Wow, ok.. turns out this has been attempted before in #1058 and #1286.

I'm still hitting a case where everything just freezes with my all Y values being NaN buffer, so I can't really validate the code.. I'm not sure if it's related to the arr[2:]['c'] = isfinite which was patched in b61c7c1 .. will have to investigate more. Running on Py3.8 with PySide2 5.15.1 with pyqtgraph 0.11.0 + this patch and the mentioned commit.

Also, could it be that with the fill-in solution, that for time-series, where the X-axis is generally the (finite) timestamp, filling-in values will overwrite the timestamps of the points that aren't rendered and thus cause auto-scaling to behave differently since the NaN-cleaned version of the buffer will have a different starting point?
To solve that, I suppose we'd have to clean-up only the non-finite Y value with something like the code below. Since the NaN-turned-zero values are non-connect they shouldn't appear.

    if qtver >= [5, 12, 3]:
        isfinite = np.isfinite(x) & np.isfinite(y)
        isfinite_x = np.isfinite(x)
        isfinite_y = np.isfinite(y)
        if not np.all(isfinite_x):
            # credit: Divakar https://stackoverflow.com/a/41191127/643629
            mask = ~isfinite_x
            idx = np.arange(len(x))
            idx[mask] = -1
            np.maximum.accumulate(idx, out=idx)
            first = np.searchsorted(idx, 0)
            if first < len(x):
                # Replace all non-finite X-axis points from beginning of arr with the first finite one
                idx[:first] = first # np.max(first, len(x) - 1)
                arr[1:-1] = arr[1:-1][idx]
        if not np.all(isfinite_y):
            mask = ~isfinite_y
            # Replace all non-finite Y-axis points with 0
            arr[1:-1]['y'][mask] = 0