RUNTIME: Variable forcing now uses limited list and buffer instead of systematical instance tree traversal and in-tree "fvalue" to keep track of forced value for pointed variables (external, located). Pointer swapping is performed when forcing externals and located, with backup being restored when forcing is reset. Retain still uses tree traversal.
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# This file is part of Beremiz
# Copyright (C) 2021: Edouard TISSERANT
#
# See COPYING file for copyrights details.
# Based on Eelco Hoogendoorn stackoverflow answer about RingBuffer with numpy
import numpy as np
class RingBuffer(object):
def __init__(self, width=None, size=131072, padding=None):
self.size = size
self.padding = size if padding is None else padding
shape = (self.size+self.padding,)
if width :
shape += (width,)
self.buffer = np.zeros(shape)
self.cursor = 0
def append(self, data):
"""this is an O(n) operation"""
data = data[-self.size:]
n = len(data)
if self.size + self.padding - self.cursor < n:
self.compact()
self.buffer[self.cursor:][:n] = data
self.cursor += n
@property
def count(self):
return min(self.size, self.cursor)
@property
def view(self):
"""this is always an O(1) operation"""
return self.buffer[max(0, self.cursor - self.size):][:self.count]
def compact(self):
"""
note: only when this function is called, is an O(size) performance hit incurred,
and this cost is amortized over the whole padding space
"""
print 'compacting'
self.buffer[:self.count] = self.view
self.cursor -= self.size