NA / NaN detection

[DavisVaughan]

NA and NaN functions can generally be found in two places, src/main/arithmetic.c and src/include/R_ext/Arith.h.

All of the functions / objects / macros below can be accessed by including R.h, as the ones in arithmetic.c are re-exposed in R_ext/Arith.h, which is include-ed in R.h.

The three most important are R_IsNA(), R_IsNaN() and ISNAN().

---

src/main/arithmetic.c contains:

• Initialization of missing value objects such as R_NaInt, R_NaReal, R_NaN:

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/main/arithmetic.c#L163-L177

• R_IsNA() to detect if a double value is NA_real_

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/main/arithmetic.c#L120-L128

• R_IsNaN() to detect if a double value is NaN

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/main/arithmetic.c#L130-L138

Note: Both R_IsNA() and R_IsNaN() call isnan() internally, and then do additional processing if it returns TRUE to determine which type of nan value it is. You should only use these if you need to be able to tell the difference between the missing value types. If you just want to know if a value is either NA or NaN, use isnan() directly or ISNAN() (see below for more).

---

src/include/R_ext/Arith.h additionally contains:

• Definitions of the following macros, which just use the above missing value objects.

  • #define NA_LOGICAL R_NaInt
  • #define NA_INTEGER R_NaInt
  • #define NA_REAL R_NaReal

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/include/R_ext/Arith.h#L55-L58

• The ISNA() macro, which is just R_IsNA()

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/include/R_ext/Arith.h#L64

• The ISNAN() macro, see below for further explanation.

  https://github.com/wch/r-source/blob/62a4c351ef455c9233f2565ab48453818258dd51/src/include/R_ext/Arith.h#L66-L77

---

A developer's decision to use C's isnan() vs R's ISNAN() is dependent on a few things. Both functions can be used to detect if a double is either NA_real_ or NaN, which is very useful as a single check for either value, rather than calls to both R_IsNA() and R_IsNaN().

• If using C++, you will not have access to isnan() as the C++ math headers undefine it, so you must use ISNAN().

• If using C, you can use either, but be aware that isnan() is not guaranteed to return 1 for TRUE. On an AWS Windows 2019 Server for example, it returned -1. You can therefore safely use isnan() inside of an if (isnan(x)) {} block, but if you need to return the value of isnan(), then you should use R's ISNAN() which guarantees a result of 1 or 0 by wrapping it with isnan(x) != 0.