[PonyORM-list] Algebra of sets

[Vsevolod Novikov]

Hi Alexander,

> Is this inconvinience particularly one which you meant, or something
> other?
>
> Don't remember I said something like that, I spoke purely about
> performance of INTERSECT / UNION in SQL
>

I meant this "This way of filtering should works in the development version
of Pony (it was accidentally broken in the last release 0.4.8, but will be
restored in the upcoming release 0.4.9)", but I see now that the
development version works fine as described.

It will be non-intuitive if parameter name can be arbitrary when query has
> only one for-loop, but becomes non-arbitrary with multiple for-loops.


I didnt mean that.

Look here for example:

>>> d = {
...   'a':[1,2,3],
...   'b':[4,5],
...   'c':[6,7]
... }
>>> [(v,k) for k in d for v in d[k]]
[(1, 'a'), (2, 'a'), (3, 'a'), (6, 'c'), (7, 'c'), (4, 'b'), (5, 'b')]


You can see that the generator turns the first circle and then the second
one, but always returns only one item - the (v,k) pair - every turn. After
the evaluating of the generator the only this item remains available for
the following operations. So, filtering for example, the returned sequence
we may use only the item value in a filter expression:

>>> filter(lambda x:x[0]==2,[(v,k) for k in d for v in d[k]])
[(2, 'a')]


The same rule should be appied IMHO to the lambda-filter expression which
is used to filter the query: the only item returned in the query result may
be used. This particular item is an only lambda parameter of the
lambda-filter, so we are free to use any name for it.

Another syntax case should be applied IMHO if the filter is included
DIRECTLY into the generator like here:

>>> [(v,k) for k in d for v in d[k] if v != 1 and k != 'c']
[(2, 'a'), (3, 'a'), (4, 'b'), (5, 'b')]

The filter here may use any variable in the context of the generator and
this is usual python language syntax and semantic.

Regards,
Vsevolod
-------------- next part --------------
An HTML attachment was scrubbed...
URL: </ponyorm-list/attachments/20131025/82ccedde/attachment.html>