Skip to content

_arrays

Attributes

PyObject = Union[dict, list, str, float, int, np.dtype, None, Any, pd.DataFrame, pd.Series, np.ndarray, az.InferenceData, xr.DataArray, xr.Dataset] module-attribute

Union of common Python-side objects produced by R→Python conversion.

This is intentionally broad: brmspy frequently returns standard scientific Python types (NumPy/pandas/xarray/ArviZ), plus plain dict/list primitives.

Note

Avoid adding Any here unless absolutely necessary; it defeats the purpose of having this alias.

Classes

ShmPool

Minimal interface for allocating and attaching shared-memory blocks.

The concrete implementation lives in brmspy._session.transport.ShmPool and tracks blocks so they can be closed on teardown.

Source code in brmspy/types/shm.py 
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
class ShmPool:
    """
    Minimal interface for allocating and attaching shared-memory blocks.

    The concrete implementation lives in
    [`brmspy._session.transport.ShmPool`][brmspy._session.transport.ShmPool] and tracks
    blocks so they can be closed on teardown.
    """

    def __init__(self, manager: SharedMemoryManager) -> None:
        """
        Create a pool bound to an existing `SharedMemoryManager`.

        Parameters
        ----------
        manager : multiprocessing.managers.SharedMemoryManager
            Manager used to allocate blocks.
        """
        ...

    def alloc(self, size: int, temporary: bool = False) -> ShmBlock:
        """
        Allocate a new shared-memory block.

        Parameters
        ----------
        size : int
            Size in bytes.

        Returns
        -------
        ShmBlock
            Newly allocated block.
        """
        ...

    def attach(self, ref: ShmRef) -> ShmBlock:
        """
        Attach to an existing shared-memory block by name.

        Returns
        -------
        ShmBlock
            Attached block.
        """
        ...

    def close_all(self) -> None:
        """
        Close all tracked shared-memory handles owned by this pool.

        Returns
        -------
        None
        """
        ...

    def gc(self, name: str | None = None) -> None: ...

Functions

__init__(manager)

Create a pool bound to an existing SharedMemoryManager.

Parameters:

Name Type Description Default
manager SharedMemoryManager

Manager used to allocate blocks.

 required
Source code in brmspy/types/shm.py 
87
88
89
90
91
92
93
94
95
96
def __init__(self, manager: SharedMemoryManager) -> None:
    """
    Create a pool bound to an existing `SharedMemoryManager`.

    Parameters
    ----------
    manager : multiprocessing.managers.SharedMemoryManager
        Manager used to allocate blocks.
    """
    ...
alloc(size, temporary=False)

Allocate a new shared-memory block.

Parameters:

Name Type Description Default
size int

Size in bytes.

 required

Returns:

Type Description
ShmBlock

Newly allocated block.
Source code in brmspy/types/shm.py 
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
def alloc(self, size: int, temporary: bool = False) -> ShmBlock:
    """
    Allocate a new shared-memory block.

    Parameters
    ----------
    size : int
        Size in bytes.

    Returns
    -------
    ShmBlock
        Newly allocated block.
    """
    ...
attach(ref)

Attach to an existing shared-memory block by name.

Returns:

Type Description
ShmBlock

Attached block.
Source code in brmspy/types/shm.py 
114
115
116
117
118
119
120
121
122
123
def attach(self, ref: ShmRef) -> ShmBlock:
    """
    Attach to an existing shared-memory block by name.

    Returns
    -------
    ShmBlock
        Attached block.
    """
    ...
close_all()

Close all tracked shared-memory handles owned by this pool.

Returns:

Type Description
None
Source code in brmspy/types/shm.py 
125
126
127
128
129
130
131
132
133
def close_all(self) -> None:
    """
    Close all tracked shared-memory handles owned by this pool.

    Returns
    -------
    None
    """
    ...
gc(name=None)
Source code in brmspy/types/shm.py 
135
def gc(self, name: str | None = None) -> None: ...

ShmArray

Bases: ndarray

NumPy array view backed by a shared-memory block.

Attributes:

Name Type Description
block ShmRef

Reference to the shared-memory block backing the array data.
Notes

This is a view over SharedMemory.buf. Closing/unlinking the underlying shared memory while the array is still in use will lead to undefined behavior.

Source code in brmspy/types/shm_extensions.py 
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
class ShmArray(np.ndarray):
    """
    NumPy array view backed by a shared-memory block.

    Attributes
    ----------
    block : ShmRef
        Reference to the shared-memory block backing the array data.

    Notes
    -----
    This is a *view* over `SharedMemory.buf`. Closing/unlinking the underlying
    shared memory while the array is still in use will lead to undefined
    behavior.
    """

    _shm_metadata: ShmRef  # for type checkers

    @classmethod
    def from_metadata(
        cls, meta: ShmArrayMetadata | dict[str, Any], block: ShmBlock
    ) -> np.ndarray:
        dtype = np.dtype(meta["dtype"])
        shape = tuple(meta["shape"])
        order = meta["order"]

        return ShmArray.from_block(block=block, shape=shape, dtype=dtype, order=order)

    @classmethod
    def from_block(
        cls, block: ShmBlock, shape: tuple[int, ...], dtype: np.dtype, **kwargs
    ) -> Union["ShmArray", np.ndarray]:
        """
        Create an array view backed by an existing shared-memory block.

        Parameters
        ----------
        block : ShmBlock
            Attached shared-memory block.
        shape : tuple[int, ...]
            Desired array shape.
        dtype : numpy.dtype
            NumPy dtype of the array.
        **kwargs
            Reserved for future compatibility. Currently unused.

        Returns
        -------
        ShmArray
            Array view into the shared-memory buffer.
        """
        is_object = np.dtype(dtype) == np.dtype("O")

        if not is_object:
            if block.shm.buf:
                view = memoryview(block.shm.buf)
                view = view[: block.content_size]
            else:
                view = None
            base = np.ndarray(
                shape=shape,
                dtype=dtype,
                buffer=view,
                order=kwargs.get("order", "F"),
            )
            obj = base.view(ShmArray)
            obj._shm_metadata = block.to_ref()
        else:
            assert block.shm.buf
            view = memoryview(block.shm.buf)
            view = view[: block.content_size]
            payload = bytes(view)
            obj = pickle.loads(payload)
            assert isinstance(obj, np.ndarray)

        return obj

    @classmethod
    def array_order(cls, a: np.ndarray) -> Literal["C", "F", "non-contiguous"]:
        """
        Determine how an array can be reconstructed from a raw buffer.

        Returns `"C"` for C-contiguous arrays, `"F"` for Fortran-contiguous arrays,
        otherwise `"non-contiguous"` (meaning: bytes were obtained by forcing
        a contiguous copy during encoding).
        """
        if a.flags["C_CONTIGUOUS"]:
            return "C"
        if a.flags["F_CONTIGUOUS"]:
            return "F"
        return "non-contiguous"

    @classmethod
    def is_string_object(cls, a: np.ndarray, sample: int = 1000):
        if np.dtype(a.dtype) != np.dtype("O"):
            return False
        it = a.flat
        for _ in range(min(sample, a.size)):
            v = next(it, None)
            if v is not None and not isinstance(v, str):
                return False
        return True

    @classmethod
    def to_shm(
        cls, obj: np.ndarray | pd.Series | list, shm_pool: Any
    ) -> tuple[np.ndarray | None, ShmRef, str, list[int], str]:
        if isinstance(obj, pd.Series):
            if isinstance(obj.dtype, pd.CategoricalDtype):
                arr = obj.cat.codes.to_numpy(copy=False)
            else:
                arr = obj.to_numpy(copy=False)
        elif not isinstance(obj, np.ndarray):
            arr = np.asarray(obj)
        else:
            arr = obj

        is_object = np.dtype(arr.dtype) == np.dtype("O")
        is_string = cls.is_string_object(arr)

        arr_modified = None
        if isinstance(arr, ShmArray):
            ref = arr._shm_metadata

        else:
            temporary = False
            if not is_object:
                data = arr.tobytes(order="C")
            elif is_string:
                arr = arr.astype("U")
                arr_modified = arr
                data = arr.tobytes(order="C")
            else:
                data = pickle.dumps(arr, protocol=pickle.HIGHEST_PROTOCOL)
                temporary = True

            nbytes = len(data)

            # Ask for exactly nbytes; OS may round up internally, that's fine.
            block = shm_pool.alloc(nbytes, temporary=temporary)
            block.shm.buf[:nbytes] = data
            ref = block.to_ref()

        ref, dtype, shape, order = (
            ref,
            str(arr.dtype),
            list(arr.shape),
            cls.array_order(arr),
        )

        return arr_modified, ref, dtype, shape, order

Attributes

_shm_metadata instance-attribute

Functions

from_metadata(meta, block) classmethod
Source code in brmspy/types/shm_extensions.py 
63
64
65
66
67
68
69
70
71
@classmethod
def from_metadata(
    cls, meta: ShmArrayMetadata | dict[str, Any], block: ShmBlock
) -> np.ndarray:
    dtype = np.dtype(meta["dtype"])
    shape = tuple(meta["shape"])
    order = meta["order"]

    return ShmArray.from_block(block=block, shape=shape, dtype=dtype, order=order)
from_block(block, shape, dtype, **kwargs) classmethod

Create an array view backed by an existing shared-memory block.

Parameters:

Name Type Description Default
block ShmBlock

Attached shared-memory block.

 required
shape tuple[int, ...]

Desired array shape.

 required
dtype dtype

NumPy dtype of the array.

 required
**kwargs

Reserved for future compatibility. Currently unused.

 {}

Returns:

Type Description
ShmArray

Array view into the shared-memory buffer.
Source code in brmspy/types/shm_extensions.py 
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
@classmethod
def from_block(
    cls, block: ShmBlock, shape: tuple[int, ...], dtype: np.dtype, **kwargs
) -> Union["ShmArray", np.ndarray]:
    """
    Create an array view backed by an existing shared-memory block.

    Parameters
    ----------
    block : ShmBlock
        Attached shared-memory block.
    shape : tuple[int, ...]
        Desired array shape.
    dtype : numpy.dtype
        NumPy dtype of the array.
    **kwargs
        Reserved for future compatibility. Currently unused.

    Returns
    -------
    ShmArray
        Array view into the shared-memory buffer.
    """
    is_object = np.dtype(dtype) == np.dtype("O")

    if not is_object:
        if block.shm.buf:
            view = memoryview(block.shm.buf)
            view = view[: block.content_size]
        else:
            view = None
        base = np.ndarray(
            shape=shape,
            dtype=dtype,
            buffer=view,
            order=kwargs.get("order", "F"),
        )
        obj = base.view(ShmArray)
        obj._shm_metadata = block.to_ref()
    else:
        assert block.shm.buf
        view = memoryview(block.shm.buf)
        view = view[: block.content_size]
        payload = bytes(view)
        obj = pickle.loads(payload)
        assert isinstance(obj, np.ndarray)

    return obj
array_order(a) classmethod

Determine how an array can be reconstructed from a raw buffer.

Returns "C" for C-contiguous arrays, "F" for Fortran-contiguous arrays, otherwise "non-contiguous" (meaning: bytes were obtained by forcing a contiguous copy during encoding).

Source code in brmspy/types/shm_extensions.py 
122
123
124
125
126
127
128
129
130
131
132
133
134
135
@classmethod
def array_order(cls, a: np.ndarray) -> Literal["C", "F", "non-contiguous"]:
    """
    Determine how an array can be reconstructed from a raw buffer.

    Returns `"C"` for C-contiguous arrays, `"F"` for Fortran-contiguous arrays,
    otherwise `"non-contiguous"` (meaning: bytes were obtained by forcing
    a contiguous copy during encoding).
    """
    if a.flags["C_CONTIGUOUS"]:
        return "C"
    if a.flags["F_CONTIGUOUS"]:
        return "F"
    return "non-contiguous"
is_string_object(a, sample=1000) classmethod
Source code in brmspy/types/shm_extensions.py 
137
138
139
140
141
142
143
144
145
146
@classmethod
def is_string_object(cls, a: np.ndarray, sample: int = 1000):
    if np.dtype(a.dtype) != np.dtype("O"):
        return False
    it = a.flat
    for _ in range(min(sample, a.size)):
        v = next(it, None)
        if v is not None and not isinstance(v, str):
            return False
    return True
to_shm(obj, shm_pool) classmethod
Source code in brmspy/types/shm_extensions.py 
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
@classmethod
def to_shm(
    cls, obj: np.ndarray | pd.Series | list, shm_pool: Any
) -> tuple[np.ndarray | None, ShmRef, str, list[int], str]:
    if isinstance(obj, pd.Series):
        if isinstance(obj.dtype, pd.CategoricalDtype):
            arr = obj.cat.codes.to_numpy(copy=False)
        else:
            arr = obj.to_numpy(copy=False)
    elif not isinstance(obj, np.ndarray):
        arr = np.asarray(obj)
    else:
        arr = obj

    is_object = np.dtype(arr.dtype) == np.dtype("O")
    is_string = cls.is_string_object(arr)

    arr_modified = None
    if isinstance(arr, ShmArray):
        ref = arr._shm_metadata

    else:
        temporary = False
        if not is_object:
            data = arr.tobytes(order="C")
        elif is_string:
            arr = arr.astype("U")
            arr_modified = arr
            data = arr.tobytes(order="C")
        else:
            data = pickle.dumps(arr, protocol=pickle.HIGHEST_PROTOCOL)
            temporary = True

        nbytes = len(data)

        # Ask for exactly nbytes; OS may round up internally, that's fine.
        block = shm_pool.alloc(nbytes, temporary=temporary)
        block.shm.buf[:nbytes] = data
        ref = block.to_ref()

    ref, dtype, shape, order = (
        ref,
        str(arr.dtype),
        list(arr.shape),
        cls.array_order(arr),
    )

    return arr_modified, ref, dtype, shape, order

ShmDataFrameColumns

Bases: DataFrame

pandas DataFrame backed by per-column shared-memory blocks (numeric only).

Attributes:

Name Type Description
_blocks_columns dict[str, PandasColumnMetadata]

Mapping from column name to data required for its reconstruction
Source code in brmspy/types/shm_extensions.py 
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
class ShmDataFrameColumns(pd.DataFrame):
    """
    pandas DataFrame backed by per-column shared-memory blocks (numeric only).

    Attributes
    ----------
    _blocks_columns : dict[str, PandasColumnMetadata]
        Mapping from column name to data required for its reconstruction
    """

    _metadata = ["_shm_metadata"]
    _shm_metadata: dict[str, ShmSeriesMetadata]

    @property
    def _constructor(self):
        # We INTENTIONALLY do not return ShmSeriesMetadata
        # whenever the dataframe is reindexed, slices, we want to get rid of all _shm_metadata,
        # as otherwise we will have immediate problems with buffer alignment
        return pd.DataFrame

    @classmethod
    def _create_col_metadata(
        cls, series: pd.Series, block: ShmRef, arr: np.ndarray | None = None, **params
    ) -> ShmSeriesMetadata:
        pd_dtype = series.dtype

        # Decide what ndarray is actually stored in SHM
        if isinstance(pd_dtype, pd.CategoricalDtype):
            # store dtype params
            params["categories"] = pd_dtype.categories.to_numpy(dtype=object).tolist()
            params["ordered"] = bool(pd_dtype.ordered)
            params["categories_dtype"] = pd_dtype.categories.dtype.name

            # IMPORTANT: store integer codes, not values
            # -1 means missing
            array = series.cat.codes.to_numpy(copy=False)
        elif arr is not None:
            assert isinstance(arr, np.ndarray)
            array = arr
        else:
            # for numeric-only SHM: require a real numpy array output here
            array = series.to_numpy(copy=False)

        # Optional param dtypes you mentioned (only if you support them)
        if isinstance(pd_dtype, pd.PeriodDtype):
            params["freq"] = str(pd_dtype.freq)

        if isinstance(pd_dtype, pd.IntervalDtype):
            params["subtype"] = np.dtype(pd_dtype.subtype).str
            params["closed"] = str(pd_dtype.closed)  # type: ignore[attr-defined]

        meta: ShmSeriesMetadata = {
            "name": series.name,
            "np_dtype": str(array.dtype),
            "pd_dtype": str(pd_dtype.name),
            "block": block,
            "params": params,
        }

        if np.dtype(meta["np_dtype"]) == np.dtype("O"):
            # Sanity check. If this goes wrong, it will be frustrating to debug
            if ShmArray.is_string_object(array, sample=25):
                raise Exception(
                    f"{series.name} column is string, but stored as object!"
                )

        return meta

    def _set_col_raw(self, col: str, value) -> None:
        # bypass our __setitem__
        pd.DataFrame.__setitem__(self, col, value)

    def _set_shm_metadata(self, meta: dict[str, ShmSeriesMetadata]):
        self._shm_metadata = meta

    @classmethod
    def _put_col_in_shm(
        cls, df: "ShmDataFrameColumns", col: str, shm_pool: Any, replace=False
    ):
        vals = df[col].to_numpy(copy=False)
        if (
            isinstance(vals, ShmArray)
            and hasattr(vals, "_shm_metadata")
            and not replace
        ):
            pass
        elif isinstance(vals, np.ndarray):
            if col in df._shm_metadata:
                del df._shm_metadata[col]
            arr_modified, ref, dtype, shape, order = ShmArray.to_shm(df[col], shm_pool)

            if arr_modified is not None:
                # Only needed for string-object normalization; for numeric/codes it's None
                df._set_col_raw(
                    col, pd.Series(arr_modified, index=df.index, name=col, copy=False)
                )
            df._shm_metadata[col] = cls._create_col_metadata(df[col], ref, arr_modified)
            return
        else:
            print(
                f"Failed to update shm metadata for column '{col}' dtype {vals.dtype}"
            )
            return

    @classmethod
    def _reconstruct_series(
        cls,
        meta: ShmSeriesMetadata,
        block: ShmBlock,
        nrows: int,
        index: list | None,
    ) -> pd.Series:
        col_name = meta["name"]
        col_name = str(col_name)
        dtype = np.dtype(meta["np_dtype"])
        pd_dtype = meta["pd_dtype"]
        params = meta["params"]

        arr = ShmArray.from_block(block=block, shape=(nrows,), dtype=dtype, order="C")

        if pd_dtype == "category":
            cats = params.get("categories", None)
            ordered = bool(params.get("ordered", False))

            if cats is None:
                raise ValueError("category dtype requires params['categories']")

            cats_pd_dtype = params.get("categories_dtype")
            if cats_pd_dtype is not None:
                cats_index = pd.Index(cats, dtype=str(cats_pd_dtype))
            else:
                cats_index = pd.Index(cats)

            cat_dtype = pd.CategoricalDtype(categories=cats_index, ordered=ordered)

            # arr should hold integer codes
            # If arr holds codes: build categorical from codes without copying codes.
            # Pandas uses -1 for missing.
            cat = pd.Categorical.from_codes(cast(Sequence[int], arr), dtype=cat_dtype)
            return pd.Series(cat, name=col_name, index=index)

        # 2) tz-aware datetimes
        # Expect arr to be int64 ns timestamps
        if pd_dtype.startswith("datetime64[ns,") or pd_dtype == "datetime64[ns, tz]":
            tz = params.get("tz")
            if not tz:
                # if stored as a plain datetime64[ns] string, just fall through
                pass
            else:
                dt = pd.to_datetime(arr, unit="ns", utc=True).tz_convert(tz)
                return pd.Series(dt, name=col_name, index=index)

        return pd.Series(arr, name=col_name, index=index)

    def __setitem__(self, key, value):
        is_existing = key in self.columns

        super().__setitem__(key, value)

        if is_existing:
            self._on_column_replaced(key)
        else:
            self._on_column_added(key)

    def _on_column_added(self, col: str):
        from brmspy._singleton._shm_singleton import _get_shm

        shm = _get_shm()
        if not shm:
            return
        self._put_col_in_shm(self, col, shm)

    def _on_column_replaced(self, col: str):
        from brmspy._singleton._shm_singleton import _get_shm

        shm = _get_shm()
        if not shm:
            return
        self._put_col_in_shm(self, col, shm, replace=True)

Attributes

_metadata = ['_shm_metadata'] class-attribute instance-attribute
_shm_metadata instance-attribute
_constructor property

Functions

_create_col_metadata(series, block, arr=None, **params) classmethod
Source code in brmspy/types/shm_extensions.py 
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
@classmethod
def _create_col_metadata(
    cls, series: pd.Series, block: ShmRef, arr: np.ndarray | None = None, **params
) -> ShmSeriesMetadata:
    pd_dtype = series.dtype

    # Decide what ndarray is actually stored in SHM
    if isinstance(pd_dtype, pd.CategoricalDtype):
        # store dtype params
        params["categories"] = pd_dtype.categories.to_numpy(dtype=object).tolist()
        params["ordered"] = bool(pd_dtype.ordered)
        params["categories_dtype"] = pd_dtype.categories.dtype.name

        # IMPORTANT: store integer codes, not values
        # -1 means missing
        array = series.cat.codes.to_numpy(copy=False)
    elif arr is not None:
        assert isinstance(arr, np.ndarray)
        array = arr
    else:
        # for numeric-only SHM: require a real numpy array output here
        array = series.to_numpy(copy=False)

    # Optional param dtypes you mentioned (only if you support them)
    if isinstance(pd_dtype, pd.PeriodDtype):
        params["freq"] = str(pd_dtype.freq)

    if isinstance(pd_dtype, pd.IntervalDtype):
        params["subtype"] = np.dtype(pd_dtype.subtype).str
        params["closed"] = str(pd_dtype.closed)  # type: ignore[attr-defined]

    meta: ShmSeriesMetadata = {
        "name": series.name,
        "np_dtype": str(array.dtype),
        "pd_dtype": str(pd_dtype.name),
        "block": block,
        "params": params,
    }

    if np.dtype(meta["np_dtype"]) == np.dtype("O"):
        # Sanity check. If this goes wrong, it will be frustrating to debug
        if ShmArray.is_string_object(array, sample=25):
            raise Exception(
                f"{series.name} column is string, but stored as object!"
            )

    return meta
_set_col_raw(col, value)
Source code in brmspy/types/shm_extensions.py 
318
319
320
def _set_col_raw(self, col: str, value) -> None:
    # bypass our __setitem__
    pd.DataFrame.__setitem__(self, col, value)
_set_shm_metadata(meta)
Source code in brmspy/types/shm_extensions.py 
322
323
def _set_shm_metadata(self, meta: dict[str, ShmSeriesMetadata]):
    self._shm_metadata = meta
_put_col_in_shm(df, col, shm_pool, replace=False) classmethod
Source code in brmspy/types/shm_extensions.py 
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
@classmethod
def _put_col_in_shm(
    cls, df: "ShmDataFrameColumns", col: str, shm_pool: Any, replace=False
):
    vals = df[col].to_numpy(copy=False)
    if (
        isinstance(vals, ShmArray)
        and hasattr(vals, "_shm_metadata")
        and not replace
    ):
        pass
    elif isinstance(vals, np.ndarray):
        if col in df._shm_metadata:
            del df._shm_metadata[col]
        arr_modified, ref, dtype, shape, order = ShmArray.to_shm(df[col], shm_pool)

        if arr_modified is not None:
            # Only needed for string-object normalization; for numeric/codes it's None
            df._set_col_raw(
                col, pd.Series(arr_modified, index=df.index, name=col, copy=False)
            )
        df._shm_metadata[col] = cls._create_col_metadata(df[col], ref, arr_modified)
        return
    else:
        print(
            f"Failed to update shm metadata for column '{col}' dtype {vals.dtype}"
        )
        return
_reconstruct_series(meta, block, nrows, index) classmethod
Source code in brmspy/types/shm_extensions.py 
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
@classmethod
def _reconstruct_series(
    cls,
    meta: ShmSeriesMetadata,
    block: ShmBlock,
    nrows: int,
    index: list | None,
) -> pd.Series:
    col_name = meta["name"]
    col_name = str(col_name)
    dtype = np.dtype(meta["np_dtype"])
    pd_dtype = meta["pd_dtype"]
    params = meta["params"]

    arr = ShmArray.from_block(block=block, shape=(nrows,), dtype=dtype, order="C")

    if pd_dtype == "category":
        cats = params.get("categories", None)
        ordered = bool(params.get("ordered", False))

        if cats is None:
            raise ValueError("category dtype requires params['categories']")

        cats_pd_dtype = params.get("categories_dtype")
        if cats_pd_dtype is not None:
            cats_index = pd.Index(cats, dtype=str(cats_pd_dtype))
        else:
            cats_index = pd.Index(cats)

        cat_dtype = pd.CategoricalDtype(categories=cats_index, ordered=ordered)

        # arr should hold integer codes
        # If arr holds codes: build categorical from codes without copying codes.
        # Pandas uses -1 for missing.
        cat = pd.Categorical.from_codes(cast(Sequence[int], arr), dtype=cat_dtype)
        return pd.Series(cat, name=col_name, index=index)

    # 2) tz-aware datetimes
    # Expect arr to be int64 ns timestamps
    if pd_dtype.startswith("datetime64[ns,") or pd_dtype == "datetime64[ns, tz]":
        tz = params.get("tz")
        if not tz:
            # if stored as a plain datetime64[ns] string, just fall through
            pass
        else:
            dt = pd.to_datetime(arr, unit="ns", utc=True).tz_convert(tz)
            return pd.Series(dt, name=col_name, index=index)

    return pd.Series(arr, name=col_name, index=index)
__setitem__(key, value)
Source code in brmspy/types/shm_extensions.py 
404
405
406
407
408
409
410
411
412
def __setitem__(self, key, value):
    is_existing = key in self.columns

    super().__setitem__(key, value)

    if is_existing:
        self._on_column_replaced(key)
    else:
        self._on_column_added(key)
_on_column_added(col)
Source code in brmspy/types/shm_extensions.py 
414
415
416
417
418
419
420
def _on_column_added(self, col: str):
    from brmspy._singleton._shm_singleton import _get_shm

    shm = _get_shm()
    if not shm:
        return
    self._put_col_in_shm(self, col, shm)
_on_column_replaced(col)
Source code in brmspy/types/shm_extensions.py 
422
423
424
425
426
427
428
def _on_column_replaced(self, col: str):
    from brmspy._singleton._shm_singleton import _get_shm

    shm = _get_shm()
    if not shm:
        return
    self._put_col_in_shm(self, col, shm, replace=True)

ShmDataFrameSimple

Bases: DataFrame

pandas DataFrame backed by a single shared-memory block (numeric only).

Attributes:

Name Type Description
block ShmRef

Reference to the shared-memory block backing the DataFrame's values.
Source code in brmspy/types/shm_extensions.py 
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
class ShmDataFrameSimple(pd.DataFrame):
    """
    pandas DataFrame backed by a single shared-memory block (numeric only).

    Attributes
    ----------
    block : ShmRef
        Reference to the shared-memory block backing the DataFrame's values.
    """

    _metadata = ["_shm_metadata"]
    _shm_metadata: ShmRef

    @classmethod
    def from_block(
        cls,
        block: ShmBlock,
        nrows: int,
        ncols: int,
        columns: list[Any] | None,
        index: list[Any] | None,
        dtype: str | np.dtype,
    ) -> "ShmDataFrameSimple":
        """
        Construct a DataFrame backed by a single SHM block.

        Parameters
        ----------
        block : ShmBlock
            Attached shared-memory block containing a contiguous 2D numeric matrix.
        nrows, ncols : int
            DataFrame shape.
        columns, index : list[Any] or None
            Column/index labels.
        dtype : str or numpy.dtype
            Dtype of the matrix stored in the block.

        Returns
        -------
        ShmDataFrameSimple
        """
        _dtype = np.dtype(dtype)
        arr = ShmArray.from_block(shape=(ncols, nrows), dtype=_dtype, block=block)

        df = ShmDataFrameSimple(data=arr.T, index=index, columns=columns)
        df._set_shm_metadata(block.to_ref())
        return df

    def _set_shm_metadata(self, meta: ShmRef):
        self._shm_metadata = meta

Attributes

_metadata = ['_shm_metadata'] class-attribute instance-attribute
_shm_metadata instance-attribute

Functions

from_block(block, nrows, ncols, columns, index, dtype) classmethod

Construct a DataFrame backed by a single SHM block.

Parameters:

Name Type Description Default
block ShmBlock

Attached shared-memory block containing a contiguous 2D numeric matrix.

 required
nrows int

DataFrame shape.

 required
ncols int

DataFrame shape.

 required
columns list[Any] or None

Column/index labels.

 required
index list[Any] or None

Column/index labels.

 required
dtype str or dtype

Dtype of the matrix stored in the block.

 required

Returns:

Type Description
ShmDataFrameSimple
Source code in brmspy/types/shm_extensions.py 
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
@classmethod
def from_block(
    cls,
    block: ShmBlock,
    nrows: int,
    ncols: int,
    columns: list[Any] | None,
    index: list[Any] | None,
    dtype: str | np.dtype,
) -> "ShmDataFrameSimple":
    """
    Construct a DataFrame backed by a single SHM block.

    Parameters
    ----------
    block : ShmBlock
        Attached shared-memory block containing a contiguous 2D numeric matrix.
    nrows, ncols : int
        DataFrame shape.
    columns, index : list[Any] or None
        Column/index labels.
    dtype : str or numpy.dtype
        Dtype of the matrix stored in the block.

    Returns
    -------
    ShmDataFrameSimple
    """
    _dtype = np.dtype(dtype)
    arr = ShmArray.from_block(shape=(ncols, nrows), dtype=_dtype, block=block)

    df = ShmDataFrameSimple(data=arr.T, index=index, columns=columns)
    df._set_shm_metadata(block.to_ref())
    return df
_set_shm_metadata(meta)
Source code in brmspy/types/shm_extensions.py 
246
247
def _set_shm_metadata(self, meta: ShmRef):
    self._shm_metadata = meta

Functions

_get_rvector_memview(obj)

Source code in brmspy/helpers/_rpy2/_converters/_vectors.py 
35
36
37
38
39
40
41
42
43
44
45
46
47
48
def _get_rvector_memview(
    obj: Any,
) -> tuple[SexpVectorWithNumpyInterface | None, memoryview | None]:
    try:
        assert isinstance(obj, SexpVectorWithNumpyInterface) and isinstance(
            obj, SexpVector
        )  # assert types, shouldnt error by itself
        if hasattr(obj, "memoryview"):
            src = cast(Any, obj).memoryview()
            return obj, src
        else:
            return None, None
    except:
        return None, None

_get_rvector_types(obj)

Source code in brmspy/helpers/_rpy2/_converters/_vectors.py 
22
23
24
25
26
27
28
29
30
31
32
def _get_rvector_types(obj: Any) -> tuple[None | str, None | int]:
    if not isinstance(obj, SexpVectorWithNumpyInterface):
        return None, None

    dtypestr = obj._NP_TYPESTR
    itemsize = obj._R_SIZEOF_ELT

    if not dtypestr or not itemsize:
        return None, None

    return dtypestr, itemsize

_r2py_vector(obj, shm=None, allow_scalar=True)

Source code in brmspy/helpers/_rpy2/_converters/_vectors.py 
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
def _r2py_vector(
    obj: "Vector", shm: ShmPool | None = None, allow_scalar: bool | None = True
) -> PyObject:
    import rpy2.robjects as ro
    from rpy2.robjects import default_converter
    from rpy2.robjects.conversion import localconverter

    assert not isinstance(obj, ro.ListVector)

    if allow_scalar:
        obj_any = cast(Any, obj)
        # length 1 → scalar
        if obj_any.__len__ and len(obj_any) == 1:
            # Try default R→Python conversion
            with localconverter(default_converter) as cv:
                py = cv.rpy2py(obj[0])
            return py

    is_factor = isinstance(obj, ro.FactorVector)

    dtypestr, itemsize = _get_rvector_types(obj)
    rvecnp, src = _get_rvector_memview(obj)

    # fallback
    if not dtypestr or not itemsize or not shm or not rvecnp or not src:
        return _fallback_rvector_iter(obj)

    # numpy convertible
    N = len(rvecnp)
    expected_bytes = itemsize * N
    dtype = np.dtype(dtypestr)

    if src.nbytes != expected_bytes:
        raise RuntimeError(f"R vector bytes={src.nbytes}, expected={expected_bytes}")

    # Allocate shm once
    block = shm.alloc(expected_bytes)
    assert block.shm.buf

    # Single bulk copy: R → shm, no intermediate ndarray
    src_bytes = src.cast("B")
    block.shm.buf[:expected_bytes] = src_bytes

    arr = ShmArray.from_block(block=block, shape=(N,), dtype=dtype)
    if is_factor:
        return _to_pandas_factor(arr, obj)

    return arr

log_warning(msg, method_name=None)

Log a warning message.

Parameters:

Name Type Description Default
msg str

The warning message to log

 required
method_name str

The name of the method/function. If None, will auto-detect from call stack.

 None
Source code in brmspy/helpers/log.py 
200
201
202
203
204
205
206
207
208
209
210
211
212
def log_warning(msg: str, method_name: str | None = None):
    """
    Log a warning message.

    Parameters
    ----------
    msg : str
        The warning message to log
    method_name : str, optional
        The name of the method/function. If None, will auto-detect from call stack.

    """
    log(msg, method_name=method_name, level=logging.WARNING)

_rmatrix_info(obj)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
30
31
32
33
34
35
36
37
38
39
40
41
42
def _rmatrix_info(obj: "Matrix") -> tuple[int, int, list[str] | None, list[str] | None]:
    nrow, ncol = obj.dim

    if obj.colnames != NULL:
        colnames = [str(el) for el in obj.colnames]
    else:
        colnames = None
    if obj.rownames != NULL:
        rownames = [str(el) for el in obj.rownames]
    else:
        rownames = None

    return nrow, ncol, rownames, colnames

_rmatrix_to_py_default(obj)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
45
46
47
48
49
50
51
52
53
54
55
56
def _rmatrix_to_py_default(obj: "Matrix") -> pd.DataFrame | np.ndarray:
    nrow, ncol, rownames, colnames = _rmatrix_info(obj)

    if not rownames and not colnames:
        return np.array(obj)

    df = pd.DataFrame(data=np.array(obj), columns=colnames, index=rownames)
    if "_obs_id_" in df.columns and not df["_obs_id_"].duplicated().any():
        df.index = df["_obs_id_"]
    elif "obs_id" in df.columns and not df["obs_id"].duplicated().any():
        df.index = df["obs_id"]
    return df

_rmatrix_to_py(obj, shm=None)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
def _rmatrix_to_py(
    obj: "Matrix", shm: ShmPool | None = None
) -> pd.DataFrame | np.ndarray | ShmArray | ShmDataFrameSimple:
    if len(obj.dim) != 2:
        raise Exception("Matrix with dims != 2. Unimplemented conversion")

    # No shm, fall back to regular numpy
    if shm is None:
        return np.array(obj)

    dtypestr, itemsize = _get_rvector_types(obj)

    if not dtypestr or not itemsize:
        return _rmatrix_to_py_default(obj)

    dtype = np.dtype(dtypestr)

    rvecnp, src = _get_rvector_memview(obj)
    if rvecnp is None or src is None:
        return _rmatrix_to_py_default(obj)

    nrow, ncol, rownames, colnames = _rmatrix_info(obj)

    expected_bytes = nrow * ncol * itemsize

    # Raw buffer view over R's underlying data (column-major)
    if src.nbytes != expected_bytes:
        raise RuntimeError(f"R matrix bytes={src.nbytes}, expected={expected_bytes}")

    # Allocate shm once
    block = shm.alloc(expected_bytes)
    assert block.shm.buf

    # Single bulk copy: R → shm, no intermediate ndarray
    src_bytes = src.cast("B")
    block.shm.buf[:expected_bytes] = src_bytes

    # Wrap shm buffer as a numpy array, matching R's column-major layout
    if not rownames and not colnames:
        return ShmArray.from_block(block=block, shape=(nrow, ncol), dtype=dtype)

    df = ShmDataFrameSimple.from_block(
        block=block,
        nrows=nrow,
        ncols=ncol,
        columns=colnames,
        index=rownames,
        dtype=dtype,
    )

    if "_obs_id_" in df.columns and not df["_obs_id_"].duplicated().any():
        df.index = df["_obs_id_"]
    elif "obs_id" in df.columns and not df["obs_id"].duplicated().any():
        df.index = df["obs_id"]

    return df

_r2py_matrix(obj, shm=None)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
120
121
def _r2py_matrix(obj: "Matrix", shm: ShmPool | None = None) -> PyObject:
    return _rmatrix_to_py(obj=obj, shm=shm)

_r2py_dataframe_fallback(obj)

Fallback conversion for R data.frame -> pandas.DataFrame.

Notes

In some environments, rpy2/pandas2ri may convert R factors to their underlying integer codes instead of pandas.Categorical. Since brmspy relies on factors roundtripping as categoricals (and we have custom factor handling in _r2py_vector()), we patch factor columns explicitly here.

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
def _r2py_dataframe_fallback(obj: "DataFrame") -> PyObject:
    """
    Fallback conversion for R data.frame -> pandas.DataFrame.

    Notes
    -----
    In some environments, rpy2/pandas2ri may convert R factors to their underlying
    integer codes instead of `pandas.Categorical`. Since brmspy relies on factors
    roundtripping as categoricals (and we have custom factor handling in
    [`_r2py_vector()`][brmspy.helpers._rpy2._converters._vectors._r2py_vector]),
    we patch factor columns explicitly here.
    """
    import rpy2.robjects as ro
    from rpy2.robjects import pandas2ri
    from rpy2.robjects.conversion import localconverter

    with localconverter(pandas2ri.converter) as cv:
        df = cv.rpy2py(obj)

    # Ensure factor columns come back as pandas categoricals.
    # (Otherwise they can appear as int32 codes with NA_INTEGER sentinel values.)
    try:
        if obj.names is not NULL:
            for name in list(obj.names):
                col_name = str(name)
                if col_name not in df.columns:
                    continue

                col_r = obj.rx2(name)
                if isinstance(col_r, ro.FactorVector):
                    cat = _r2py_vector(col_r, shm=None, allow_scalar=False)
                    df[col_name] = pd.Series(
                        cast(np.ndarray, cat), index=df.index, name=col_name, copy=False
                    )
    except Exception:
        # Best-effort: never let fallback conversion fail due to factor patching.
        pass

    return _adjust_df_for_py(df)

_r2py_dataframe(obj, shm=None)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
def _r2py_dataframe(obj: "DataFrame", shm: ShmPool | None = None) -> PyObject:
    if not shm:
        return _r2py_dataframe_fallback(obj)

    try:
        from rpy2.robjects.pandas2ri import _flatten_dataframe

        rownames = list(obj.rownames)
        if len(rownames) == 0:
            return pd.DataFrame({})

        # convert straight into ShmDataframeColumns
        colnames_lst = []
        od = OrderedDict()
        od_r = OrderedDict()
        for i, col in enumerate(_flatten_dataframe(obj, colnames_lst)):
            arr = _r2py_vector(col, shm, allow_scalar=False)
            od_r[i] = col
            od[i] = arr

        res = ShmDataFrameColumns(od)
        res.columns = tuple(
            ".".join(_) if isinstance(_, list) else _ for _ in colnames_lst
        )
        res.index = rownames
        res._set_shm_metadata({})

        if res.empty:
            return res

        for col in res.columns:
            # Side-effect: ShmDataFrameColumns overrides __setitem__/events;
            # assigning the same Series forces the "put column to SHM" path
            # DO NOT REMOVE!
            res[col] = res[col]
        res = _adjust_df_for_py(res)

        return res
    except Exception as e:
        log_warning(
            f"ShmDataFrameColumns conversion failed, falling back to default. Reason: {e}"
        )
        return _r2py_dataframe_fallback(obj)

_adjust_df_for_r(obj)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
def _adjust_df_for_r(obj: pd.DataFrame) -> pd.DataFrame:
    if "_obs_id_" not in obj.columns:
        if obj.index.nlevels == 1:
            obj["_obs_id_"] = obj.index
        else:
            obj = obj.assign(_obs_id_=range(len(obj)))

    for c in obj.columns:
        s = obj[c]
        if isinstance(s.dtype, pd.CategoricalDtype):
            cats = s.dtype.categories
            # needs string categories for factor conversion
            if cats.inferred_type != "string":
                # log_warning(
                #    f"Column {c} has non-string categories, converting to string."
                # )
                obj[c] = s.cat.rename_categories(cats.map(str))

        elif pd.api.types.is_integer_dtype(s.dtype) and s.dtype != np.int32:
            obj[c] = obj[c].astype("int32")
        elif pd.api.types.is_float_dtype(s.dtype) and s.dtype != np.float64:
            obj[c] = obj[c].astype("float64")

    return obj

_adjust_df_for_py(df)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
236
237
238
239
def _adjust_df_for_py(df: pd.DataFrame) -> pd.DataFrame:
    if "_obs_id_" in df.columns:
        df = df.set_index("_obs_id_", drop=True)
    return df

_py2r_dataframe(obj)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
242
243
244
245
246
247
248
249
def _py2r_dataframe(obj: pd.DataFrame) -> Sexp:
    from rpy2.robjects import pandas2ri
    from rpy2.robjects.conversion import localconverter

    obj = _adjust_df_for_r(obj)

    with localconverter(pandas2ri.converter) as cv:
        return cv.py2rpy(obj)

_py2r_numpy(obj)

Source code in brmspy/helpers/_rpy2/_converters/_arrays.py 
252
253
254
255
256
257
def _py2r_numpy(obj: np.ndarray) -> Sexp:
    from rpy2.robjects import numpy2ri, pandas2ri
    from rpy2.robjects.conversion import localconverter

    with localconverter(pandas2ri.converter + numpy2ri.converter) as cv:
        return cv.py2rpy(obj)