Skip to content

no-real-suffix (C021)

What it does

Checks for floating point literal constants that don't have their kinds explicitly specified.

Why is this bad?

Floating point literals use the default 'real' kind unless given an explicit kind suffix. This can cause surprising loss of precision:

use, intrinsic :: iso_fortran_env, only: dp => real64

real(dp), parameter :: pi_1 = 3.14159265358979
real(dp), parameter :: pi_2 = 3.14159265358979_dp

print *, pi_1  ! Gives: 3.1415927410125732
print *, pi_2  ! Gives: 3.1415926535897900

There are cases where the difference in precision doesn't matter, such as:

real(dp) :: x, y

x = 1.0
y = real(2.0, kind=dp)

A case where a missing suffix may be intentional is when using a kind statement:

integer, parameter :: sp = kind(0.0)

This rule will try to avoid catching these case. However, even for 'nice' numbers, it's possible to accidentally lose precision in surprising ways:

real(dp) :: x

x = sqrt(2.0)

This rule will therefore require an explicit kind statement in the majority of cases where a floating point literal is found in an expression.

References