dtype for building the temporary `corners` array. Examples:
+          - None (default): auto -> float32 if output dtype is wider, else output dtype.
+          - np.float32: memory-friendly; auto recenters & restores origin.
+          - np.float64: highest precision (more memory).
+
+    Returns
+    -------
+    pv.ExplicitStructuredGrid
+    """
+    import numpy as np
+    pv = optional_requirements.require_pyvista()
+
+    # Output dtype follows source grid points (usually float64)
+    out_dtype = getattr(rectl_grid.points, "dtype", np.float64)
+
+    # Auto-pick temp dtype: prefer float32 when output is wider (e.g., float64)
+    if temp_dtype is None:
+        temp_dtype = np.float32 if (
+                    np.dtype(out_dtype).kind == 'f' and np.dtype(out_dtype).itemsize > 4) else out_dtype
+
+    # Coordinate arrays
+    x = np.asarray(rectl_grid.x)
+    y = np.asarray(rectl_grid.y)
+    z = np.asarray(rectl_grid.z)
+
+    # Decide if we must recenter (when temp dtype is lower precision than output dtype)
+    def _is_lower_precision(src, dst):
+        s, d = np.dtype(src), np.dtype(dst)
+        if s.kind != 'f' or d.kind != 'f':
+            return s != d
+        return s.itemsize < d.itemsize
+
+    if _is_lower_precision(temp_dtype, out_dtype):
+        origin = np.array([x[0], y[0], z[0]], dtype=np.float64)
+        x_base, y_base, z_base = x - origin[0], y - origin[1], z - origin[2]
+    else:
+        origin = None
+        x_base, y_base, z_base = x, y, z
+
+    # Double coordinates (interior duplication expected by ExplicitStructuredGrid ctor)
+    def _doubled(arr):
+        # [a,b,c,d] -> [a, b,b, c,c, d]
+        return np.repeat(arr, 2)[1:-1]
+
+    xcorn = _doubled(x_base)
+    ycorn = _doubled(y_base)
+    zcorn = _doubled(z_base)
+
+    nx2, ny2, nz2 = len(xcorn), len(ycorn), len(zcorn)
+    slab = ny2 * nz2
+    N = nx2 * slab
+
+    # Build corners via slab/chunked fill (avoids N-sized intermediates)
+    yz = np.empty((slab, 2), dtype=temp_dtype)
+    yz[:, 0] = np.repeat(ycorn, nz2).astype(temp_dtype, copy=False)  # Y pattern
+    yz[:, 1] = np.tile(zcorn, ny2).astype(temp_dtype, copy=False)  # Z pattern
+
+    corners = np.empty((N, 3), dtype=temp_dtype)
+    for i, xv in enumerate(xcorn):
+        start = i * slab
+        end = start + slab
+        corners[start:end, 0] = xv
+        corners[start:end, 1:3] = yz
+
+    # Construct explicit grid
+    dims = (len(x), len(y), len(z))
+    explicit = pv.ExplicitStructuredGrid(dims, corners)
+    explicit.compute_connectivity()
+
+    # Always return world coordinates; add origin back and cast to out_dtype in one fused pass
+    if origin is not None:
+        new_pts = np.empty_like(explicit.points, dtype=out_dtype)
+        np.add(explicit.points, origin, out=new_pts, dtype=out_dtype)
+        explicit.points = new_pts
+    else:
+        if explicit.points.dtype != out_dtype:
+            explicit.points = explicit.points.astype(out_dtype, copy=False)
+
+    # Shallow-transfer all data arrays (no deep copies)
+    for name, arr in rectl_grid.cell_data.items():
+        explicit.cell_data[name] = arr
+    for name, arr in rectl_grid.point_data.items():
+        explicit.point_data[name] = arr
+    for name, arr in rectl_grid.field_data.items():
+        explicit.field_data[name] = arr
+
+    __validate_rectilinear_to_explicit_conversion(rectl_grid, explicit)
+
+    return explicit
+
+
+def __validate_rectilinear_to_explicit_conversion(rectl_grid, explicit_grid, *, atol=1e-6, rtol=1e-8) -> None:
+    """
+    Validate core equivalence between a RectilinearGrid and its ExplicitStructuredGrid.
+    Raises ValueError on mismatch. Avoids large 3D uniques / big temporaries.
+    """
+    import numpy as np
+
+    # dims & counts
+    nx, ny, nz = map(int, rectl_grid.dimensions)
+    if tuple(map(int, explicit_grid.dimensions)) != (nx, ny, nz):
+        raise ValueError(f"Dimensions differ: explicit {tuple(explicit_grid.dimensions)} vs rect {tuple(rectl_grid.dimensions)}")
+
+    expected_cells = (nx - 1) * (ny - 1) * (nz - 1)
+    if explicit_grid.n_cells != expected_cells:
+        raise ValueError(f"Cell count mismatch: explicit {explicit_grid.n_cells} vs expected {expected_cells}")
+
+    expected_points = nx * ny * nz
+    if explicit_grid.n_points != expected_points:
+        raise ValueError(f"Point count mismatch: explicit {explicit_grid.n_points} vs expected {expected_points}")
+
+    # bounds
+    if not np.allclose(explicit_grid.bounds, rectl_grid.bounds, rtol=rtol, atol=atol):
+        raise ValueError(f"Bounds differ: explicit {explicit_grid.bounds} vs rect {rectl_grid.bounds}")
+
+    # axis coordinates (order-independent)
+    pts = explicit_grid.points  # shape (M,3), M = nx*ny*nz (already deduped/sorted by unique)
+    x_exp = np.unique(pts[:, 0])
+    y_exp = np.unique(pts[:, 1])
+    z_exp = np.unique(pts[:, 2])
+
+    x_rect = np.asarray(rectl_grid.x)
+    y_rect = np.asarray(rectl_grid.y)
+    z_rect = np.asarray(rectl_grid.z)
+
+    if len(x_exp) != len(x_rect) or not np.allclose(x_exp, x_rect, rtol=rtol, atol=atol):
+        raise ValueError("X axis coordinates differ.")
+    if len(y_exp) != len(y_rect) or not np.allclose(y_exp, y_rect, rtol=rtol, atol=atol):
+        raise ValueError("Y axis coordinates differ.")
+    if len(z_exp) != len(z_rect) or not np.allclose(z_exp, z_rect, rtol=rtol, atol=atol):
+        raise ValueError("Z axis coordinates differ.")