fix: line_length assertion error for odd paths

[JeppeKlitgaard]

This fixes #278, see discussion there.

The proposed fix is likely not the optimal one – I am happy to change it.

@anthrotype

[anthrotype]

hm that would effectively mean the _farthest method would yield two Vector(0.0, 0.0), Vector(0.0, 0.0) in this situation, where the arc endpoint sits exactly on the previous command end point.

Maybe we should simply catch that situation somewhere near

https://github.com/googlefonts/picosvg/blob/c19605d818528329ebd3af90715bfc9054f517db/src/picosvg/svg_reuse.py#L120-131

and simply continue and skip it altogether. That arc is as if it's not there IIUC.

/cc @rsheeter

[anthrotype]

though on the other end, we don't do similar filtering out of such overlapping segments for non arcs in the normalization method, so maybe we should let it through like you are currently doing.

Rod is the author of that code and he knows better what he was thinking when he added that assertion

[rsheeter]

I think there is a general question of whether we should eliminate pointless commands, tracked in #280, but until we address that I think this makes sense.

Ty for the PR.