Filippo Battaglia wrote:

Thanks for your answers. Sorry for the title in upper case. I didn't want to create troubles. :)

I've an important question for you: is it possible that a large python module, created using SWIG and with a hundred of routines, makes slower the execution (i.e. the job of ceval.c) of the Python interpreter ?

If you were running on a PC with what is now considered to be very small memory, I would hypothesize that you had filled memory so that the interpreter or parts thereof were being swapped in and out of memory from and to disk. Is any thing like that possible with the PSP? Next, I would wonder whether any module, as part of its initialization, was doing anything 'unusual' with respect to its interaction with the interpreter.

We've observed that, if we don't import ndpsp.pyc at startup, the time of execution of a loop containing the pass instruction becomes near normal.

What happens if you divide the imported stuff in half? Do both halves slow it down? Neither? Just one? The answer to that would be a start to answering whether the specific problem is quantitative or qualitative. Terry Jan Reedy