******************* Fortran style guide ******************* The main principle when contributing code to Fortran projects should be readability. If your code can not be easily read and understood by others it will be hard to maintain and extend. It should also fit well with the existing parts of the code (in style as well as in its programming paradigms) maintaining the principle of least surprise. Below you will find some explicit coding rules we try to follow. The list can not cover all aspects, so also look at the existing source code and try to follow the conventions being used. If you use Emacs as editor, consider adding `appropriate customisation settings `_ to your config file in order to automatically enforce some of the conventions below. You may also wish to use `ws-butler `_. Line length and indentation =========================== * Maximal **line length** is **100** characters. For lines longer than that, use continuation lines. * **Nested blocks** are indented by **2** white spaces:: write(*, *) "Nested block follows" do ii = 1, 100 write(*, *) "This is the nested block" if (ii == 50) then write(*, *) "Next nested block" end if end do * **Continuation lines** are indented by **4** white spaces. Make sure to place continuation characters (`&`) both at the end of the line as well as at the beginning of the continuation line:: call someRoutineWithManyParameters(param1, param2, param3, param4,& & param5) Try to break lines at natural places (e.g. at white space characters) and include one white space character after the opening ampersand in the continuation line. * **Single line preprocessor directives** are indented as normal code:: @:ASSERT(someCondition) call someRoutine(...) * **Preprocessor block directives** (directives with starting and ending constructs) are outdented by **2** characters with respect of the code they enclose. The enclosed code must be aligned as if the preprocessor directives were not present:: call doSomething() #:if WITH_SCALAPACK call someRoutineScalapackVersion(...) #:else call someRoutineSerialVersion(...) #:endif do iKS = 1, nKS #:if WITH_SCALAPACK call someRoutineScalapackVersion(iKS, ...) #:else call someRoutineSerialVersion(iKS, ...) #:endif end do Naming ====== The naming conventions basically follow those in the `Google Style Guide for Java naming convention `_, with minor modifications. * **Variable** names follow the **lowerCamelCase** convention:: logical :: hasComponent * **Constants** (parameters) use the **lowerCamelCase** convention similar to variables :: integer, parameter :: maxArraySize = 100 with the exception of the constants used to define the kind parameter for intrinsic types, which should be all lowercase (and short):: integer, parameter :: dp = kind(1.0d0) real(dp) :: val * **Subroutine** and **function** names follow also the **lowerCamelCase** notation:: subroutine testSomeFunctionality() myValue = getSomeValue(...) * **Type** (object) names are written **UpperCamelCase**:: type :: TRealList type(TRealList) :: myList All type names should be prefixed with a capital 'T', in order to clarify the distinction between type names and variable names:: type :: TBroydenMixer : end type TBroydenMixer : type(TBroydenMixer) :: broydenMixer * **Instances** referenced out of type-bound procedures are to be named `this`:: subroutine typeBoundProcedure(this, ...) class(TType), intent(inout) :: this : end subroutine typeBoundProcedure * **Module** names follow **lower_case_with_underscore** convention:: use dftb_common_accuracy Underscores are used for name-spacing only, so the module above would be typically found at the path `dftb/common/accuracy.f90`. The individual component names (``dftb``, ``common``, ``accuracy``) may not contain any underscores and must be shorter than 15 characters. * **Preprocessor** variables and macros follow **UPPER_CASE_WITH_UNDERSCORE** convention:: #:if WITH_MPI withMpi = ${FORTRAN_LOGICAL(WITH_MPI)}$ #:endif White spaces ============ Please use white spaces to make the code readable. In general, you **must use** white spaces in following situations: * Around arithmetic operators:: 2 + 2 * Around assignment and pointer assignment operators:: aa = 3 + 2 pWindow => array(1:3) * Around the ``::`` separator in declarations:: integer :: ind * After commas (``,``) in general and especially in declarations, calls and lists:: real(wp), allocatable :: array(:) type, extends(TBaseType) :: TDerivedType subroutine myRoutine(par1, par2) call myRoutine(val1, val2) print *, 'My value:', val do ii = 1, 3 array(1:3) = [1, 2, 3] * When separating array indices, when the actual index value for an index contains an expression:: myArray(ii + 2, jj) = 12 You **may omit** white space in following cases: * When separating array indices and the actual index values are simple and short (typically two letters) variable names, one or two digit integers or the range operator ``:``:: myArray(:,1) = vector latVecs(1,1) = 1.0_wp myArray(ii,jj) = myArray(jj,ii) You **must omit** white spaces in following cases: * Around opening and closing braces of any kind:: call mySubroutine(aa, bb) ! and NOT call mySubroutine( aa, bb ) myVector(:) = [1, 2, 3] ! instead of myVector(:) = [ 1, 2, 3 ] tmp = 2 * (aa + bb) ! instead of 2 * ( aa + bb ) * Around the equal (``=``) sign, when passing named arguments to a function or subroutine:: call mySubroutine(aa, optionalArgument=.true.) * Around the power operator:: val = base**power (instead of val = base ** power) **Avoid** white spaces for **visual aligning** of code, use:: integer, intent(in) :: nNeighbors real(wp), intent(out) :: interaction instead of:: integer, intent(in) :: nNeighbors real(wp), intent(out) :: energy Although latter may look more readable, it makes rather difficult to track real changes in the code with the revision control system. For example when a new line is added to the block making the realignment of previous (but otherwise unchanged) lines necessary :: integer, intent(in) :: nNeighbors real(wp), intent(out) :: energy real(wp), intent(out), optional :: forces(:) the version control system will indicate all of those lines having been modified, although only the alignment (but not the actual instructions) were changed. Comments ======== * **Module**, **Subroutine** and **function** comments should be consistent with `doxygen `_ / `FORD `_ literate comments for publicly visible interfaces and variables. * Variable/module/routine comments on multiple lines should use a double-bang for the second and subsequent lines, with the first line capitalised :: !> This is a multi-line comment for something, !! it continues on a second line. * Adapt your comment if it starts with a term that is expected to be in lower-case :: !> The k-point instead of :: !> K-point * Comments are indented to the same position as the code they document:: ! Take spin degeneracy into account energy = 2.0_wp * energy * Generally, write the comment *before* the code snippet it documents:: ! Loop over all neighbours do iNeigh = 1, nNeighbours : end do * Try to avoid mixing code and comments within one line as this is often hard to read:: bb = 2 * aa ! this comment should be before the line. * Never use multi-line suffix comments, as an indenting editor would mess up the indentation of subsequent lines:: bb = 2 * aa ! This comment goes over multiple lines, therefore, it ! should stay ALWAYS before the code snippet and NOT HERE. * Specifically comment any workarounds, include the compiler name and the version number for which the workaround had to be made. Always use the following pattern, so that searching for workarounds which can be possibly removed is easy:: ! Workaround: gfortran 4.8 ! Finalisation not working, we have to deallocate explicitly deallocate(myPointer) * Comments should always start with one bang only. Comments with two bangs are reserved for source code documentation systems:: ! This block needs a documentation do ii = 1, 2 : end do * If you need a comment for a longer block of code, consider instead packaging that block of code into a properly named function (if the additional function call would be performance critical, write it as an internal procedure):: somePreviousStatement ind = getFirstNonZero(array) someStatementAfter instead of :: somePreviousStatement ! Look for the first nonzero element found = .false. do ind = 1, size(array) if (array(ind) > 0) then found = .true. exit end if end do if (.not. found) then ind = 0 end if someStatementAfter Block constructs ================ * Block constructs are normally used in their verbose form, with block opening and corresponding block closing in separate lines:: if (some_conditions) then ! Do something ... end if * The closing form should have a space between the ``end`` keyword and the construct type:: do ii = 1, 10 ... end do ! instead of "enddo" * If the block construct contains an expression within obligatory parentheses, insert one space between the block type and the opening parenthesis:: if (some_condition) then ! instead of "if(some_condition)" where (aa == 0) ! instead of "where(aa == 0)" * Some block constructs have alternative one-line short forms without closing statements (e.g. ``if``, ``where``). Only use their short form, if it is readable and fits into a single line:: if (ioStat /= 0) return if (allocated(someArray)) someArray(:,:) = 0.0_dp where (abs(aa) >= epsilon(0.0_dp)) aa = 1.0_dp / aa Allocation status ================= At several places, the allocation status of a variable is used to signal choices about logical flow in the code:: !> SCC module internal variables type(TScc), allocatable :: sccCalc . . . if (allocated(sccCalc)) then end if This is to be preferred to the use of additional logical variables if possible. Part of the reason for this choice is that from Fortran 2008 onwards, optional arguments to subroutines and functions are treated as not-present if not allocated. File I/O ======== All files must be opened (i.e., connected to a descriptor) by the ``openFile()`` routine, which initializes a ``type(TFileDescriptor)`` instance. Whenever possible, use the ``mode`` argument to specify the file opening type:: call openFile(fd, "test.dat", mode="r") The ``mode`` specifier accepts the following possible options: * ``r``: read (file must exist, the descriptor is at the start of the file contents), * ``r+``: read and write (file must exist, the descriptor is at the start of the file contents), * ``w``: write (file will be replaced if already existing, otherwise created) * ``w+``: read and write (file will be replaced if it already exists, otherwise created) * ``a``: appended write (file will be opened if it already exists, otherwise created; the descriptor will be positioned at its end) * ``a+``: appended read and write (file will be opened if it already exists, otherwise created; the descriptor will be positioned at its end) Additionally the letter ``b`` can be appended to open the file in binary (unformatted) mode (e.g. ``rb`` for reading a binary file or ``a+b`` for appending to a binary file in read and write mode). For reading, writing and rewinding, the ``%unit`` field of the descriptor should be used. Do not change the value of ``%unit``. Do not close the file with the ``close`` statement, but use the ``closeFile()`` routine instead. (Actually, files are automatically closed, if the connected descriptor leaves code scope, but for better readability of the code, we close them explicitely by calling the ``closeFile()`` routine.) Calling ``closeFile()`` with an unconnected descriptor is fine, it will simply do nothing. This should allow you to eliminate most guarding ``if`` statements around any ``closeFile()`` calls.