Ugh, haha

Why is this sufficient? Can't the contents be the same even if the paths are different?

I think you can only fastpath if the paths are the same, otherwise you have to fall back to data comparison.

This is already queued, but I think we need to drop it if I'm right here.

Ryan and I talked offline -- but surprisingly, the default filectx.cmp function only compares contents:

1:~/1$ repo['.']['A'].cmp(repo['.']['B'])
Out[1]: False

2:~/1$ repo['.']['A'].cmp(repo['.']['A'])
Out[2]: False

3:~/1$ repo['.']['A'].cmp(repo['.']['C'])
Out[3]: True

(here, A and B have the same content).

And filecmp seems to behave the same way:

7:~/1$ filecmp.cmp('A', 'B')
Out[7]: True

8:~/1$ filecmp.cmp('A', 'A')
Out[8]: True

9:~/1$ filecmp.cmp('A', 'C')
Out[9]: False

That doesn't mean that it's wrong, but I think it's consistent.

Ah, but some experimenting revealed that filecmp does follow symlinks, but our filectx comparators do not. Otherwise, both filecmp and filectx.cmp ignore any flag differences.

Thus, you can demonstrate a discrepancy:

A contains "foo"
real_A contains "A"
sym_A link to A

repo['.']['real_A'].cmp(repo['.']['sym_A']) # claims the same
filecmp.cmp('real_A', 'sym_A') # claims a difference, because "foo" != "A"

Note this simpler case doesn't trigger the discrepancy, because the linked file is otherwise identical:

A contains "A"
sym_A link to A

repo['.']['A'].cmp(repo['.']['sym_A']) # claims the same
filecmp.cmp('A', 'sym_A') # claims the same

The easiest fix is just to skip the fast-comparison path if either side is a symlink, and that's what I'll do unless others have other ideas. (@ryanmce thought we should think about what this API _should_ do).