@sgiehl Tests are in progress. From the first results, It seems it's a little better but I can't import more than 800-810req/s (which is still quite good compared to the results without any cache). I'm making some more tests but I think it's because with the factory, we're calling the parse() method for everything and not only for bots.
DeviceDetector skips parsing additional data when a bot is detected. So it
should be the optimal way as parsing is always only done once.
The only thing that could be improved might be caching more devicedetector
instances. But I guess that would only make sense using memcache or
something similar as file cache is too slow for that.
Yes, I saw the tests after posting.
After several tests with 2.4.0-b5 and this branch, that's OK for me, the performances are nearly equal, using the same logs.
That was a great Pull request... well done Stefan, you're the man! Merging now... feel free to make edits as you see fit!