So I noticed stellarsolver finally made it into kstars. Great work Rob!
I've compiled latest git of both, to have a quick test. It's daylight at the moment, so no real star tests. I only fired up the simulators, and then wanted to use 'Load&Slew', using one of the images of last week as target. That was quite a disappointment. In the default setting it seemed to use 4 threads, and it chewed some 3-4 minutes on the image. I switched back to the old, local astrometry.net. That one's set up to use all 8 threads, and finished faster (2 minutes). But still, this is about the longest time I've seen it using here so far.
So I downgraded to the previous version (kstars-3.5.0-1442_gd604b2835). That one (also using the local installation of astrometry.net) solved the same image in 2 seconds. I assume this is because it found WCS info in the header and used that as starting point (?). So there's hope that things work better on real-sky plate solving, using the mount pointing and FOV of the camera as starter. Not sure though why the local version also took so long in the new setup. Is stellarsolver hiding info from astrometry.net that it did get in the old version?