Field / ScalarField

C++: phynexis::fields::Field<T> Python: phynexis.fields.ScalarField / Int32Field / Int64Field / BoolField Header: src/fields/core/field.hpp

Concrete one-dimensional homogeneous field container, the workhorse type of the fields module. The C++ template Field<T> is exposed in Python as one non-templated class per primitive value type. Field is an alias for ScalarField (i.e. Field<double>), and Idx32Field / Idx64Field are aliases for Int32Field / Int64Field.

Python class	C++ instantiation	Element type
ScalarField (alias Field)	Field<double>	float (64-bit)
Int32Field (alias Idx32Field)	Field<Int32>	int (32-bit)
Int64Field (alias Idx64Field)	Field<Int64>	int (64-bit)
BoolField	Field<bool>	bool

Constructors

Signature	Description
ScalarField()	Empty field, no name.
ScalarField(size)	Default-initialised field of length size.
ScalarField(size, value)	Field of length size filled with value.
ScalarField(name, size=0)	Named field, optionally pre-sized.
ScalarField(name, size, value)	Named field, pre-sized and filled.
ScalarField(data, name)	Build from a VecX<T> buffer with a name.
ScalarField(other)	Copy constructor.

The ScalarField(data: VecX<T>) and ScalarField(name, data: VecX<T>) overloads exist in the binding but require a VecX<T> instance, which is not exposed in Python. Use from_numpy (see below) for the numpy round-trip instead.

Example:

import phynexis

F = phynexis.fields

print(repr(F.ScalarField()))
print(repr(F.ScalarField(5)))
print(repr(F.ScalarField(5, 3.14)))
print(repr(F.ScalarField("rho", 5)))
print(repr(F.ScalarField("rho", 5, 1000.0)))

Output:

ScalarField(size=0, name='')
ScalarField(size=5, name='')
ScalarField(size=5, name='')
ScalarField(size=5, name='rho')
ScalarField(size=5, name='rho')

NumPy Interop

to_numpy()

Returns a 1-D numpy.ndarray with a copy of the field data.

Returns: numpy.ndarray — same dtype as the field element type.

class method from_numpy(array)

Creates a new field whose contents are copied from array. The element type is fixed by the binding type, so the dtype must match (float64 for ScalarField, int32 for Int32Field, etc.).

Parameters:

Parameter	Type	Description
array	numpy.ndarray	1-D array; dtype must match the field type

Example:

import numpy as np
import phynexis

F = phynexis.fields

sf = F.ScalarField.from_numpy(np.linspace(0.0, 1.0, 5))
print(repr(sf))
print(sf.to_numpy())

Output:

ScalarField(size=5, name='')
[0.   0.25 0.5  0.75 1.  ]

Element Access

Method / operator	Description
field[idx]	Read/write element. No bounds check — out-of-range indices read uninitialised memory.
len(field)	Equivalent to field.size().
at(index)	Bounds-checked read/write. Raises IndexError if out of range.
at_const(index)	Bounds-checked read-only access.
data()	Returns the value at index 0 (not a buffer). Mostly useful as a presence check.

Example:

import phynexis

F = phynexis.fields

sf = F.ScalarField("v", 4, 0.0)
sf[0], sf[1], sf[2], sf[3] = 1.0, 2.0, 3.0, 4.0
print("len(sf):", len(sf))
print("sf[2]:", sf[2])
print("sf.at(2):", sf.at(2))

try:
    sf.at(99)
except IndexError as e:
    print("sf.at(99):", repr(e))

Output:

len(sf): 4
sf[2]: 3.0
sf.at(2): IndexError('Field index out of range')

注意

field[idx] does not validate the index. Negative indices or indices beyond size() will return arbitrary memory contents instead of raising. Use field.at(idx) whenever the index is not statically known.

Capacity

Method	Description
size()	Number of elements currently stored.
empty()	True if size() == 0.
capacity()	Allocated buffer size; always >= size().
resize(new_size)	Grow / shrink to new_size; new elements are default-initialised.
resize(new_size, value)	Grow to new_size, padding with value.
reserve(capacity)	Pre-allocate to avoid reallocation. Does not change size().
clear()	Sets size() to 0. Capacity is retained.
push_back(value)	Append a single element.
pop_back()	Remove the last element.

Example:

import phynexis

F = phynexis.fields

sf = F.ScalarField()
sf.reserve(100)
print("after reserve(100): size=", sf.size(), "capacity=", sf.capacity())

sf.push_back(1.0)
sf.push_back(2.0)
print("after 2 push_back:", sf.to_numpy(),
      "size=", sf.size(), "capacity=", sf.capacity())

sf.pop_back()
print("after pop_back:", sf.to_numpy())

sf.clear()
print("after clear: size=", sf.size(), "capacity=", sf.capacity())

sf.resize(3)
print("resize(3):", sf.to_numpy())

sf.resize(5, 9.0)
print("resize(5, 9.0):", sf.to_numpy())

Output:

after reserve(100): size= 0 capacity= 100
after 2 push_back: [1. 2.] size= 2 capacity= 100
after pop_back: [1.]
after clear: size= 0 capacity= 100
resize(3): [0. 0. 0.]
resize(5, 9.0): [0. 0. 0. 9. 9.]

Bulk Operations

Method	Description
fill(value)	Set all elements to value.
append(other)	Concatenate other (same field type) onto the end of this field.
append_span(buffer)	Append from a VecX<T> buffer (not exposed in Python — see notes below).
swap(other)	Swap contents with another field of the same type, including names.
subset(mask)	Build a FieldHolder containing only elements where mask[i] == True. The mask must be a BoolField of the same length.

Example:

import numpy as np
import phynexis

F = phynexis.fields

# fill
sf = F.ScalarField("a", 4, 0.0)
sf.fill(7.0)
print("after fill(7):", sf.to_numpy())

# append
a = F.ScalarField("a", 3, 1.0)
b = F.ScalarField("b", 2, 9.0)
a.append(b)
print("a.append(b):", a.to_numpy())

# swap (note: name is swapped too)
a = F.ScalarField("a", 3, 1.0)
b = F.ScalarField("b", 3, 2.0)
a.swap(b)
print(f"after swap: a name={a.name} data={a.to_numpy()}")
print(f"            b name={b.name} data={b.to_numpy()}")

# subset
sf = F.ScalarField.from_numpy(np.array([10.0, 20.0, 30.0, 40.0, 50.0]))
mask = F.BoolField.from_numpy(np.array([True, False, True, False, True]))
holder = sf.subset(mask)
print("subset returns:", type(holder).__name__)

Output:

after fill(7): [7. 7. 7. 7.]
a.append(b): [1. 1. 1. 9. 9.]
after swap: a name=b data=[2. 2. 2.]
            b name=a data=[1. 1. 1.]
subset returns: FieldHolder

Arithmetic (Double only)

ScalarField (i.e. Field<double>) supports element-wise arithmetic with both scalars and other ScalarFields of the same length. Integer and bool fields do not implement these operators — only Field<double>.

Operator	Field, Field	Field, scalar	scalar, Field
+	element-wise	broadcast	broadcast
-	element-wise	broadcast	broadcast
*	element-wise	broadcast	broadcast
/	element-wise	broadcast	—

Example:

import phynexis

F = phynexis.fields

a = F.ScalarField("a", 4, 1.0)
b = F.ScalarField("b", 4, 2.0)

print("a + b:", (a + b).to_numpy())
print("a - b:", (a - b).to_numpy())
print("a * b:", (a * b).to_numpy())
print("b / a:", (b / a).to_numpy())
print("a + 5:", (a + 5).to_numpy())
print("5 + a:", (5 + a).to_numpy())
print("5 - a:", (5 - a).to_numpy())
print("3 * a:", (3 * a).to_numpy())

Output:

a + b: [3. 3. 3. 3.]
a - b: [-1. -1. -1. -1.]
a * b: [2. 2. 2. 2.]
b / a: [2. 2. 2. 2.]
a + 5: [6. 6. 6. 6.]
5 + a: [6. 6. 6. 6.]
5 - a: [4. 4. 4. 4.]
3 * a: [3. 3. 3. 3.]

The mismatched-size case raises RuntimeError("Field sizes must match").

Comparison Semantics

== and != test identity, not value equality:

return data_ == other.data_ && name_ == other.name_;

That is, two fields compare equal only if they share the same internal pointer and have the same name — effectively only when comparing a field with itself or an alias.

import phynexis

F = phynexis.fields

a = F.ScalarField("x", 3, 1.0)
b = F.ScalarField("x", 3, 1.0)  # same name and contents, different storage
print("identical contents, separate fields a == b:", a == b)
print("self-comparison           a == a:", a == a)

Output:

identical contents, separate fields a == b: False
self-comparison           a == a: True

For value-equality, compare to_numpy() arrays.

Properties

Property	Type	Access	Description
name	str	read/write	Field name (used by FieldManager, I/O, etc.).

Type Aliases and Integer / Bool Fields

import phynexis

F = phynexis.fields

print("Field is ScalarField:    ", F.Field is F.ScalarField)
print("Idx32Field is Int32Field:", F.Idx32Field is F.Int32Field)
print("Idx64Field is Int64Field:", F.Idx64Field is F.Int64Field)

i32 = F.Int32Field("idx", 5, 7)
i32[2] = 999
print("Int32Field:", i32.to_numpy())

i64 = F.Int64Field("idx", 3, 10**12)
print("Int64Field:", i64.to_numpy())

bf = F.BoolField("active", 4, True)
bf[1] = False
print("BoolField :", bf.to_numpy())

Output:

Field is ScalarField:     True
Idx32Field is Int32Field: True
Idx64Field is Int64Field: True
Int32Field: [  7   7 999   7   7]
Int64Field: [1000000000000 1000000000000 1000000000000]
BoolField : [ True False  True  True]

Known Issues

• field[idx] is unchecked. Use field.at(idx) whenever the index might be out of range or negative — field[-1] returns garbage instead of the last element.
• data() returns the first element, not a buffer pointer. Use to_numpy() to obtain the data as an array.
• np.array(field, copy=False) segfaults. The class declares pybind11::buffer_protocol() but no def_buffer callback. Use to_numpy().
• == is identity equality, not value equality (see above).
• Field<T> is not registered as a subclass of FieldBase in pybind, so isinstance(field, FieldBase) is False and the FieldBase-only methods (ensure_capacity, is_same_type, erase_indices, reorder, reset_value, swap_elements, copy_element) are unreachable from Python field instances. Tracked in binding-repair-backlog.md (2026-05-03 entry).

Unexposed C++ API

The following C++ methods exist on Field<T> but are not bound:

• value_type() / field_type() — type tag queries (ValueType, FieldType).
• is_same_type(other) — type-tag equivalent of isinstance for fields.
• erase_indices(indices), reorder(new_to_old) — bulk index operations.
• reset_value(i) / reset_values() / set_value(i, void*) / get_value_ptr(i) — generic value pokes used by FieldHolder internals.
• subset_inplace(mask) — in-place variant of subset.
• save_to(path, file, opt) / load_from(path, file, opt) — file persistence; the alternative free-function entry points are write_binary / read_binary (documented separately under field I/O).
• to_binary(out) / from_binary(in) / pack(out, idx) / unpack(in, idx) — binary serialisation hooks consumed by the manager.
• Field<T>(name, comp_id, size, value) — component-field constructor used by schema-driven layouts.
• Iteration via begin() / end() — Python-side iteration is available through to_numpy() or [i] access.