Re: Some questions about file handling + some other things

In message <[email protected]>,
Ralf Schlatterbeck writes:
>On Wed, Mar 24, 2021 at 09:31:42AM -0400, John P. Rouillard wrote:
>> >Looks good (without having tested anything) but I'm not sure to_raw is
>> >the correct method to use here in sorting.
>> 
>> I don't understand. sort_repr is the method I am using. to_raw is
>> just a (poorly??) chosen variable name.
>
>Ah. The hyperdb has "from_raw" methods, I thought there would be a
>"to_raw", too and that you did reuse that.

Nope, I sort of figured that might be the confusion. Glad we got it
sorted.

>> >It is also worth a regression test: If sorting of numbers and/or dates
>> >works in SQL (for everything but Multilink properties) I'm quite sure
>> >that it would break for anydbm which always sorts in python.
>> 
>> Yes, Probably needs more data in the db to actually test the ordering
>> across multiple items. Anydbm I'll bet is currently broken since it
>> would use a string sort order not an integer/number sort order in its
>> sorting calls.
>
>No, the regression-test tests that.

Well at the very least the test testMultilinkOrdering: "all backends
should return the Multilink numeric-id-sorted" is commented out.

Also are you sure that the sort test is sorting values >= 10? Under
that string sort and integer sort of 1...9 is the same. Can you point
me to an applicable test?

>But it may be that sorting values that can contain NULL values will not
>work.

Ok.

>> >From an implementation point of view, sorting null last requires a
>> value to replace null/None so sort(key=...) doesn't throw an error
>> trying to sort None against any other type. Without knowing something
>> about the property's possible values I don't see how the code can
>> choose a suitable value.
>
>I'm doing this explicitly for Postgres (and probably other backends),
>the generated SQL for filter calls looks like this:
>
>...
>order by (lower(_someclass._title) is not NULL),lower(_someclass._title)
>...
>
>So its perfectly fine to put NULL values first or last. And it doesn't
>need many changes to make it work.

Does that form the tuples like: (TRUE,"A real title"), (FALSE,None) So
you can sort FALSE first or last then subsort TRUE to get it ordered
by title?

>> Consider a multilink called surveys with a 'rating' property. We want
>> to display the links in rating order. 'rating' has values [-5,5] where
>> 0 is neutral. You want "no rating given"/unset to sort as 0. I am sure
>> there are other examples. Note if 'rating' had values [0,10] then 'no
>> rating' should be 5. If you wanted all 'no value given' to sort at the
>> end, you choose 11 as the 'no rating given" value.
>
>Hmm, in that case I would not allow NULL values. Should be set to 0 by
>an auditor.

Except no answer or N/A (not applicable) is a valid value represented
by None. You could use a -6 or +6 sentinal for N/A etc. But again you
can't sort first/last etc.

>> Thoughts?
>
>I'm reluctant to add an explicit property where you can sort NULL values
>into the middle of everything else. Doesn't feel right :-)

Yeah but that may be where the use case wants/needs them.

>Note that so far I was considering sorting by hyperdb.filter (and html
>wrapper classes) and you were considering only the multilink sorting as
>far as I understand.

I think you are correct. We could have two different methods for these
use cases. So nullfirst=True|False for the property setting in
hyperdb.filter and html wrapper classes.

Then for the case of None in the property of a mutilink item we could
set a value.

But I think that might be confusing.

--
				-- rouilj
John Rouillard
===========================================================================
My employers don't acknowledge my existence much less my opinions.