Given the input `a.b.c.d.e.f`, the previous code would call `_eval` 57 times:

```
1 a.b
2 a.b.c
3 a.b
4 a.b
5 a.b.c.d
6 a.b.c
7 a.b
8 a.b
9 a.b.c
10 a.b
11 a.b
12 a.b.c.d.e
13 a.b.c.d
14 a.b.c
15 a.b
16 a.b
17 a.b.c
18 a.b
19 a.b
20 a.b.c.d
21 a.b.c
22 a.b
23 a.b
24 a.b.c
25 a.b
26 a.b
27 a.b.c.d.e.f
28 a.b.c.d.e
29 a.b.c.d
30 a.b.c
31 a.b
32 a.b
33 a.b.c
34 a.b
35 a.b
36 a.b.c.d
37 a.b.c
38 a.b
39 a.b
40 a.b.c
41 a.b
42 a.b
43 a.b.c.d.e
44 a.b.c.d
45 a.b.c
46 a.b
47 a.b
48 a.b.c
49 a.b
50 a.b
51 a.b.c.d
52 a.b.c
53 a.b
54 a.b
55 a.b.c
56 a.b
57 a.b
```

The first optimization is to only call `expression._eval(…)` a single time per evaluation of a `AST_PropAccess`. We've already called it to determine if we're a nullish, so there's no reason to do it a second time in the unsafe path. This alone brings us down to 15 calls:

```
1 a.b
2 a.b.c
3 a.b
4 a.b.c.d
5 a.b.c
6 a.b
7 a.b.c.d.e
8 a.b.c.d
9 a.b.c
10 a.b
11 a.b.c.d.e.f
12 a.b.c.d.e
13 a.b.c.d
14 a.b.c
15 a.b
```

But, we can do even better by avoiding the evaluation mid–chain. There's no reason to evaluation unless the evaluation comes from the top of the chain (the rightmost property access). If anything can be optimized, it can optimize while we descend one time from that point. This brings us to 6 evaluations:

```
1 a.b
2 a.b.c.d.e.f
3 a.b.c.d.e
4 a.b.c.d
5 a.b.c
6 a.b
```

(I'm not sure why that first one happens)