bug in inequality of words

[mantepse]

Using 7.4.beta 6:

sage: shape1 = Word([1,2], alphabet=[1,2])
sage: shape2 = Word([1,1], alphabet=[1,2])
sage: shape1 == shape2
False
sage: shape1 != shape2
False

[mantepse]

However:

sage: Word([1,1], alphabet=[1,2]) != Word([1,2], alphabet=[1,2])
True

???

[tscrim]

It seems to be an issue with WordDatatype_char.__richcmp__ in that if the cmp is non-zero, it needs to include 3 (which is !=):

         cdef int test = (<WordDatatype_char> self)._lexico_cmp(other)
         if test < 0:
             return op < 2 or op == 3
         elif test > 0:
-           return op > 3
+           return op >= 3
         else:
             return op == 1 or op == 2 or op == 5

[mantepse]

Thank you Travis! This worked, but I must say that I hate python.

[mantepse]

Can you explain why it fails only when assigning to a variable?

[slabbe]

Looking at the fix suggested by Travis, I believe the problem is not the variable but what is on the left side vs right side: A != B versus B != A.

I guess the problem comes from the recent upgrade of cmp to richcmp for Python 3 support done in #21435.

[slabbe]

Hmmm, looking at this further. I do not think this bug was included by #21435. It seems to come from the original implementation of WordDatatype_char.

I can confirm this using ​sagecell which runs 7.3:

shape1 = Word([1,2], alphabet=[1,2])
shape2 = Word([1,1], alphabet=[1,2])
print shape1 == shape2, "should be False"
print shape2 == shape1, "should be False"
print shape1 != shape2, "should be True"
print shape2 != shape1, "should be True"
print shape2 == shape2, "should be True"
print shape2 != shape2, "should be False"
print version()

returns

False should be False
False should be False
False should be True
True should be True
True should be True
False should be False
SageMath version 7.3, Release Date: 2016-08-04

[slabbe]

New commits:

​6b8ac18

21609: Fixing A!=B for words char when A > B

[mantepse]

Thanks for fixing. I have a very minor suggestion. Wouldn't it be more readable to code

        if test < 0:
            return op == 0 or op == 1 or op == 3
        elif test > 0:
            return op == 3 or op == 4 or op == 5
        else:
            return op == 1 or op == 2 or op == 5

[slabbe]

Yes, I agree. But, a little bit less efficient if op == 5 when test >0 for instance...

I must leave now to take a train, I won't have acces to internet until monday.

[tscrim]

I agree that it is a reasonable shortcut to sacrifice the slight bit of readability, but considering the comparison types are (arbitrarily) assigned integers, readability is already somewhat out the window...