Constraints might not be fulfilled without a warning

[nbruder]

If the constraints are too restrictive, adoptr cannot find a suitable design. But no warning is returned, so the user does not know whether the returned design does indeed fulfill the constraints. A small example:

null        <- PointMassPrior(.0, 1)
alternative <- PointMassPrior(.4, 1)
datadist    <- Normal()
ess   <- ExpectedSampleSize(datadist, alternative)
toer  <- Power(datadist, null)
power <- Power(datadist, alternative)

initial_design <- get_initial_design(.4, 0.025, 0.2, "two-stage", datadist)
incorrect_design <- minimize(ess, subject_to(toer <= 0.025, power >= 0.8, MaximumSampleSize() <= 90), 
                         initial_design = initial_design)

evaluate(toer, incorrect_design$design)
evaluate(power, incorrect_design$design)

This could be an issue of the underlying optimization algorithm from the nloptr package.