Consecutive underscores in camelize/underscore transformations aren't reversible

[levic]

Every underscore/camelize cycle drops an underscore in the case of consecutive underscores.

Consider the following snippet:

import inflection
x = 'a____b_c'
for i in range(4):
	x = inflection.camelize(x)
	print(x)
	x = inflection.underscore(x)
	print(x)

Output:

A__bC
a__b_c
A_bC
a_b_c
ABC
abc
Abc
abc

I understand that not every transformation is going to be cleanly reversible.

Given that underscores are already word separators a number of python packages use double underscores as higher-level separators though (most notably django's ORM but also eg fixture generators factoryboy and model mommy). Preserving reversibility of consecutive underscores is a useful property.

The tests and docs don't seem cover the behaviour of consecutive underscores -- it seems to be undefined.

Would you accept a patch that changed the behaviour to preserve consecutive underscores? My proposal is:

• For underscores in the middle of a word, camelizing drops one underscore:

camelize('a___b_c') == 'A__bC'
camelize('a__b_c') == 'A_bC'
camelize('a_b_c') == 'ABC'
underscore('A__bC') == 'a___b_c'
underscore('A_bC') == 'a__b_c'
underscore('ABC') == 'a_b_c'

• For underscores at the start of a word, camelizing preserves the number of underscores and doesn't uppercase the first non-underscore letter

camelize('___a') == '___a'
camelize('__a') == '__a'
camelize('_a') == '_a'
underscore('___a') == '___a'
underscore('__a') == '__a'
underscore('_a') == '_a'