[ruby-core:118783] [Ruby master Bug#20662] pack("g") completely discards any actual NaN value and always packs the same single-precision bytes for a NaN

3 Aug 2024

      Issue #20662 has been updated by nobu (Nobuyoshi Nakada).

Description updated

cabo (Carsten Bormann) wrote:
...
pack("G")/unpack("G") works great with NaN values. However,
First, Ruby provides only `Float::NAN`, and does not consider payloads of NaN values.
...
- `pack("g")` completely discards any actual NaN value and always packs the same bytes for a NaN
  ("bug as implemented" in `VALUE_to_float`)
Since `"G"` preserves the payload, it may be better to preserve it for `"g"` as well, IMO.
...
- `unpack("g")` always sets the quiet bit to 1 in the `Float` result
  (location of bug not obvious to me)
Ruby's `Float` uses `double` always.

Although I don't remember about NaN and its conversion in IEEE-754 well, clang on macOS appears not to wrap around it at casts between `float` and `double`, for instance.

```C
#include <float.h>
#include <stdint.h>
#include <stdio.h>

int main(void)
{
    union {
        uint32_t b;
        float f;
    } x;
    x.b = 0x7fbff000;
    printf("%x\n", x.b);
    double f = x.f;
    printf("%f\n", f);
    x.f = (float)f;
    printf("%x\n", x.b);
    return 0;
}
```
```shell-session
$ clang -ggdb nan-dbl.c && ./a.out
7fbff000
nan
7ffff000
```
...
pack("G")/unpack("G") works great with NaN values.  However,
# - pack("g") completely discards any actual NaN value and always packs the same bytes for a NaN
#   ("bug as implemented" in VALUE_to_float)
Also:
# - unpack("g") always sets the quiet bit to 1 in the Float result
#   (location of bug not obvious to me)
----------------------------------------
Bug #20662: pack("g") completely discards any actual NaN value and always packs the same single-precision bytes for a NaN
https://bugs.ruby-lang.org/issues/20662#change-109335

* Author: cabo (Carsten Bormann)
* Status: Open
* ruby -v: ruby 3.3.4 (2024-07-09 revision be1089c8ec) [arm64-darwin23]
* Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN
----------------------------------------
`pack("G")`/`unpack("G")` works great with NaN values.  However,

- `pack("g")` completely discards any actual NaN value and always packs the same bytes for a NaN
  ("bug as implemented" in `VALUE_to_float`)

Also:

- `unpack("g")` always sets the quiet bit to 1 in the `Float` result
  (location of bug not obvious to me)

---Files--------------------------------
showbug.rb (3.21 KB)
showbug-arm.txt (5.48 KB)
showbug-intel.txt (5.48 KB)

-- 
https://bugs.ruby-lang.org/