On Wed, Mar 27, 2024, at 17:52, Bruno Haible wrote:
> Collin Funk wrote:
>> I have also made a conscious choice not to use more abstract
>> collections like 'Iterable' even though they are recommended [4].
>>
>> The first is because some functions check the type:
>>
>> def setModules(self, modules: list[str] | tuple[str]) -> None:
>> '''Set the modules list.'''
>> if type(modules) is list or type(modules) is tuple:
>> # do something, or raise a TypeError
>>
>> It is temping to use Iterable here:
>>
>> print(iter(list()))
>> <list_iterator object at 0x7f7c13f5b1c0>
>> print(iter(tuple()))
>> <tuple_iterator object at 0x7f7c13f5b1c0>
>>
>> but a what about a set?
>>
>> print(iter(set()))
>> <set_iterator object at 0x7f7c13f56b80>
>>
>> We would think it is okay to pass a set, but it would fail at runtime.
>> Using the uglier Union would let us know by looking or a warning if
>> using a lsp program.
>>
>> The second reason is sorting. Since it is easy to mess up the sorting
>> for the test cases against gnulib-tool.sh, it is probably best to only
>> pass lists for now. We can change functions to Iterable or something
>> similar once we are sure it will not affect the output.
>
> I agree. My main considerations are:
>
> - Can someone with little exposure to Python understand and modify the
> code? When you write 'list' and the developer knows that 'list' means
> what is known as "array" in other programming languages, they can work
> with the code. Using the abstract interfaces that Python has does not
> help in this aspect.
Hi Bruno,
I would argue against this line of thinking. Sure, keep it simple, but don't
reduce the subset of Python available only to something that is easily known
to, for e.g. a C developer. I made the same decision when writing the testsuite
for Wget almost 10 years ago, and in hindsight I regret it.
The thing is, when you restrict it something that can be understood by someone
with little exposure to Python, you end up with a codebase that is alien to
both groups, the ones that know Python and the ones that don't. You get a
codebase that neither enjoy working with and want to maintain.
Instead, I would argue that following the standard Python best practices is a
better choice. The person maintaining the script needs to be a Python
developer anyways. Might as well make it easy for them
>
> - In some cases, it is useful to access the data structure with indices.
> Then an 'Iterable' will not suffice, a 'list' or 'tuple' is needed.
> What would be the point of declaring "this function needs to traverse
> the given data only from the beginning to the end, never in the opposite
> direction, never with random indices"? There is no advantage. It is
> useless information. So, why bother?
>
> Bruno
