sNaNs

Signaling NaN, or sNaN, is a category of NaN values defined in IEEE 754 floating-point arithmetic to indicate an invalid or uninitialized result. An sNaN is a NaN value whose encoding designates it as signaling, in contrast to a quiet NaN (qNaN). When an operation encounters a signaling NaN, the processor is specified to raise a floating-point invalid-operation exception. The exception may be delivered to software as a trap, a status flag, or may be masked depending on the environment. In many modern systems, arithmetic with sNaNs yields a NaN result and a corresponding exception flag; implementations may silence the exception by converting sNaN to qNaN, effectively hiding the signaling. sNaNs originated in the IEEE 754 standard as a mechanism to detect invalid or uninitialized operands.

Representation: In binary formats, NaN values have an exponent of all ones and a nonzero significand. The most significant bit of the significand (the quiet bit) differentiates sNaN from qNaN: a zero in that bit typically denotes a signaling NaN, while a one denotes a quiet NaN. The remaining significand bits can carry diagnostic payload.

Behavior: Any operation involving a NaN yields NaN; comparisons with NaN are unordered. Signaling NaNs are primarily used to catch errors during development or in diagnostic tools. In practice, many languages treat signaling NaNs as quiet NaNs once an operation is performed, to avoid frequent traps.