provide explicit namespace to eval in user_defined_transform()

[bobby-payne]

Summary

This PR addresses #45: a NameError: name 'np' is not defined error occurring when applying certain user-defined transforms (e.g., np.log(x)) in user_defined_transform(). The root cause was that eval() was executed without a namespace, leaving references like np undefined inside the evaluated expression.

Changes

Adds an evaluation context ({"np": np, "x": x}) to eval() so transforms can safely access numpy operations and the variable x.
The execution environment remains isolated — no global variables are leaked or modified.

Modified file

nc2pt/nc2pt/computations.py

Before (beginning on line 30):

    for transform in var.transform:
        try:
            x = 1.0  # noqa: F841
            eval(transform)  # x is implicitly a variable from the config
        except SyntaxError:
            raise SyntaxError(f"Invalid transform in config {transform}.")

        def func(x):
            return eval(transform)

        logging.info(f"🧮 Applying transform {transform} to {var.name}...")
        ds[var.name] = xr.apply_ufunc(func, ds[var.name], dask="parallelized")

After:

    for transform in var.transform:
        try:
            x = 1.0
            eval(transform, {"np": np, "x": x})  # x is implicitly a variable from the config
        except SyntaxError:
            raise SyntaxError(f"Invalid transform in config {transform}.")

        def func(x):
            return eval(transform, {"np": np, "x": x})

        logging.info(f"🧮 Applying transform {transform} to {var.name}...")
        ds[var.name] = xr.apply_ufunc(func, ds[var.name], dask="parallelized")

Notes

Tested with np.nan_to_num(x) as transform on indices calculated from ERA5 and USask-WRF data. Produced no errors and worked as intended.

eval() remains restricted — no access to built-ins or globals other than np and x.

Could later replace with numexpr for improved safety, as eval allows for arbitrary code execution.

[SBeairsto]

Looks good!