performance of country_name_for_number

[kputland]

Why is country_name_for_number so slow compared to region_code_for_number? I'd expect the a lookup for the region_code to country_name to be quite quick. It appears to check validity of the number. I haven't figured out where most time is spent yet. Suppose there's any way to speed things up?

--Karl

from phonenumbers import format_number, is_possible_number, is_possible_number_with_reason, is_valid_number, number_type
from phonenumbers.geocoder import region_code_for_number, country_name_for_number, description_for_number

# setup p for test
p = parse('13035551212', 'US')
print("parse_number:", timeit(lambda: parse('13035551212','US'), number=10000))
print("format_number:", timeit(lambda: format_number(p, phonenumbers.PhoneNumberFormat.E164), number=10000))
print("is_possible_number:", timeit(lambda: is_possible_number(p), number=10000))
print("is_possible_number_with_reason:", timeit(lambda: is_possible_number_with_reason(p), number=10000))
print("is_valid_number:", timeit(lambda: is_valid_number(p), number=10000))
print("number_type:", timeit(lambda: number_type(p), number=10000))
print("region_code_for_number:", timeit(lambda: region_code_for_number(p), number=10000))
print("country_name_for_number:", timeit(lambda: country_name_for_number(p, 'en'), number=10000))
print("desctription_for_number:", timeit(lambda: description_for_number(p, 'en'), number=10000))

results

('parse_number:', 1.0535039901733398) ('format_number:', 0.021859169006347656) ('is_possible_number:', 0.036605119705200195) ('is_possible_number_with_reason:', 0.0437769889831543) ('is_valid_number:', 1.0846400260925293) ('number_type:', 1.1673779487609863) ('region_code_for_number:', 0.5287020206451416) ('country_name_for_number:', 4.455769062042236) ('desctription_for_number:', 1.7113380432128906)

[kevin-brown]

This probably has to do with the fact that country_name_for_number will check all countries to ensure that only a single country is found for a number. Most of the other methods appear to exit as soon as it finds one, even if there might be duplicates.

The change that enforced no duplicates was added in https://github.com/daviddrysdale/python-phonenumbers/commit/0a9a7359c87326b50e06e762615081b273ec3ba0, which points to r701 in the upstream libphonenumber. You can find that revision at https://github.com/googlei18n/libphonenumber/commit/af007418cf449e72bb6e2bd4689d249c8af087d6, and based on the test case this was done because of NANP toll-free numbers.

[kputland]

I understand that logic. It would be nice if it could be optimized at least for country_code == 1 since NPA should be enough to cache a result and there'd be < 1000 values

I guess I could decorate it for my own needs....

[kputland]

Example for anyone else that may want a similar solution.

from functools import wraps
def phonenumber_memoizer(e164_prefix_len=5):
    """e164_prefix_len should be long enough to uniquely identify region information
    
    e164_prefix_len == 5 # +1NPA should be enough to identify a country
    e164_prefix_len == 8 # +1NPANXX should be enough to identify a description/city
    """
    def decorator(f):
        memo = {}
        @wraps(f)
        def wrapper(numobj, *args, **kwargs):
            #print("country_for_number_memoizer: country_code: {}".format(numobj.country_code))
            if numobj.country_code == 1:
                e164_num = format_number(numobj, phonenumbers.PhoneNumberFormat.E164) 
                npa = e164_num[:e164_prefix_len]
                if npa not in memo:
                    memo[npa] = f(numobj, *args, **kwargs)
                return memo[npa]
            else:
                return f(numobj, *args **kwargs)
        return wrapper
    return decorator

memo_number_type = phonenumber_memoizer(e164_prefix_len=8)(number_type)
memo_region_code_for_number = phonenumber_memoizer(e164_prefix_len=5)(region_code_for_number)
memo_country_name_for_number = phonenumber_memoizer(e164_prefix_len=5)(country_name_for_number)
memo_description_for_number = phonenumber_memoizer(e164_prefix_len=8)(description_for_number)

# setup p for test
p = parse('+13035551212', 'US')

print("parse_number:", timeit(lambda: parse('+13035551212','US'), number=10000))
print("format_number:", timeit(lambda: format_number(p, phonenumbers.PhoneNumberFormat.E164), number=10000))
print("is_possible_number:", timeit(lambda: is_possible_number(p), number=10000))
print("is_possible_number_with_reason:", timeit(lambda: is_possible_number_with_reason(p), number=10000))
print("is_valid_number:", timeit(lambda: is_valid_number(p), number=10000))
print
print("number_type:", timeit(lambda: number_type(p), number=10000))
print("region_code_for_number:", timeit(lambda: region_code_for_number(p), number=10000))
print("country_name_for_number:", timeit(lambda: country_name_for_number(p, 'en'), number=10000))
print("description_for_number:", timeit(lambda: description_for_number(p, 'en'), number=10000))
print
print("memo_number_type:", timeit(lambda: memo_number_type(p), number=10000))
print("memo_region_code_for_number:", timeit(lambda: memo_region_code_for_number(p), number=10000))
print("memo_country_name_for_number:", timeit(lambda: memo_country_name_for_number(p, 'en'), number=10000))
print("memo_description_for_number:", timeit(lambda: memo_description_for_number(p, 'en'), number=10000))

results

('parse_number:', 0.9178240299224854) ('format_number:', 0.028091907501220703) ('is_possible_number:', 0.04349517822265625) ('is_possible_number_with_reason:', 0.05940389633178711) ('is_valid_number:', 1.1776368618011475)

('number_type:', 1.135775089263916) ('region_code_for_number:', 0.5933191776275635) ('country_name_for_number:', 4.652255058288574) ('description_for_number:', 1.7402870655059814)

('memo_number_type:', 0.03579902648925781) ('memo_region_code_for_number:', 0.03642702102661133) ('memo_country_name_for_number:', 0.0368959903717041) ('memo_description_for_number:', 0.03756093978881836)