Why is `0.0 / 0.0` preferred over using NaN

[dhruvan2006]

Description

Encountered an error when compiling native add-ons with MSVC.

Why do we prefer returning 0.0 / 0.0 instead of NAN from <math.h>?

Example:

stdlib/lib/node_modules/@stdlib/stats/base/dists/negative-binomial/skewness/src/main.c

Line 42 in a1f8442

return 0.0 / 0.0; // NaN

Related Issues

No response

Questions

No.

Demo

No response

Reproduction

Expected Results

Actual Results

Version

No response

Environments

Node.js

Browser Version

No response

Node.js / npm Version

No response

Platform

No response

Checklist

• Read and understood the Code of Conduct.
• Searched for existing issues and pull requests.

[kgryte]

@dhruvan2006 What is the error you encountered?

For context, this is a pattern established by following other numerical libraries implementing special math functions (e.g., https://github.com/JuliaMath/openlibm/blob/97af075844906c664ff2b59d674479a281fbc7ad/src/e_log.c#L100). We follow a similar pattern. The original reason has to do with quiet and signaling NaNs. Some downstream users want to interpret NaNs as a result of operations as errors, while others do not. In order to trap NaNs, you need to actually perform an arithmetic operation which generates NaN and not just return a specific bit sequence.

In the future, we will replace the use of 0.0 / 0.0 by a stdlib macro, similar to how we use macros for positive and negative infinity. In the meantime, we inline 0.0 / 0.0 which is known to generate NaN.

[manit2004]

@kgryte Should we start replacing all the inline 0.0 / 0.0 occurrences with the existing stdlib NaN macro, such as STDLIB_CONSTANT_FLOAT64_NAN in the C sources?