bug#24450: [PATCHv2] Re: pypi importer outputs strange character series in optional dependency case.

[Maxim Cournoyer]

Hey Ricardo :-)

Ricardo Wurmus <address@hidden> writes:

> Hi Maxim,
>
>>>>    (call-with-input-file requires.txt
>>>>      (lambda (port)
>>>> -      (let loop ((result '()))
>>>> +      (let loop ((required-deps '())
>>>> +                 (test-deps '())
>>>> +                 (inside-test-section? #f)
>>>> +                 (optional? #f))
>>>>          (let ((line (read-line port)))
>>>> -          ;; Stop when a section is encountered, as sections contains 
>>>> optional
>>>> -          ;; (extra) requirements.  Non-optional requirements must appear
>>>> -          ;; before any section is defined.
>>>> -          (if (or (eof-object? line) (section-header? line))
>>>> +          (if (eof-object? line)
>>>>                ;; Duplicates can occur, since the same requirement can be
>>>>                ;; listed multiple times with different conditional 
>>>> markers, e.g.
>>>>                ;; pytest >= 3 ; python_version >= "3.3"
>>>>                ;; pytest < 3 ; python_version < "3.3"
>>>> -              (reverse (delete-duplicates result))
>>>> +              (map (compose reverse delete-duplicates)
>>>> +                   (list required-deps test-deps))
>>>
>>> Looks like a list of lists to me.  “delete-duplicates” now won’t delete
>>> a name that is in both “required-deps” as well as in “test-deps”.  Is
>>> this acceptable?
>>
>> It is acceptable, as this corner case cannot exist given the current
>> code (a requirement can exist in either required-deps or test-deps, but
>> never in both). It also doesn't make sense that a run time requirement
>> would also be listed as a test requirement, so that corner case is not
>> likely to exist in the future either.
>
> I mentioned it because I believe I’ve seen this in the past where the
> importer would return some of the same inputs as both regular inputs and
> test dependencies.

OK!

>>> Personally, I’m not a fan of using data structures for returning
>>> multiple values, because we can simply return multiple values.
>>
>> I thought the Guile supported multiple values return value would be
>> great here as well, but I've found that for this specific case here, a
>> list of lists worked better, since the two lists contain requirements to
>> be processed the same, which "map" can readily do (i.e. less ceremony is
>> required).
>
> “map” can also operate on more than one list at a time:
>
>     (call-with-values
>       (lambda ()
>         (values (list 1 2 3)
>                 (list 9 8 7)))
>       (lambda (a b) (map + a b)))
>
>     => (10 10 10)

That's what I meant by "requires more ceremony".  I can simply apply
"map" to the return value of the function and get what I need, rather
than having to use "values" in the callee, then "call-with-values" in
the caller and establish a binding for each list.
  
> Of course, it would be simpler to just use a single list of tagged
> items.

Do you feel strongly about it? I don't; I'm open to try to use a tagged
list if you feel this is worth it.

Maxim