overflow checks

disable generating code for time profiling

quadratic cost due to use of ++, use 'if (level==0)' instead
of 'if0' in function pL to reduce memory allocation (because if0
is lazy in last two arguments) and improve speed.

compute line width only for new lines when lines are added

fix printing of heap allocation between 1<<31 and 1<<33 bytes.

- removed paths for ArgEnv and Directory

in StartCodeGenerator

flags, instead of GetMessage