Using Python as an intermediate to access external resources allows
writing such interfaces with less effort by a much wider community. The
resulting interface is often also more robust due to well defined data
conversion and sound memory management that you get for free.
Nevertheless, Python often accesses resources with a C or C++ API. We
can also create this bridge directly, bypassing Python. That avoids one
layer of data conversion and preserves the excellent multi-threading
capabilities of SWI-Prolog. As is, Python operations are synchronized
using the Python GIL, a global lock that allows for only a
single thread to use Python at the same time.9There
are rumors that Python's multi threading will be able to use multiple
cores.
Writing an interface for SWI-Prolog is typically easier that for
Python/C because memory management is easier. Where we need to manage
reference counts to Python objects through all possibly paths of the C
functions, SWI-Prolog term_t merely has to be allocated
once in the function. All failure parts will discard the Prolog data
automatically through backtracking and all success paths will do so
through the Prolog garbage collector.10Using
a Python C++ interface such as pybind11
simplifies memory management for a Python interface.
Summarizing, Janus is ideal to get started quickly. Applications that
need to access C/C++ resources and need either exploit all cores of your
hardware or get the best performance on calls or exchanging data should
consider using the C or C++ interfaces of SWI-Prolog.
