set update in 2.5

[Duncan Smith]

Hello,
In moving from 2.4 to 2.5 I find that some of my unit tests are now
failing. I've worked out that the problem relates to the set update
method. In 2.4 I could update a set with an iterable type derived from
dict as the argument. I now find that the set is updated with the hash
values of the items in my iterable, rather than the items themselves.
Converting to a list first gets round the problem.

My iterable type has the same API as 'set' but requires items to have
hashable 'uid' attributes, so they can also be looked up by uid. I hope
this, and the fact that this worked fine in 2.4 will be enough to track
down the issue without me having to paste reams of code (other than the
following, pasted from IDLE). Any ideas? Cheers.

Duncan

from graphItems import Node
from keyed_sets import KeyedSet
n = Node(1)
n.uid 1
k = KeyedSet([n])
k KeyedSet([1])
type(iter(k).next())

type(k[1])

s = set()
s.update(list(k))
type(iter(s).next())

s = set()
s.update(k)
type(iter(s).next())

s = set()
s.add(n)
type(iter(s).next())

hash(n) 1

 

[Peter Otten]

In moving from 2.4 to 2.5 I find that some of my unit tests are now
failing. I've worked out that the problem relates to the set update
method. In 2.4 I could update a set with an iterable type derived from
dict as the argument. I now find that the set is updated with the hash
values of the items in my iterable, rather than the items themselves.
Converting to a list first gets round the problem.

My iterable type has the same API as 'set' but requires items to have
hashable 'uid' attributes, so they can also be looked up by uid. I hope
this, and the fact that this worked fine in 2.4 will be enough to track
down the issue without me having to paste reams of code (other than the
following, pasted from IDLE). Any ideas? Cheers.

Duncan

from graphItems import Node
from keyed_sets import KeyedSet
n = Node(1)
n.uid 1
k = KeyedSet([n])
k KeyedSet([1])
type(iter(k).next())

type(k[1])

s = set()
s.update(list(k))
type(iter(s).next())

s = set()
s.update(k)
type(iter(s).next())

s = set()
s.add(n)
type(iter(s).next())

hash(n) 1

Is your KeyedSet derived from dict?

$ python2.5
[snip].... def __iter__(self): return self.itervalues()
....

d = D(a=1, b=2)
set(d) set(['a', 'b'])

$ python2.4
[snip].... def __iter__(self): return self.itervalues()
....set([1, 2])


I think you should file a bug report. The fix is probably

--- setobject.c 2006-09-08 08:02:26.000000000 +0200
+++ setobject_new.c 2007-01-28 10:02:24.071688248 +0100
@@ -854,7 +854,7 @@
if (PyAnySet_Check(other))
return set_merge(so, other);

- if (PyDict_Check(other)) {
+ if (PyDict_CheckExact(other)) {
PyObject *value;
Py_ssize_t pos = 0;
while (PyDict_Next(other, &pos, &key, &value)) {

Peter

 

[Duncan Smith]

Duncan Smith wrote:

In moving from 2.4 to 2.5 I find that some of my unit tests are now
failing. I've worked out that the problem relates to the set update
method. In 2.4 I could update a set with an iterable type derived from
dict as the argument. I now find that the set is updated with the hash
values of the items in my iterable, rather than the items themselves.
Converting to a list first gets round the problem.

My iterable type has the same API as 'set' but requires items to have
hashable 'uid' attributes, so they can also be looked up by uid. I hope
this, and the fact that this worked fine in 2.4 will be enough to track
down the issue without me having to paste reams of code (other than the
following, pasted from IDLE). Any ideas? Cheers.

Duncan

from graphItems import Node
from keyed_sets import KeyedSet
n = Node(1)
n.uid
1

k = KeyedSet([n])
k
KeyedSet([1])

type(iter(k).next())

type(k[1])

s = set()
s.update(list(k))
type(iter(s).next())

s = set()
s.update(k)
type(iter(s).next())

s = set()
s.add(n)
type(iter(s).next())

hash(n)

1

Is your KeyedSet derived from dict?

Thanks Peter. It is derived from dict.

$ python2.5
[snip]

... def __iter__(self): return self.itervalues()
...

set(['a', 'b'])


$ python2.4
[snip]

... def __iter__(self): return self.itervalues()
...

set([1, 2])


I think you should file a bug report. The fix is probably

--- setobject.c 2006-09-08 08:02:26.000000000 +0200
+++ setobject_new.c 2007-01-28 10:02:24.071688248 +0100
@@ -854,7 +854,7 @@
if (PyAnySet_Check(other))
return set_merge(so, other);

- if (PyDict_Check(other)) {
+ if (PyDict_CheckExact(other)) {
PyObject *value;
Py_ssize_t pos = 0;
while (PyDict_Next(other, &pos, &key, &value)) {

Peter

I will file a report. Cheers.

Duncan