Do you think you could revisit this PR? The changes all look nice and reasonable to me, modulo some style nits, and it'd be great to merge this.

Ok, I'll take a peek at this when I get a chance. The description of it worries me, but maybe I'll see my problem easier with fresh eyes. (I didn't remember that I had any outstanding unmerged for loop code, I thought only the Eigen proof of concept was left.)

Any hint on what the style nits are? Do you want all loop bodies, etc. bracketed? (I normally do that, so seeing a few not like that was the only style thing that jumped out at me rereading this on my phone, at least.). Should be easy enough for me to bisect which change causes problems since most of those diff hunks look independent.

If it turns out the "nitpicks" are purely formatting you could use clang-format:
https://github.com/solvespace/solvespace/blob/master/CONTRIBUTING.md#coding-style

Any hint on what the style nits are?

Oh sorry, I forgot to actually mention. It's just one thing, in a few places you use a reference when iterating then immediately convert it to a pointer and access through that (extraneous) pointer.

Ah, that was to perform minimum changes. No problem, I can fix that.

@rpavlik Any word on this PR or #475 or #460 ? It's fine if you still want to make changes, or have any of them merged. But if they're obsolete I'd like to close them. Thanks!