Constrained layout works pretty well for a number of cases, but the code is pretty convoluted. It has bothered me for a while that it uses the constraint solver, but the constraints are so regulated that they really can just be solved with a bit of calculation.

I probably wouldn't have sat down to redo this on my own, but I burnt up some of my Claude tokens and made a new layout engine that is pretty close to feature compatible with constrained layout.

Pros:

• no kiwisolver dependency: the kiwisolver folks have been great, but it is a C dependency and hence a bit fragile to packaging quirks.
• the code is less indirect, and I'd say easier to follow.
• It is faster for large grids (though such grids are perhaps well beyond anything most people should be making anyway)
  

Cons:

• it is still in it infancy compared to constrained layout
• it does not deal with fixed aspect ratio plots (yet), like compressed layout does.
• it is not pixel perfect comparison (see the comparison here). You can also see how it does on the test images at https://github.com/jklymak/mpl-direct-layout/tree/main/tests/baseline_images

Posting for feedback on strategy here. It could be a standalone package, it could be another layout manager, or we could try and replace constrained layout altogether with it (though I'd want to wait a bit to put it through more tests).