1) Due to a problem with code not relying on 64bit floats, compiling
mclust with optimization level -O2 may give incorrect results: 
z-values are computed larger than 1 (to be more exact: 1 +
.Machine$double.eps), which should not happen. 
If this is the case, mclust will not pass the check stage. 
This has been observed with at least one particular version the g77 compiler
(GNU Fortran 0.5.25 19991030 (prerelease)); the problem disappeared when f2c
was used, and also when no optimization was performed with g77.

2) When incorporating noise in the cluster model, the classification vector
is incorrect in summary.emclust and summary.emclust1. To be more precise:
the model with the maximal BIC value contains fewer clusters than the
classification vector.
