Colormap.__call__ with a scalar and bytes=True raises ValueError

[matthiasschabel]

• cmap version: 0.7.2 (also on main, 0.7.3.dev11+g02fa073)
• Python version: 3.12.12
• Operating System: macOS 26.5.1 (arm64), numpy 2.4.6

Description

Colormap.__call__ with a scalar and bytes=True always raises. Non-scalar array input in
byte mode works and scalar input in float mode works, so it is only the combination that
fails.

Byte mode converts the LUT to uint8, and scalar input wraps the result in Color, which
accepts a 3-element integer array but not a 4-element one (parse_rgba in _color.py).

Two parts of the contract disagree about what should come back, which is why this is an issue
rather than a PR:

• the scalar overload and the docstring promise a Color whatever bytes is set to,
• the bytes documentation promises uint8 values in 0-255.

Either scalar byte mode returns a (4,) uint8 array and the overloads split on bytes,
or it quantizes to 8 bits and returns a Color, which keeps the scalar contract and makes
bytes=True visible only in the rounding (Color(rgba8 / 255) gives #20908C against
#21918C in float mode).

What I Did

import numpy as np
from cmap import Colormap

cmap = Colormap("viridis")
print(cmap(np.array([0.5]), bytes=True))   # [[ 32 144 140 255]]
print(cmap(0.5))                           # #21918C
print(cmap(0.5, bytes=True))               # raises

Traceback (most recent call last):
  File "<string>", line 7, in <module>
  File "/tmp/cmap-072/src/cmap/_colormap.py", line 420, in __call__
    return rgba if np.iterable(x) else Color(rgba)
                                       ^^^^^^^^^^^
  File "/tmp/cmap-072/src/cmap/_color.py", line 529, in __new__
    rgba = parse_rgba(value)
           ^^^^^^^^^^^^^^^^^
  File "/tmp/cmap-072/src/cmap/_color.py", line 467, in parse_rgba
    raise ValueError(f"Invalid color array: {value!r}")  # pragma: no cover
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: Invalid color array: array([ 32, 144, 140, 255], dtype=uint8)

A Python float or int, a numpy scalar, and a 0-d array all take this path, since the
return branches on np.iterable(x).