On Sat, 21 Feb 2026 00:34:11 GMT, Andy Goryachev <[email protected]> wrote:
>> In this specific test: `assertSameDouble(0.19999999999999998335, >> "0.19999999999999998335")` >> we find that: >> 1. The Java compiler interprets the literal (first argument) as >> 0.19999999999999998 >> 2. Both `Double.parseDouble()` and `CssNumberParser.parseDouble()` return >> the value 0.19999999999999998. >> >> So everyone agrees. In fact, everyone agrees for _all_ numbers where the >> significand can fit into 64 bits. For numbers with more digits than can fit >> in 64 bits, there is a difference: >> 1. The Java compiler and `Double.parseDouble()` convert with infinite >> precision, and then round to the nearest representable double. >> 2. CssNumberParser truncates to 64 bits, and then rounds to the nearest >> representable double. This almost always results in the same value, but I >> think there can be cases very close to the midpoint between two adjacent >> doubles that the result can be different by one ulp. > > thank you for clarifications! that's why I want to test the adjacent values > (+/- N ulps). > > I'll try to come up with some additional tests next week. I've changed the implementation slightly to fall back to `Double.parseDouble()` for values that don't fit into 64 bits. This means that `CssNumberParser` now always returns the same value as `Double.parseDouble()`. ------------- PR Review Comment: https://git.openjdk.org/jfx/pull/2069#discussion_r2838081130
