Oscar Dowson

@blegat: am I right in reading that a PSDVAR cannot itself appear in an affine PSD constraint? Getting this right seems pretty nasty.

It can with Mosek, but not with CBF: https://docs.mosek.com/latest/capi/cbf-format.html  $G_i$ does not include $X_j$ variables, and HCOORD explicitly mentions only scalar variables: 

> If the CBF writer sees PSD variables in a VAF-in-PSD constraint, it should error then. Currently we don't write any PSDVAR, so this isn't a problem. We can add...

> that variables declared PSD can't be used in affine constraints There's a subtle correction needed: they can be used in affine-in-cone constraints, but NOT affine-in-PSD constraints.

> Forgive my ignorance, but can't one refer to "inner" variables of a PSDVAR by their scalar indices? No, because `PSDVAR` are special "matrix" variables. They are not like `x...

A new version of CBF could introduce the vectorized PSD cone and deprecate PSDVAR and PSDCON. There's no need for special handling, and I'm reticent to spend too much time...

Currently we have one that supports only PSDCON. I think it's fair to leave it at that. It's still a valid file and model. The solver reader could be clever...

We're not going to start throwing an error because that would be breaking. This is also not a real-world issue. No one, apart from @rbassett3 has ever asked about it,...

> Isn't the first comment of this issue saying that PSD variables are currently ignored ? No. * Currently, all `VectorOfVariables-in-PositiveSemidefiniteConeTriangle` and `VectorAffineFunction-in-PositiveSemidefiniteConeTriangle` constraints get written in the `PSDCON` section....

Did you try `ZerosBridge`? It still failed in a bunch of places