# (In)equality semantics of Coords are confusing

XMLWordPrintable

#### Details

• Type: Improvement
• Status: Done
• Resolution: Done
• Fix Version/s: None
• Component/s:
• Labels:
None
• Story Points:
1
• Sprint:
Science Pipelines DM-S15-1
• Team:
Data Release Production

#### Description

Viz:

 In [1]: from lsst.afw.coord import Coord In [2]: c1 = Coord("11:11:11", "22:22:22") In [3]: c1 == c1, c1 != c1 Out[3]: (True, False) In [4]: c2 = Coord("33:33:33", "44:44:44") In [5]: c1 == c2, c1 != c2 Out[5]: (False, True) In [6]: c3 = Coord("11:11:11", "22:22:22") In [7]: c1 == c3, c1 != c3 Out[7]: (True, True)

c1 is simultaneously equal to and not equal to c3!

#### Activity

No builds found.
John Swinbank created issue -
Field Original Value New Value
Description Viz:

{code}
In [1]: from lsst.afw.coord import Coord
In [2]: c1 = Coord("11:11:11", "22:22:22")
In [3]: c1 == c1, c1 != c1
Out[3]: (True, False)
In [4]: c2 = Coord("33:33:33", "44:44:44")
In [5]: c1 == c2, c1 != c2
Out[5]: (False, True)
In [6]: c3 = Coord("11:11:11", "22:22:22")
In [7]: c1 == c3, c1 != c3
Out[7]: (True, True)
{code}

c1 is simultaneously equal to *and* not equal to c3!
Viz:

{code}
In [1]: from lsst.afw.coord import Coord
In [2]: c1 = Coord("11:11:11", "22:22:22")
In [3]: c1 == c1, c1 != c1
Out[3]: (True, False)
In [4]: c2 = Coord("33:33:33", "44:44:44")
In [5]: c1 == c2, c1 != c2
Out[5]: (False, True)
In [6]: c3 = Coord("11:11:11", "22:22:22")
In [7]: c1 == c3, c1 != c3
Out[7]: (True, True)
{code}

{{c1}} is simultaneously equal to *and* not equal to {{c3}}!
Hide
John Swinbank added a comment -

This issue is, of course, that we provide an operator==() for Coord, which compares by value, but no corresponding operator!=(), so that the check for inequality is done by identity.

An easy fix is just to add something along the lines of:

 inline bool lsst::afw::coord::Coord::operator!=(lsst::afw::coord::Coord const &rhs) const {  return !operator==(rhs); }

Jim Bosch, you know this code better than I do – any reason not to go ahead and do just that?

Show
John Swinbank added a comment - This issue is, of course, that we provide an operator==() for Coord , which compares by value, but no corresponding operator!=() , so that the check for inequality is done by identity. An easy fix is just to add something along the lines of: inline bool lsst::afw::coord::Coord::operator!=(lsst::afw::coord::Coord const &rhs) const { return !operator==(rhs); } Jim Bosch , you know this code better than I do – any reason not to go ahead and do just that?
Hide
Paul Price added a comment - - edited

I think you want to do:

 %include "lsst/p_lsstSwig.i" %useValueEquality(lsst::afw::coord::Coord);

Although I've noticed that this macro (and %usePointerEquality) contain:

  try:  return self._eq_impl(rhs)  except Exception:  return NotImplemented

i.e., in the event the equality throws an exception, they're returning a class (which always evaluates to True) instead of raising an exception instance.

Show
Paul Price added a comment - - edited I think you want to do: %include "lsst/p_lsstSwig.i" %useValueEquality(lsst::afw::coord::Coord); Although I've noticed that this macro (and %usePointerEquality) contain: try: return self._eq_impl(rhs) except Exception: return NotImplemented i.e., in the event the equality throws an exception, they're returning a class (which always evaluates to True ) instead of raising an exception instance.
Hide
Paul Price added a comment -

Created DM-2348 for that.

Show
Paul Price added a comment - Created DM-2348 for that.
Hide
John Swinbank added a comment -

The macro is neat, but isn't comparing coordinates in C++ potentially useful as well?

Show
John Swinbank added a comment - The macro is neat, but isn't comparing coordinates in C++ potentially useful as well?
Hide
Jim Bosch added a comment - - edited

I think we do want to add the C++ operator!=, but we should be using %useValueEquality here too to get the right Python behavior when comparing Coords to other types. I'm not sure yet whether this has revealed a problem in %useValueEquality, but I'll figure that out on DM-2348.

Show
Jim Bosch added a comment - - edited I think we do want to add the C++ operator!= , but we should be using %useValueEquality here too to get the right Python behavior when comparing Coords to other types. I'm not sure yet whether this has revealed a problem in %useValueEquality , but I'll figure that out on DM-2348 .
Hide
John Swinbank added a comment -

The downside to that is that %useValueEquality delegates to the C++ operator== and just negates the result when checking for inequality. By also defining an operator!= in C++ we could quite easily end up with different semantics in C++ and Python, which seems unfortunate.

(Of course, if somebody wants to define an operator!= which isn't just the complement of operator== they should probably think again, but...)

Show
John Swinbank added a comment - The downside to that is that %useValueEquality delegates to the C++ operator== and just negates the result when checking for inequality. By also defining an operator!= in C++ we could quite easily end up with different semantics in C++ and Python, which seems unfortunate. (Of course, if somebody wants to define an operator!= which isn't just the complement of operator== they should probably think again, but...)
Hide
Jim Bosch added a comment -

I suppose we could essentially require the existence of operator!= in C++ by having %useValueEquality delegate to that for __ne__.

Show
Jim Bosch added a comment - I suppose we could essentially require the existence of operator!= in C++ by having %useValueEquality delegate to that for __ne__ .
 Epic Link DM-1910 [ 15942 ]
 Story Points 1
Hide
John Swinbank added a comment -

I watched an entertaining rant by Alex Stepanov over the weekend, to the effect that it's basically always a design flaw if !(thing == other) is not equivalent to thing != other. I'm persuaded that I like that solution best, iff it doesn't introduce too much overhead – I'm not sure how many places we're using %useValueEquality, and I don't think it's worth spending a long time checking them down. I made the change as suggested, fixed a couple of places where I know it'll break, and submitted to buildbot – let's see what fails.

Show
John Swinbank added a comment - I watched an entertaining rant by Alex Stepanov over the weekend, to the effect that it's basically always a design flaw if !(thing == other) is not equivalent to thing != other . I'm persuaded that I like that solution best, iff it doesn't introduce too much overhead – I'm not sure how many places we're using %useValueEquality , and I don't think it's worth spending a long time checking them down. I made the change as suggested, fixed a couple of places where I know it'll break, and submitted to buildbot – let's see what fails.
Hide
John Swinbank added a comment -

It turns out that it took only a few lines of changes in utils and afw to delegate Python's __ne__ to C++ and then provide operator!= everywhere that it's required. Given that, I think this is the approach I prefer.

Jim, the changes are on the tickets/DM-2347 branch on utils and afw. Please check that you're happy not just with those changes but with the approach adopted. Thanks!

Show
John Swinbank added a comment - It turns out that it took only a few lines of changes in utils and afw to delegate Python's __ne__ to C++ and then provide operator!= everywhere that it's required. Given that, I think this is the approach I prefer. Jim, the changes are on the tickets/ DM-2347 branch on utils and afw . Please check that you're happy not just with those changes but with the approach adopted. Thanks!
 Reviewers Jim Bosch [ jbosch ] Status To Do [ 10001 ] In Review [ 10004 ]
Hide
Jim Bosch added a comment -

Looks good to me. I think I agree that's the best approach (I certainly agree that everything that implements == should implement != in C++; I'm ambivalent as to how we enforce that).

Show
Jim Bosch added a comment - Looks good to me. I think I agree that's the best approach (I certainly agree that everything that implements == should implement != in C++; I'm ambivalent as to how we enforce that).
 Status In Review [ 10004 ] Reviewed [ 10101 ]
 Resolution Done [ 10000 ] Status Reviewed [ 10101 ] Done [ 10002 ]

#### People

Assignee:
John Swinbank
Reporter:
John Swinbank
Reviewers:
Jim Bosch
Watchers:
Jim Bosch, John Swinbank, Paul Price