Changed behavior of QueryResult

[andreymal]

This change broke my website:

-class QueryResult(list):
+class QueryResult(object):

0.7.3:

>>> [7] + select(p for p in Person if p.age > 20)[:]
[7, Person[2], Person[3]]

0.7.5:

>>> [7] + select(p for p in Person if p.age > 20)[:]
TypeError: can only concatenate list (not "QueryResult") to list

If this is a bug, please fix this.

If this is a feature, please update the documentation and the changelog.

[andreymal]

There's also another case: random.shuffle(query_result)

Not sure if it should work by design, but it works in 0.7.3 :)

(In general, class QueryResult is not documented, it would be good if it will mentioned in the documentation)

[kozlovsky]

Hi, thanks for reporting. This is a bug, and we will fix it

[kozlovsky]

QueryResult now is indeed not a subclass of list. The reason for this change is the following: in 0.7.3, when someone write:

for obj in query:
    ...

iter(query) immediately fetches query result.

But now it is possible to base a new query on a previous query:

query2 = select(x for x in query)

when query is used as a source of a new generator, Python immediately takes iter(query) even before the select function is called. In 0.7.3 it forces query execution. But we don't want to execute first query, it just a part of a final query.

The solution that we use in 0.7.4 is to make query result iterator lazy, and actually execute the query only when iterator.next is called the first time.

That is to say, the result of query[:] is still fetched immediately, while iter(query) is lazy now. Also, query.limit(x) now returns query instead of a query result.

And with this changes it is more clear not to base potentially lazy QueryResult on list, but to have a separate class for it.

To fix the problem that you described we need to add

For compatibility we can add QueryResult.__add__ which allows adding QueryResult to list. But I'm against adding mutable methods __iadd__, __setitem__, __delitem__, append, clear, extend, insert, pop, remove, reverse, sort, because mutating query result looks strange. So, the code like random.shuffle(query_result) will not work. You will need to replace it with:

query_result = list(query_result)
random.shuffle(query_result)

[kozlovsky]

Now QueryResult is not based on list, but expressions like

[7] + select(p for p in Person if p.age > 20)[:]

should work again.

Also, I added .sort(), .reverse() and .shuffle() methods directly to QueryResult

For more complex mutations of QueryResult you need to convert it to list first, either with list(query_result), or by calling query_result.to_list()