Repository navigation
Add fractional float cases at the int64/uint64 boundaries #430
Description
Per the ADR, whole floats in [-2^63, 2^64-1] must be normalized to CBOR integers before hashing. The "Fractional operator with floats limits" scenario doesn't cover the edges of that range; we've found bugs at these boundaries in several implementations that the suite didn't catch.
Suggested Float rows (buckets computed with the Go reference, including the fix in open-feature/flagd#2011):
| hashing_input | Expected encoding | value |
|---|---|---|
-9223372036854775808.0 (-2^63) |
int64 | bucket3 |
9223372036854775808.0 (2^63) |
uint64 | bucket16 |
1e19 |
uint64 | bucket14 |
18446744073709549568.0 (largest double < 2^64) |
uint64 | bucket16 |
18446744073709551616.0 (2^64) |
float | bucket15 |
Activity
toddbaert commentedon Oct 7, 2026
cc @m-olko @NeaguGeorgiana23
m-olko commentedon Oct 8, 2026
At the beginning we wanted to add much more tests around those limits, not only float but also ints, but we had issues when trying to use integers in this range due to the differences between languages parsing json structures. Because of that we dropped this idea for now, until unification of handling numbers in the [2^53, 2^63] happens. This addition by limiting to only float makes a lot of sense. Thanks!