Fix comparison of integers

[krlmlr]

Revert to default comparison also if x is integer.

Current behavior:

> testthat::expect_equal(0L, as.Date("2015-10-10Z"))
Error in `-.Date`(x, y) : can only subtract from "Date" objects
9: stop("can only subtract from \"Date\" objects")
8: `-.Date`(x, y)
7: format(x - y, digits = 3)
6: paste0(format(x, digits = 3), " - ", format(y, digits = 3), " == ",
       format(x - y, digits = 3))
5: compare.numeric(actual, expected, ...)
4: compare(actual, expected, ...)
3: condition(object)
2: expect_that(object, equals(expected, label = expected.label,
       ...), info = info, label = label)
1: testthat::expect_equal(0L, as.Date("2015-10-10Z"))

> testthat::expect_equal(0L, "0")
Error in x - y : non-numeric argument to binary operator
7: format(x - y, digits = 3)
6: paste0(format(x, digits = 3), " - ", format(y, digits = 3), " == ",
       format(x - y, digits = 3))
5: compare.numeric(actual, expected, ...)
4: compare(actual, expected, ...)
3: condition(object)
2: expect_that(object, equals(expected, label = expected.label,
       ...), info = info, label = label)
1: testthat::expect_equal(0L, "0")

[hadley]

Maybe a better check would be !identical(class(x), class(y)) ?

[krlmlr]

Currently:

> testthat::expect_equal(1L, 0)
Error: 1L not equal to 0
1 - 0 == 1

I guess this case (and others) will be handled differently if comparing classes of x and y.

[hadley]

I think that's probably reasonable - it doesn't really matter what the values are if the classes are different. But a bigger problem is expect_equal(1, 1L) which for convenience shouldn't fail. Maybe we should special case integer vs. numeric comparison?

This current behaviour is also bad:

expect_equal(1L, factor(1))
#  Error: 1L not equal to factor(1)
# 1 - 1 == NA In addition: Warning message:
# In Ops.factor(x, y) : ‘-’ not meaningful for factors

[hadley]

And this gives an uninformative message:

x <- structure(1L, class = "myclass")
expect_equal(1L, x)
#  Error: 1L not equal to x
# 1 - 1 == 0 

[krlmlr]

Updated: Added support for complex, fallback to string comparison if length different, otherwise use numeric comparison only for "true" numbers.

[krlmlr]

This probably conflicts with #313.

[krlmlr]

Merged with master.