from_str_radix incorrectly parses out-of-range repeating constants...

[huonw]

...when the radix is a power of 2. e.g. 100 1's is an invalid u64/i64 in any base, but for powers of 2 it gives a "correct" parse:

static hundred_ones: &'static str  = "1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111";

fn main() {
    for uint::range(2, 37) |i| {
        io::println(
            fmt!("%u %? %?", i,
                 u64::from_str_radix(hundred_ones, i),
                 i64::from_str_radix(hundred_ones, i)));
    }
}

outputs

2 Some(18446744073709551615) None
3 None None
4 Some(6148914691236517205) Some(6148914691236517205)
5 None None
6 None None
7 None None
8 Some(10540996613548315209) None
9 None None
10 None None
11 None None
12 None None
13 None None
14 None None
15 None None
16 Some(1229782938247303441) Some(1229782938247303441)
17 None None
18 None None
19 None None
20 None None
21 None None
22 None None
23 None None
24 None None
25 None None
26 None None
27 None None
28 None None
29 None None
30 None None
31 None None
32 Some(1190112520884487201) Some(1190112520884487201)
33 None None
34 None None
35 None None
36 None None

[Kimundi]

The problem here is, I think, that the code that attempts to detect overflows only does so by detecting if the accumulator grows in the wrong direction, which only covers overflow for signed values. For it to detect this condition here, it also needs to check if the accumulator does not change, even though it should, or more generally, if it changes by a smaller value than it is supposed to.

The reported behavior happens because, after the algorithm reads in the first 64 1s, this happens for the accumulator at each new digit:

        1111...1111
*= 2 -> 1111...1110
+= 1 -> 1111...1111

so it never sees that the value gets smaller

[huonw]

I agree that that is the issue, @Kimundi.

I have a fix, along with one for #5544.