Sheehan Olver

The divide by `x` is probably an issue. Try multiplying through

I fixed the transform on master. But now it returns `BigFloat(NaN)`

This works however: `roots(abs(x^BigFloat(2)))` This is because `x^2.0` returns a `Fun` with `JacobiWeight` space.

The reason `roots(abs2(x))` fails is because `roots` for `BigFloat` calls standard roots and then does a small number of Newton iterations: https://github.com/JuliaApproximation/ApproxFun.jl/blob/72d88dfe546687a0c1e7e4aaf039bf9997bf11e0/src/Extras/roots.jl#L89 I think this is misguided, as one reason...

Newton iteration defeats the point of allow `BigFloat`, which is to ramp up the accuracy to get better results. Take for example: ```julia 1E-15sin(1000x) + sin(x) ``` If you use...

Sorry, meant `Sin(100000000000 x)` 😛

Please post a MWE

Can you please include the error message when running the code

The `Warning: Maximum length 1000000 reached` means something has failed to converge. Likely your initial guess is not accurate enough and the Newton solver has spun out of control.

May I suggest a homotopy method for producing the initial guess? That is, choose "nicer" parameters that converge and use that as initial guesses as you approach the desired parameters.