Comment 9 for bug 1506715

Confirmed - there are situations where filling fails unexpectedly for the user; for example depending on whether or not a traced area is bounded at the right or bottom within the rasterized area (visible screen at zoom level * 1.6). Other aborted traces seem to differ for bounds determined on the left or top side (any bounds detected on the left side may depend on objects fully within or exceeding the screen area to the left, the mirrored situation along the top side though does not AFAICT).

Possibly limitations of the chosen scan-line-processing of the area traced for pixels within the threshold - this would have to be analyzed by a developer familiar with the details of the implementation (whether it is a bug in the implementation or details which could be improved (wishlist)):
https://bazaar.launchpad.net/~inkscape.dev/inkscape/trunk/view/14419/src/ui/tools/flood-tool.cpp#L713
https://bazaar.launchpad.net/~inkscape.dev/inkscape/trunk/view/14419/src/ui/tools/flood-tool.cpp#L582

For visual debugging, comment out these lines:
https://bazaar.launchpad.net/~inkscape.dev/inkscape/trunk/view/14419/src/ui/tools/flood-tool.cpp#L796
to dump the rasterized area used for the scanline checks to a PNG file (in current working dir).