rscase 1.1.0

Python package for string case formatting; implemented in Rust.

This module offers a handful of case-formatting utility functions. It is a very simple Python package, written in Rust and implemented using pyo3 which offers you easy Rust bindings for the Python interpreter.

Installation

Install with pip using:

pip install rscase

Note: This package requires Rust nightly 2020-02-06 or an equivalent future release.

Usage

The package provides utility functions for generating strings formatted in several different case standards.

The case-standards and their functions are listed below.

Supported cases	Function	Format example
camel case	camel_case	camelCasedValue
snake case	snake_case	snake_cased_value
pascal case	pascal_case	PascalCasedValue
kebab case	kebab_case	kebab-cased-value
train case	train_case	TRAIN-CASED-VALUE

All functions are imported and accessed the same way:

>> [in] from rscase import rscase
>> [in] rscase.camel_case('this_is-a_Test')
>> [out] thisIsATest

If you want to use this package, please note that the case functions are written to successfully convert camel case and snake case to the remaining formats. Formatting train case to itself doesn’t really make sense, and the way I would use this would be to, e.g., serialize out response data to a camelCased format.

Benchmarking Performance

This repo is a bit of an experiment, and because the functions contained in this package only do some very simple string manipulation, they seem like they might actually be good candidates for Python vs Rust performance benchmarking.

To try and make this a fair comparison - to make sure we’re comparing apples to apples - I decided to test the Rust function snake_case (see the Rust function here) to an identical Python function. The Python version is shown below:

from rscase import rscase

test_string = "thisIsALongCamelCasedAlphabeticKey"

# Test functions

def original_snake_case():
    string = test_string
    new_string = ""
    dash = "-"
    for index in range(len(string)):
        if index == 0:
            new_string += string[index].lower()
        elif string[index] == dash:
            new_string += '_'
        elif string[index].upper() == string[index]:
            new_string += f'_{string[index]}'
        else:
            new_string += string[index]
    return new_string

def rust_snake_case():
    string = test_string
    return rscase.snake_case(string)

The main difference between the two functions, flow-wise, is only that Rust won’t let you just iterate over a string, so you have to create a vector of char’s instead - or at least that’s how I did it.

Results

After running the tests, the results seems to be pretty promising - in favor of the Rust implementation.

Reps	Rust Execution Time	Python Execution Time	Difference
1	18.30 us	14.20 us	0.78x*
10	55.20 us	114.20 us	2.07x
100	.49 ms	1.11 ms	2.27x
1000	4.88 ms	11.18 ms	2.28x
10 000	47.20 ms	109.13 ms	2.31x
100 000	.47 s	1.08 s	2.31x
1000 000	4.83 s	11.12 s	2.30x
10 000 000	46.67 s	109.27 s	2.34x
100 000 000	484 s	1102 s	2.28x

The results are pretty clear: after only 100 reps, the results seem to stabilize, and flatten out at around a 2.3x longer execution time for the Python implementation.

* the 1-rep result seems to show that Python actually outperforms Rust in the scenario that would normally actually matter. Since it makes sense that variance would be high when trying to measure something at the microsecond level I decided to run this individual scenario again, another one million times, to increase the sample size. With a larger sample, the average difference for 1 rep averages to 1.85x slower in Python, and the median is 1.88x. In short, the Rust implementation seems to outperform the Python across the board.

Benchmarking Performance - Update

Thanks to Thomas Hartmann for suggesting a significant performance improvement in the packaged Rust code.

Using some experimental features, we’re able to improve the performance of the Rust code considerably. The snake_case test from above is replicated below, with the performance difference settling at 5x the Python performance.

Reps	Rust Execution Time	Python Execution Time	Difference
1	10.70 us	15.20 us	1.42x
10	28.70 us	113.30 us	3.95x
100	.24 ms	1.11 ms	4.56x
1000	2.24 ms	11.28 ms	5.03x
10 000	22.16 ms	107.79 ms	4.86x
100 000	.24 s	1.09 s	4.44x
1000 000	2.21 s	11.02 s	4.99x
10 000 000	22.09 s	110.47 s	5.00x
100 000 000	222 s	1086 s	4.88x

Running the 1 rep scenario one million times, gives an average Rust execution time of 3.84 us compared to an average Python execution time of 12.61 us (~3.3x slower for Python).

This time around, I also decided to test the camel case implementations, as the logic does behave a bit differently:

Reps	Rust Execution Time	Python Execution Time	Difference
1	10.99 us	14.40 us	1.31x
10	39.79 us	106.90 us	2.69x
100	.25 ms	1.02 ms	4.07x
1000	2.40 ms	10.24 ms	4.26x
10 000	23.55 ms	100.17 ms	4.25x
100 000	.23 s	0.98 s	4.26x
1000 000	2.34 s	9.92 s	4.23x
10 000 000	23.23 s	98.91 s	4.26x
100 000 000	232 s	990 s	4.26x

Running the 1 rep scenario one million times, gives an average Rust execution time of 3.90 us compared to an average Python execution time of 11.48 us (almost ~3x slower for Python).

In summary, the benchmarked performed similarly, with Rust pulling ahead even more, for these two implementations. At the same time, there’s probably room for improvement for both implementations still, and probably especially for the Python one.