Gotchas
=======

.. highlight:: fortran

Variable Initialization Using Initialization Expression
-------------------------------------------------------

The following code::

    integer :: a = 5

is equivalent to::

    integer, save :: a = 5

and *not* to::

    integer :: a
    a = 5

See for example this `question <http://stackoverflow.com/questions/3352741/fortran-assignment-on-declaration-and-save-attribute-gotcha>`_.

.. _floating_point_numbers_gotcha:

Floating Point Numbers
----------------------

Assuming the definitions::

    integer, parameter :: dp=kind(0.d0)           ! double precision
    integer, parameter :: sp=kind(0.0 )           ! single precision

Then the following code::

    real(dp) :: a
    a = 1.0

is equivalent to::

    real(dp) :: a
    a = 1.0_sp

and *not* to::

    real(dp) :: a
    a = 1.0_dp

As such, always use the ``_dp`` suffix as explained in
:ref:`floating_point_numbers`. However, the following code::

    real(dp) :: a
    a = 1

is equivalent to::

    real(dp) :: a
    a = 1.0_dp

And so it is safe to assign integers to floating point numbers without losing
any accuracy (but one must be careful about integer division, e.g.  ``1/2`` is
equal to ``0`` and not ``0.5``).

The declaration of integer variables with a ``_dp`` suffix does not promote them
automatically to double precision variables. The following literal::

    360_dp

is interpreted as an integer.

C/Fortran Interoperability of Logical
-------------------------------------

The Fortran standard specifies, that the Fortran type ``logical(c_bool)`` and C
type ``bool`` are interoperable (as long as ``c_bool`` returns some positive
integer). Unfortunately, for some compilers one must enable this behavior with
a specific (non-default) option. In particular, the following options must be
used:

+-----------+-----------------------+
| Compiler  | Extra Compiler Option |
+-----------+-----------------------+
| gfortran  |                       |
+-----------+-----------------------+
| ifort     | -standard-semantics   |
+-----------+-----------------------+
| PGI       | -Munixlogical         |
+-----------+-----------------------+
| Cray      |                       |
+-----------+-----------------------+
| IBM XL    |                       |
+-----------+-----------------------+

Empty `Extra Compiler Option` means that no extra option is needed and things
work by default.

If you omit these extra compiler options, then when you pass `logical` to and
from Fortran, its value will in general be corrupted when accessed from C. A
minimal code example that exemplifies this behavior is at:
https://gist.github.com/certik/9744747
When you use these extra compiler options, then everything works as expected
and there is no issue.

Conclusion: *always* use these extra compiler options when compiling your
Fortran code, unless you have a specific reason not to.