Differences between revisions 4 and 13 (spanning 9 versions)
⇤ ← Revision 4 as of 2012-02-16 20:04:27 →
Size: 1083
Editor: vdelecroix
Comment:
← Revision 13 as of 2017-03-13 22:32:31 → ⇥
Size: 3264
Editor: hivertwiki
Comment:
Deletions are marked like this. Additions are marked like this.
Line 3: Line 3:
Equality and Coercion could be harmful
======================================		 Equality using Coercion considered harmful?
===========================================
Line 6: Line 6:
The goal of this page is to gather all problems due to equality accepting coercion in borderline cases:: Currently, Sage specifies that, upon comparing with ``a==b`` two objects `a` and `b` that do not have the same parent, a coercion is attempted to put them in the same parent. To prepare a discussion on whether that specification is viable in the long run, we are gathering here (borderline?) use cases where this behaviour is harmful. (Note that there is also the related problem of broken notions of equality for elements with the same parent which is discussed at EqualityUnhashable_.)

::
Line 14: Line 16:
    sage: dd = {3.1415926535897932: 'approx', pi: 'exact'}
    sage: dd
However Python 2.7 documentation specifies::

    object.__hash__(self)
    Called by built-in function hash() and for operations on members of hashed collections including set,
    frozenset, and dict. __hash__() should return an integer. The only required property is that objects
    which compare equal have the same hash value;

As a first consequence of the above behavior::

    sage: {3.1415926535897932: 'approx', pi: 'exact'}
Line 20: Line 31:
as a consequence:: And also::
Line 33: Line 44:

pi is an integer (and transitivity of equality)
-----------------------------------------------

::

    sage: R1 = RealField(2)
    sage: R2 = RealField(56)
    sage: pi1 = R1(pi); pi1
    3.0
    sage: pi2 = R2(pi); pi2
    3.141592653589793
    sage: pi1 == pi2
    True
    sage: bool(pi == pi2)
    True
    sage: pi1 == 3
    True

And we get a nice proof that pi = 3.
Line 47: Line 78:
An example showing that Sage's specifications clashes with Python's specifications
----------------------------------------------------------------------------------

::

    sage: S = SymmetricFunctions(QQ)
    sage: x = S.s()[5]
    sage: y = S.p()(x)
    sage: x == y
    True
    sage: hash(x), hash(y)
    (-1840429907820881728, 5178019317311573726)

It's surely syntactically nice to have x == y evaluate True after a
coercion. However enforcing that the two hash functions be the same
would be simply impossible: this would force to systematically coerce
any symmetric function to some fixed base for computing the hash
function, and we just can't afford that.

Yet another example (#22594)

::

    sage: bring=ZZ
    sage: for a in cartesian_product([[bring.one()]]):
    ....: print a[0].parent()
    ....:
    Integer Ring
    sage: bring=QQ
    sage: for a in cartesian_product([[bring.one()]]):
    ....: print a[0].parent()
    ....:
    Integer Ring

Equality using Coercion considered harmful?

Currently, Sage specifies that, upon comparing with a==b two objects a and b that do not have the same parent, a coercion is attempted to put them in the same parent. To prepare a discussion on whether that specification is viable in the long run, we are gathering here (borderline?) use cases where this behaviour is harmful. (Note that there is also the related problem of broken notions of equality for elements with the same parent which is discussed at EqualityUnhashable.)

sage: bool(pi == 3.14159265358979323)
True
sage: hash(pi)
2943215493
sage: hash(3.14159265358979323)
1826823505

However Python 2.7 documentation specifies:

object.__hash__(self)
Called by built-in function hash() and for operations on members of hashed collections including set,
frozenset, and dict. __hash__() should return an integer. The only required property is that objects
which compare equal have the same hash value;

As a first consequence of the above behavior:

sage: {3.1415926535897932: 'approx', pi: 'exact'}
{3.1415926535897932: 'approx', pi: 'exact'}
sage: {0:"exact", 0.0000000000000000000:"approx"}
{0: 'approx'}

And also:

sage: pii = 3.14159265358979323
sage: bool(pii == pi)
True
sage: dd = {pi: "exact"}
sage: pi in dd
True
sage: pii in dd
False
sage: pii in dd.keys()
True

pi is an integer (and transitivity of equality)

sage: R1 = RealField(2)
sage: R2 = RealField(56)
sage: pi1 = R1(pi); pi1
3.0
sage: pi2 = R2(pi); pi2
3.141592653589793
sage: pi1 == pi2
True
sage: bool(pi == pi2)
True
sage: pi1 == 3
True

And we get a nice proof that pi = 3.

More strange consequences when using UniqueRepresentation

sage: F1 = FiniteEnumeratedSet([0.000000])
sage: F2 = FiniteEnumeratedSet([0])
sage: F1 is F2
True
sage: F2.list()
[0.000000000000000]
sage:

An example showing that Sage's specifications clashes with Python's specifications

sage: S = SymmetricFunctions(QQ)
sage: x = S.s()[5]
sage: y = S.p()(x)
sage: x == y
True
sage: hash(x), hash(y)
(-1840429907820881728, 5178019317311573726)

It's surely syntactically nice to have x == y evaluate True after a coercion. However enforcing that the two hash functions be the same would be simply impossible: this would force to systematically coerce any symmetric function to some fixed base for computing the hash function, and we just can't afford that.

Yet another example (#22594)

sage: bring=ZZ
sage: for a in cartesian_product([[bring.one()]]):
....:     print a[0].parent()
....:
Integer Ring
sage: bring=QQ
sage: for a in cartesian_product([[bring.one()]]):
....:     print a[0].parent()
....:
Integer Ring

EqualityCoercion (last edited 2017-03-13 22:32:31 by hivertwiki)