Writing Documentation

Everything under Standard Providers and Localized Providers is automatically generated using sphinx.ext.autodoc which pulls docstrings from provider methods during the sphinx-build process. This also means that the docstrings must be written in valid reStructuredText.

Furthermore, because of the nature of this library, it is imperative to include sample usage to best demonstrate the capabilities and the possibilities. Since there are so many provider methods and localized versions, keeping the docs updated would have been a nightmare if the sample usage section (with reproducible output) of each provider method were to be written by hand.

Automating sample usage sections

To ease the burden of docs maintenance, the project takes advantage of docstring preprocessing offered by sphinx.ext.autodoc to automatically generate sample usage section, complete with reproducible output, all from a couple of lines of text using a :sample: “pseudo-role” like so:

:sample[ size=SIZE][ seed=SEED]:[ KWARGS]

What this will do is generate a sample usage section by calling the provider method SIZE times using an initial seed value of SEED with optional keyword arguments KWARGS. If no SIZE is specified or if SIZE is less than 5, it defaults to 5. If no SEED is specified, it defaults to 0.

For example, let us assume that the line :sample: is present in the docstring of a provider method named method1. That short line of text will automatically generate a sample usage section like this:

>>> Faker.seed(0)
>>> for _ in range(5):
...     fake.method1()
...
# Output 1
# Output 2
# Output 3
# Output 4
# Output 5

Depending on the nature of the provider method, the default of 5 samples may not be enough, so it is possible to increase that by using size=SIZE. You may also want to supply arguments to change the behavior of the method, so that can be done using KWARGS. Putting it all together, if we use :sample size=10: a=1, b=2, c=3, the sample usage section generated will look like this:

>>> Faker.seed(0)
>>> for _ in range(10):
...     fake.method1(a=1, b=2, c=3)
...
# Output 1
# Output 2
# Output 3
# Output 4
# Output 5
# Output 6
# Output 7
# Output 8
# Output 9
# Output 10

There may also be times when it is desirable to show a particular output, but the pseudo-RNG gets in the way, e.g. very low chance of said output being generated. To work around this, you may use seed=SEED to specify an initial seed value that is known to generate the desired output. If we specify :sample seed=12345: a=2, the sample usage section generated will look like this:

>>> Faker.seed(12345)
>>> for _ in range(5):
...     fake.method1(a=2)
...
# Output 1
# Output 2
# Output 3
# Output 4
# Output 5

You can mix and match SIZE, SEED, and KWARGS, and if KWARGS is becoming too long to fit a single line, you can break KWARGS into multiple lines in the same way you can break keyword arguments across multiple lines in actual Python code. For example, let us say the docstring contains this:

:sample size=25 seed=12345: arg1='very long value, unfortunately',
                            arg2='yet another long value'

The sample section usage generated will look something like this:

>>> Faker.seed(12345)
>>> for _ in range(25):
...     fake.method1(arg1='very long value, unfortunately', arg2='yet another long value')
...
# Output 1
# Output 2
# ...
# Output 24
# Output 25

Docstring preprocessing behavior

If a provider method does not have a docstring or if the docstring does not contain properly formatted :sample: lines, a default sample usage section will automatically be generated for the benefit of insufficiently documented provider methods.

A docstring may contain multiple :sample: lines, and all prospective :sample: lines are first checked to see if they are properly formatted. Malformed instances will be discarded, and details will be logged to the console as a warning. All properly formatted :sample: lines will then be removed from the docstring and will undergo sample validation and generation, and the resulting docstring will have an :examples: section appended to the end. In code form:

# Source code docstring
def foo():
    """Summary line

    Lorem ipsum dolor sit amet, consectetur adipiscing elit.
    Fusce auctor faucibus condimentum.

    :sample:

    Duis posuere lacinia porta.
    Quisque mauris nisl, mattis sed ornare eget, accumsan sit amet mauris.

    :sample size=10 seed=1000:
    """
    return 1
# Resulting docstring (more or less) after preprocessing
def foo():
    """Summary line

    Lorem ipsum dolor sit amet, consectetur adipiscing elit.
    Fusce auctor faucibus condimentum.


    Duis posuere lacinia porta.
    Quisque mauris nisl, mattis sed ornare eget, accumsan sit amet mauris.

    :examples:

    >>> Faker.seed(0)
    >>> for _ in range(5):
    ...     fake.foo()
    ...
    1
    1
    1
    1
    1

    >>> Faker.seed(1000)
    >>> for _ in range(10):
    ...     fake.foo()
    ...
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    """
    pass

Notice how it did not remember where the :sample: lines are. Regardless of the original positions of the :sample: lines, the resulting output of all those lines will be collected and appended towards the end of the docstring. Please keep this in mind when structuring the flow of docstrings.

There are definitely benefits in allowing sample sections to be generated in place as it make the creation of richer documentation possible, but unfortunately it is not yet possible due to time constraints. Until that feature is available, please keep all :sample: lines towards the end of the docstring to help out the code reviewers.

Sample validation and security segue

Under the hood, the sample sections are generated by feeding the parsed docstring sample lines into the standard library’s eval(). This setup most definitely have some security implications out of the box, and this is why :sample: lines undergo validation prior to generation.

There are many details behind the validation process, but the long and short of it is that SIZE and SEED can only be integers, and KWARGS can only be keyword arguments with literal values or OrderedDict objects. Attempting to do anything else like calling other builtins or even just performing basic arithmetic will fail the validation. Details of failed validation will be logged to the console as a warning.

To further improve security, all of the potentially dangerous code used for this purpose have been isolated into the faker.sphinx module, and this module will be excluded from release distributions that are hosted in PyPI.

If you are interested in learning more or in performing a security audit on how sample validation is implemented, please refer to the source code and docstrings of faker.sphinx.validator.SampleCodeValidator and faker.sphinx.docstring.ProviderMethodDocstring.

Sample generation

Once a :sample: line has been validated, the sphinx-build process will attempt to generate results based on the information provided. A sample run can still fail if KWARGS contains keyword arguments that the provider method is not expecting or if executing the provider method results in an exception. Details of such instances will also be logged to the console as a warning.