Hi there.

Just some thoughts on potential bug:

The scaleBand type of scale doesn't have the invert() function we need to convert pixel space to values space (see https://d3js.org/d3-scale/band

As a result, the bounds struct will contain x as zero (if undefined, etc.).
see Brush.tsx (convertRangeToDomain())

/// Brush/Brush.tsx

convertRangeToDomain(brush: BaseBrushState) {
    const { xScale, yScale } = this.props;
    const { x0, x1, y0, y1 } = brush.extent;

    const xDomain = getDomainFromExtent(xScale, x0 || 0, x1 || 0, SAFE_PIXEL);
    const yDomain = getDomainFromExtent(yScale, y0 || 0, y1 || 0, SAFE_PIXEL);

    const domain: Bounds = {
      x0: xDomain.start || 0,
      x1: xDomain.end || 0,
      xValues: xDomain.values,
      y0: yDomain.start || 0,
      y1: yDomain.end || 0,
      yValues: yDomain.values,
    };

    return domain;
  }

As I see it, the scaleInvert function wraps cases like this but seems to be the code is focused on ordinalScale only.
see utils.tsx (getDomainFromExtent(), scaleInvert())

May it have the bug?

PS. at the moment I am unable to debug that...sorry
PS. The idea by holgergp to extend the bounds with extra info looks great for me.
If I remember, the d3-brush events contain target object etc as well.

Regards,
Den

@dsdevgit I also get wrong wrong domain report in the end - it's slightly extended (by a couple of pixel).
In onBrushEnd I update original value and then update brush value with updateBrush.
For example if I only move the right boundary - the left one also shifts. Looks like it's happening because of SAFE_PIXEL. Can you explain what's the purpose of it? If I set it to 0 - there are no issues.

Here's the reproduction https://codesandbox.io/p/sandbox/quizzical-khorana-yf9dkw?file=%2FExample.tsx%3A130%2C20 - just click on the brush or try to move/resize.