Simple example to test how much faster or slower it is to simply try and convert then catch an error vs. testing up front.

speed_of_exceptions.py

import datetime
import random
from typing import List


def main():
    random.seed(172)
    count = 1_000_000
    data = build_data(count)
    run_with_except(data)
    run_with_test(data)


def build_data(count: int):
    letters = "abcdefghijklmnopqrstuv"
    numbers = list(range(1, 11))

    combined = list(letters) + numbers

    return [
        str(random.choice(combined))
        for _ in range(count)
    ]


def run_with_except(data: List):
    t0 = datetime.datetime.now()
    numbers = 0
    for item in data:
        try:
            n = int(item)
            numbers += 1
        except:
            pass

    dt = datetime.datetime.now() - t0
    print(
        f'EXCEPT: Done in {dt.total_seconds() * 1000:,.0f} ms, {numbers:,} numbers and {len(data) - numbers:,} letters.')


def run_with_test(data: List):
    t0 = datetime.datetime.now()
    numbers = 0
    for item in data:
        if item.isnumeric():
            n = int(item)
            numbers += 1

    dt = datetime.datetime.now() - t0
    print(
        f'TEST: Done in {dt.total_seconds() * 1000:,.0f} ms, {numbers:,} numbers and {len(data) - numbers:,} letters.')


if __name__ == '__main__':
    main()

Hm, this comes as a surprise for me. I thought "It's easier to ask for forgiveness than permission" meant it was always better with try/catch.

Very good example, thanks!

You're welcome! It is easier to read and easier to maintain often. And many times you have to account for exceptions at lower levels to sorta require both. Therefore, code-wise, it's easier, but in raw perf, it's not necessarily faster. :)

Mh, int() can do more than isnumeric() lets through, e.g. int("-1") does the right thing, "-1".isnumeric() won't. All I want is a fast way of going str -> either int or float without using exceptions :(

M1 Macbook Air:

EXCEPT: Done in 315 ms, 311,656 numbers and 688,344 letters.
TEST: Done in 59 ms, 311,656 numbers and 688,344 letters.

This is an interesting case, Make me thing about Golang, instead of "try catch" it returns the error until it is handled. Maybe Python should be coded such a way if performance is an issue.

@madig Other languages (like C#) have non-exception based versions in addition to this. For example:

// similar style
int x = int.Parse(text)

// exception free (but clumsier) style
int x;
if (!int.TryParse(text, out x))
{
    // error, not an int
}

@theArtechnology Perhaps :) But major changes to the syntax are probably not worth it until this because one of the slower parts of Python. There are many more pieces of low hanging fruit before worrying about this TOO much.