multiunion error on C but pass on python

[xueyumusic]

I met a problem about multiunion function that when using default C implementation it reports error but could work correctly when using python implementation... For example

fwd_index = OOBTree()
tp = (1,2,3)
fwd_index["a"] = tp

IF.multiunion(self._fwd_index.values(("a",))

[xueyumusic]

It seems that C implement does not support tuple value type, but python implement could compatible with tuple...

[tseaver]

I'm not clear what issue you are reporting. Can you show a traceback?

[xueyumusic]

The issue I want to report could be shown as this example. If I use C implement, it reports an error, however if I use Python implement, it is right

Use C multiunion:

In [1]: import BTrees

In [2]: from BTrees.OOBTree import OOBTree

In [3]: from BTrees.IFBTree import IFSet

In [4]:

In [4]:  t=OOBTree()

In [5]: t["a"]=(1,2,3)

In [6]: t["b"]=IFSet((4,5,6))

In [7]: BTrees.family32.IF.multiunion(t.values())
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-7-b83097e5a1a2> in <module>()
----> 1 BTrees.family32.IF.multiunion(t.values())

TypeError: invalid argument

Use Python multiunion:

In [1]: import BTrees

In [2]: from BTrees.OOBTree import OOBTree

In [3]: from BTrees.IFBTree import IFSet

In [4]:

In [4]: t=OOBTree()

In [5]: t["a"]=(1,2,3)

In [6]: t["b"]=IFSet((4,5,6))

In [7]: BTrees.family32.IF.multiunion(t.values())
Out[7]: BTrees.IFBTree.IFSet([1, 2, 3, 4, 5, 6])

[jamadden]

Looks to me like the C implementation is "right". But I think that's undefined behavior (what is a union of a typed, ordered set, and a random tuple even supposed to mean?) so don't be surprised at either outcome, or if it formats your hard drive.

[jamadden]

The interface definition for multiuniion explicitly spells out which types can be used as arguments (sets and tree sets, buckets and btrees, single integers).

The online help, however, simply says "Each [argument] must be an integer set, or convertible to one via the set iteration protocol." Now, it's talking about the internal set iteration protocol using the C SetIteration struct, but it's very reasonable to imagine that it means "anything you can use as a constructor to a set" e.g., "any iterable of integers."

I don't think it would be hard to extend the internal SetIteration protocol to support arbitrary iterables of integers.