[SciPy-dev] Add bz2 support to io.mmio [not tested] <--

Pearu Peterson pearu@cens.ioc...
Thu Apr 10 15:33:07 CDT 2008


On Thu, April 10, 2008 10:57 pm, Stéfan van der Walt wrote:
....
>> While I strongly agree with the importance of unittests, we should not
>>  be too bureaucratic too, imho, because it can crush down creativity
>>  which is also important. So, we need to move forward along a middle
>>  ground.
>>
>>  In this particular case, I'd say, you have been looking for a problem
>>  that was not there. :)
>
> I can't really go along with any middle ground on testing, I'm afraid
> (do I sound like a broken record yet?).  To me that reeks of bad
> software development practice.  If it were up to me, there wouldn't be
> a single line of untested code in NumPy.
>
> More and more people depend on NumPy to do their work and research,
> and we have a duty towards them to produce reliable, high quality
> software.
>
> Your change consists of only 5 lines of code, but what happens when
> the bz2 module changes its calling convention in Python 3k?  We'd
> happily ship SciPy without even knowing that the compressed matrix
> market loading was completely unusable.

That's a convincing point for me.

> I sincerely hope that I am overreacting, and that there is nothing to
> be concerned about, but as it stands I see a very real problem here.

You are right, of course. Please keep hammering us to write unittests
for any new code.

Thanks,
Pearu



More information about the Scipy-dev mailing list