[omrzk]

_input.upload_effect() calls PyBytes_AsString() and copies from its result even when conversion failed, so a non-bytes argument crashes the interpreter. Check conversion and require enough bytes for struct ff_effect before copying. Complete byte strings (including subclasses and trailing data), ioctl errors, and the InputDevice.upload_effect() serialization path retain their behavior.

Fixes #261.

Validation: all 13 new native-extension tests pass on Python 3.14. On current main, six non-bytes cases terminate subprocesses with SIGSEGV and three short-buffer cases fail; four preservation controls pass. All 20 tests outside the uinput module pass. The full suite has 24 passes and six failures because this runner has no /dev/uinput; the same six fail on clean main. Native build, wheel/sdist build, pylint error check and changed-test Ruff pass. Documentation builds with two missing ecodes_generated warnings. No actual force-feedback/input hardware was tested.

[sezanzeb]

Thanks. The change works for me, it raises a TypeError as excpected via the code sample in #261:

Python 3.14.4 (main, Aug 20 2026, 10:41:58) [GCC 15.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import evdev._input
... 
... evdev._input.upload_effect(0, None)
... 
Traceback (most recent call last):
  File "<python-input-0>", line 3, in <module>
    evdev._input.upload_effect(0, None)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^
TypeError: expected bytes, NoneType found
>>> 

pipeline passed

[sezanzeb]

wrapping bytes(ff.Effect()) in EffectBytes seems to be redundant

[sezanzeb]

This prints true:

from evdev import _input, device, ff

class EffectBytes(bytes):
    pass

a = EffectBytes(bytes(ff.Effect()))
b = bytes(ff.Effect())
c = EffectBytes(ff.Effect())
print(a)
print(b)
print(c)
print(a == b == c)

[sezanzeb]

Or is there something I'm missing?

[omrzk]

The subclass case is intentional: equal payloads still reach the extension as different Python types. The first case covers plain bytes; EffectBytes checks that validation continues to accept bytes subclasses, and the extra-data case preserves trailing-byte acceptance.

You are right that the inner bytes() conversion is redundant: EffectBytes(ff.Effect()) produces the same subclass and payload. I checked all three expressions locally; each reaches ioctl and raises EBADF with fd -1. The subclass wrapper is what adds coverage here.