Re: [PATCH v4 02/19] flake8: Enforce shorter line length for comments and docstrings

[John Snow <jsnow@redhat.com>]

On 4/9/21 5:33 AM, Markus Armbruster wrote:
> Daniel P. Berrangé <berrange@redhat.com> writes:
> 
>> On Fri, Mar 26, 2021 at 04:44:25PM +0000, Peter Maydell wrote:
>>> On Fri, 26 Mar 2021 at 16:33, John Snow <jsnow@redhat.com> wrote:
>>>> Being less terse about it: Mostly, I don't like how it enforces this
>>>> column width even for indented structures. Generally, we claim that 72
>>>> columns is "comfortable to read" and I agree.
>>>>
>>>>                                      However, when we start in a margin, I
>>>>                                      am not convinced that this is
>>>>                                      actually more readable than the
>>>>                                      alternative. We aren't using our full
>>>>                                      72 characters here.
>>>
>>> I agree, and I don't see any strong reason to hold our Python
>>> code to a different standard to the rest of our codebase as
>>> regards line length and comment standards.
>>
>> There's one small difference with python vs the rest of the codebase when
>> it comes to API doc strings specifically. eg we have a docstring API comment
>> in python/qemu/machine.py:
>>
>> class QEMUMachine:
>>      """
>>      A QEMU VM.
>>
>>      Use this object as a context manager to ensure
>>      the QEMU process terminates::
>>
>>          with VM(binary) as vm:
>>              ...
>>          # vm is guaranteed to be shut down here
>>      """
>>
>> This formatting, including line breaks, is preserved as-is when a user
>> requests viewing of the help:
>>
>>>>> print(help(qemu.machine.QEMUMachine))
>>
>> Help on class QEMUMachine in module qemu.machine:
>>
>> class QEMUMachine(builtins.object)
>>   |  QEMUMachine(binary: str, args: Sequence[str] = (), wrapper: Sequence[str] = (), name: Optional[str] = None, test_dir: str = '/var/tmp', monitor_address: Union[Tuple[str, str], str, NoneType] = None, socket_scm_helper: Optional[str] = None, sock_dir: Optional[str] = None, drain_console: bool = False, console_log: Optional[str] = None)
>>   |
>>   |  A QEMU VM.
>>   |
>>   |  Use this object as a context manager to ensure
>>   |  the QEMU process terminates::
>>   |
>>   |      with VM(binary) as vm:
>>   |          ...
>>   |      # vm is guaranteed to be shut down here
>>   |
>>   |  Methods defined here:
>>   |
>>
>>
>> IOW, while we as QEMU maintainers may not care about keeping to a narrow
>> line width, with API docstrings, we're also declaring that none of the
>> users of the python APIs can care either. These docstrings are never
>> reflowed, so they can end up wrapping if the user's terminal is narrow
>> which looks very ugly.
>>
>>
>> So this python API docstring scenario is slightly different from our
>> main codebase, where majority of comments are only ever going to be seen
>> by QEMU maintainers, and where C API doc strings don't preserve formatting,
>> because they're turned into HTML and re-flowed.
>>
>> Having said all that, I still don't think we need to restrict ourselves
>> to 72 characters. This is not the 1980's with people using text terminals
>> with physical size constraints. I think it is fine if we let python
>> docstrings get larger - especially if the docstrings are already indented
>> 4/8/12 spaces due to the code indent context, because the code indent is
>> removed when comments are displayed. I think a 100 char line limit would
>> be fine and still not cause wrapping when using python live help().
> 
> The trouble with long lines is not text terminals, it's humans.  Humans
> tend to have trouble following long lines with their eyes (I sure do).
> Typographic manuals suggest to limit columns to roughly 60 characters
> for exactly that reason[*].
> 
> Most doc strings are indented once (classes, functions) or twice
> (methods).  72 - 8 is roughly 60.
> 

My problem with this patch isn't actually the docstrings -- it's 
one-line comments.

If you can teach flake8 to allow this:

# Pretend this is a single-line comment that's 73 chars

but disallow this:

# Pretend this is a two-line comment that's 73 chars,
# and continues to a new line that's also pretty long,
# and maybe keeps going, too.

I will happily accept that patch. Without the ability to enforce the 
style though, I am reluctant to pretend that it's even a preference that 
we have. I think it's a waste to hunt down and re-flow single-line 
comments that just barely squeak over a limit. They look worse.

We can discuss this more when we go to propose a style guide for the 
Python folder; I think it's maybe a misprioritization of our energies in 
the present context.

(I still have the style guide on my TODO list, and even began writing a 
draft at one point, but I think we'd both like to press forward on the 
Typing bits first.)

> With nesting, doc strings can become indented more.  Nesting sufficient
> to squeeze the doc string width to column 72 under roughly 60 is pretty
> rare.  Going beyond 72 colums to keep such doc strings readable is
> exactly what PEP 8 wants you to do.
> 
> Again, I see no reason to deviate from PEP 8.
> 
> 
> [*] https://en.wikipedia.org/wiki/Column_(typography)#Typographic_style
>