Parsing slows down with each subsequent parse

[connormayer]

Python 3.8.10
apertium-python 0.2.3
apertium 3.8.1
Ubuntu 20.04 (running under WSL)

Reproducing code:

import apertium
import time

i = 0
parser = apertium.Analyzer('eng')
while True:
	then = time.time()
	parser.analyze("Why does this keep getting slower?")
	now = time.time()
	print("Parsing {} took {}".format(i, now - then))
	i += 1

When I run this code, the time taken for the same parse gradually increases. In the first few iterations it takes about 0.0009 seconds per parse, but by iteration 1000 it takes about 0.03 seconds, and it increases from there. Moving parser = apertium.Analyzer('eng') within the loop doesn't solve the problem. If I use the hfst library to run the transducer instead, no slowdown is observed.

[jonorthwash]

I can confirm this issue. I'm able to reproduce on Debian GNU/Linux, Python 3.10.5, apertium 3.8.1.

[jonorthwash]

The amount of memory it uses grows as well. On the first few iterations it uses about 55MB, by 1000 iterations it's up to double that.

[connormayer]

It turns out that this is the specific call that is increasing in execution time: https://github.com/apertium/apertium-python/blob/master/apertium/utils.py#L113. If you set wrappers_available = False to avoid using the wrappers, the parses maintain a constant speed, although they're much slower (at least at first) than when the wrapper is used.