In-place operators on NDArray (a *= 2) return a LazyExpr and leave the array unchanged #760

Hi Francesc,

NDArray.__imul__ and the other in-place operators (__iadd__, __isub__, __itruediv__, __ifloordiv__, __imod__, __ipow__, __iand__, __ior__, __ixor__, __ilshift__, __irshift__) return the out-of-place LazyExpr. So a *= 2 rebinds the name a to a lazy expression, and the array itself, which other names may still refer to, does not change:

import blosc2
import numpy as np

a = blosc2.ones(4, dtype=np.float32)
b = a        # e.g. the caller's reference to the same array
a *= 2
type(a)      # blosc2.lazyexpr.LazyExpr
b[:]         # array([1., 1., 1., 1.], dtype=float32): the array did not change

This is blosc2 4.14.1; the methods are the same on main (a712c65).

With NumPy, a *= 2 updates the array in place, and code written for NumPy arrays relies on that, e.g. a function that rescales an array it is given. Given an NDArray, such code silently does nothing to the caller's data, and nothing raises. I noticed it while looking at nnInteractive, which rescales stored blosc2 channels with interactions[ch] *= scale: that works because it goes through __getitem__ (a NumPy array) and __setitem__, while interactions *= scale would have replaced the attribute with a LazyExpr and left the stored channels as they were.

Would you take a PR that makes these operators work in place? (a * 2).compute(out=a) already gives the right values in place in the cases I tried (float16, float32 and int32, with several chunks), so they could evaluate the expression into self and return self. Two points I have not checked yet: whether compute(out=) is safe when the other operand overlaps self (e.g. a shifted view), and keeping self's dtype (NumPy raises for int_array *= 2.5 under its default same_kind casting). If the lazy result is intended instead, documenting it, and perhaps a warning, would already help.

AI disclosure: I used Claude Code (Anthropic) to look into this and to draft this issue, and Claude Code opened it on my behalf. I checked the examples.

Best,
Fabian