The upper limit on the number of times that we will consecutively do IR1 optimization that doesn’t introduce any new code. A finite limit is necessary, since type inference may take arbitrarily long to converge.
3
