the attached patch fixes some uninitialized memory
* pdl_initthreadstruct () checks whether the magic
value is correct,
but the value might be correct by accident.
Calls to PDL_THR_CLRMAGIC after every pdl_thread
fix this. I don't understand PP much, so there might
be a better way
than defining a new config setting...
* threadI_redodims would access threadids[-1].
This might be related to #1051810, but it's just a code
not fixing the deep causes.