[thomasklemm]

Fixes #695.

When DOM.getContentQuads fails, Node#find_position (and so Node#click) falls back to bounding_rect_coordinates. That added half of offsetWidth/offsetHeight to the bounding rect's left/top. The rect includes transforms, the offset size doesn't, so for a scaled or rotated node the point could land outside it. This uses the rect's own width/height, which gives the node's visual center for any 2D transform and the same point as before for an untransformed node.

The fallback now looks like this:

def bounding_rect_coordinates
  evaluate(<<~JS)
    function() {
      const { left, top, width, height } = this.getBoundingClientRect();
      return [left + (width / 2), top + (height / 2)];
    }
  JS
end

Test

The new example in the existing when content quads can't be computed context rotates a 200×20 link by 45° and clicks it. A rotated element covers the center of its bounding rect but neither its corners nor the old left + offsetWidth / 2 point. So the spec fails with the old formula (the URL stays at /simple) and also with a top-left-corner variant, and passes with the fix. It passed 20 of 20 repeated runs.

• spec/node_spec.rb: 166 examples, 0 failures
• Full suite: 1235 examples, 2 failures, 4 pending. The two failures (Options::Chrome.version and colors protocol lines written to a terminal) fail the same way on main in my environment: there is no chrome on PATH, and dbus messages end up in the log.
• RuboCop is clean.

Side effect

An SVG element has no offsetWidth, so the old fallback returned [NaN, NaN] and mouse.click raised JSON::GeneratorError. It now gets the center of its rect. The CHANGELOG mentions this.

Out of scope (unchanged)

• The rect is relative to the node's frame while the mouse uses page coordinates, so the fallback ignores iframe offsets, as before.
• The raise if x.zero? && y.zero? guard stays as it is. With the corrected point it can also fire for a visible transformed node whose rect center lies exactly at the viewport origin. It's an existing sentinel for "no box", and changing it would be a separate behaviour change.
• For an inline element wrapped across lines, the rect center can fall between the fragments, as before.

[thomasklemm]

The one failing test on every Ruby version, spec/contexts_spec.rb:9 ("detaches from attached targets of a type it doesn't track"), isn't caused by this change. This run got Chrome for Testing 157.0.8103.38 (chrome-version: stable), while the last green main run (944be71) used 156.0.8078.80. Locally, the spec fails on unchanged main with 157.0.8103.38 and passes with 156. It waits for at least two untracked browser_ui targets per context, and Chrome 157 seems to open fewer. spec/node_spec.rb passes with both versions.

[thomasklemm]

The contexts_spec.rb failure is fixed separately in #697. I'll rebase this PR once that's merged.

[route]

Looking good thank you!