`IntMap` `isProperSubmapOfBy` test failure about containers HOT 4 CLOSED

This is a bug in isProperSubmapOfBy.

When the first tree is in the left/right child of the second tree, and the first tree and the left/right child compare equal, the EQ should be updated to LT since it's a proper submap.

I tried to trace the source of the error and looks like it has been there since the commit which introduced IntMap bbbba97.

from containers.

whoa, indeed.

when translating the textual description to a property, with leancheck I get

ghci> checkFor 100000 $ \ f a b -> I.isProperSubmapOfBy f a b == (S.isProperSubsetOf  (I.keysSet a ) (I.keysSet b) && and (I.intersectionWith f a b))
*** Failed! Falsifiable (after 4652 tests):
(\_ _ -> True) (fromList [(0,()),(1,())]) (fromList [(-1,()),(0,()),(1,())])

Also,

ghci> I.isProperSubmapOf (I.fromList [(0,()),(1,())]) (I.fromList [(-1,()),(0,()),(1,())])
False

from containers.

This bug was fixed for IntSet in 71fe15f GHC #1762. Unfortunately they missed IntMap.

from containers.

That's an incredibly old bug. Wow.

from containers.