Skip to main content

PWR064: Precision loss in floating-point constant

Issue​

An unsuffixed real literal has default precision, which may cause precision loss if the value is actually destined for a wider variable.

Actions​

Suffix the floating-point literal with an appropriately precise kind derived from either selected_real_kind() or the iso_fortran_env module.

Relevance​

Floating-point literals without a kind suffix have default precision, which may offer inconsistent behavior across compilers or compilation flags. The default kind behavior is the same as for variables, see PWR071: Prefer real(kind=kind_value) for declaring consistent floating types for more details.

warning

Avoid the use of compilation flags such as GNU's -fdefault-real-8 to increase the precision of unsuffixed literals. While many compilers interpret these correctly, they are ultimately not part of the Fortran standard, and thus code relying on them may silently break on other compilers.

note

Unlike their C analogue, the default kind of unsuffixed literals is not that of double precision.

Code example​

In the following example, an unsuffixed literal is used to initialize a parameter with an explicitly specified kind:

! example.f90
program test_literal_without_suffix
  implicit none
  integer, parameter :: dp = selected_real_kind(15, 307)
  real(kind=dp), parameter :: e = 2.718281828459045
  print *, e
end program test_literal_without_suffix

In many implementations the default real kind is not precise enough, so rounding errors will be introduced to the literal before it is assigned. To prevent this, suffix the literal with the same kind as the variable:

! solution.f90
program test_literal_with_suffix
  implicit none
  integer, parameter :: dp = selected_real_kind(15, 307)
  real(kind=dp), parameter :: e = 2.718281828459045_dp
  print *, e
end program test_literal_with_suffix

References​