[Python-Dev] Is raising SystemError during relative import the best solution?

[Brett Cannon]

On Thu, Jan 8, 2009 at 11:33, Guido van Rossum <guido at python.org> wrote:
> On Thu, Jan 8, 2009 at 11:25 AM, Brett Cannon <brett at python.org> wrote:
>> So it turns out that if you try to do a relative import where a parent
>> is not loaded, it raises a SystemError. This has been in there since
>> Guido added package support back in the day. But this seems more like
>> an ImportError than a SystemError to me. My guess is that the original
>> purpose was to signify someone specified some relative import name
>> without the proper stuff to make the name resolve to what it should
>> be. But that to me is still an ImportError as the name came out wrong,
>> not that the system did something incorrectly.
>>
>> So I would like to propose to remove the SystemError and make it an
>> ImportError. Anyone object?
>
> Hm. The SystemError is because this is a logical impossibility -- how
> could you be doing an import (relative or otherwise) from P.M when P
> is not loaded? It could only happen if somebody has been removing
> stuff selectively from sys.modules. Why don't you want this to be a
> SystemError?
>

Doesn't fit my mental model as nicely as raising ImportError. And as
an aside this also makes testing a little bit more of a pain, but that
is not the reason I brought this up. But the way you phrased it makes
sense for me not to care enough to press this any farther.

-Brett