Andreas Noack

I don't think it's surprising that a coefficient is dropped since the default is `dropcollinear = true` and ```julia julia> cond(X) 1.7679681910558162e15 ``` It's surprising, though, that we error out...

The QR based least squares solution is ```julia julia> F = qr(X); julia> F.R\(F.Q'*df_filip.y)[1:size(F, 2)] 11-element Vector{Float64}: -1467.4896018652016 -2772.179569812206 -2316.3710641177045 -1127.9739329347135 -354.4782313151358 -75.12420126161706 -10.875317970213466 -1.0622149798557083 -0.06701911509851412 -0.002467810770121917 -4.029625231108671e-5 ```

Oh. Now I realize what is going on in the pivoted case. There is a rank determination by default, so the relevant comparison is ```julia julia> ldiv!(qr(X, ColumnNorm()), df_filip.y[:,:], 0.0)[1][1:size(X,...

> Regarding `dof`, you mean we should deprecate it because people confuse it with `dof_residual`? Yeah. The confusion seems to show up repeatedly and I think that `dof_residual` is all...

Would be good to show the before and after in a comment

The example works fine for me. Could you provide a complete example with a complete error message and info about the versions of Julia and your packages?

I just tried on 0.7 and it is indeed an issue there. Please report this as a separate issue. It used to work on 0.6 so it is a regression.

It happens [here](https://github.com/JuliaLang/julia/pull/19594/files#diff-38a12c96c28b00e80d91e466c284e3dfR86) where the logic for moving globals is implemented. cc: @amitmurthy

See the discussion in https://github.com/JuliaLang/julia/issues/19578. Right now our `pmapreduce` is hidden as `@parallel (op) for`

Just wanted mention that it also seemed that https://github.com/JuliaLang/julia/pull/22588 made adding remote workers noticeably faster.