Skip to content

zerothi/fdict

Repository files navigation

fdict

Build Status Donate

A variable and dictionary in pure fortran for retaining any data-type and a fast hash-table dictionary.

Usage

This module consists of two separate modules which co-exist for maintenance and usage reasons.

First, the variable module which is a type-free variable that can contain any variable type, and any dimension as well.

Second, the dictionary module which contains a hash-table of variables that can contain any data-type allowed by the variable module.

Downloading and installation

Installing fdict requires a download of the library hosted at github at fdict@git.

Installation can be done via 2 different back-ends. 1) smeka build system, or 2) CMake build.

smeka build system

Extract and create an setup.make file for compilation, a minimal setup.make file can look like this

FC=gfortran
FFLAGS = -g

if in a build directory, also add this:

PREFIX =

Type make and a library called libfdict.a is created. Subsequently the installation may be performed by:

make PREFIX=/path/to/fdict install

which installs the required files (modules and libraries) to the folder. It will also install pkg-config files for auto-detection.

CMake

CMake procedure can be done via the normal procedure:

cmake -S. -Bbuild-fdict
cmake --build build-fdict

fdict should also be able to be used in a sub-project. If problems occur, feel free to open up an issue.

Linking to fdict

To use the dictionary you need to add include statements for the modules as well as linking to the program.

To link fdict to your program the following can be used in a Makefile

FDICT_PATH  = /path/to/fdict/parent
FDICT_LIBS  = -L$(FDICT_PATH) -lfdict
FDICT_INC   = -I$(FDICT_PATH)

Alternatively, one can use pkg-config for obtaining the include flags and libraries.

For parent programs that uses fdict there are 2 ways of knowing which fdict version one is using:

  1. A simple header file (like C-preprocessor statements), this information is found in fdict.inc
  2. A fypp compatible include file which contains library version and which data types are included in the built library, see the file fdict.fypp

The file fdict.inc may be included in projects which exposes the following definitions:

_FDICT_MAJOR_ 0
_FDICT_MINOR_ 9
_FDICT_PATCH_ 0
_FDICT_VERSION_ 0.9.0

This is mainly meant as a feature usable when the fdict interface and e.g. modules change names.

Alternatively the fdict.fypp inclusion file exposes variables such as:

  • the library version numbers (as above)
  • which data-types are enabled
  • the number of ranks for each kind

The fdict.fypp file is handy when you are already relying on fypp whereas the regular fdict.inc header files are easy to use in standard fortran source compilation.

Controlling interfaces

Typically not needed: allows for customization of different interfaces.

By default the number of dimensions allowed by the library is 5, i.e. there is no interface created for real a(:,:,:,:,:,:), etc. However, to accomodate arbitrary dimensions you must define constants in your setup.make file.

There are several fine-tuning options that allows creating more or fewer interfaces. As the number of dimensions increases, so does the library size. If only a specific maximum range of ranks are required, it might be beneficial to reduce maximum ranks to the used range.

Currently the fdict library supports the types listed in the below table:

Type Precision format (GNU) C-type Default
type(variable_t) --- yes
character(len=1) char yes
integer selected_int_kind(2) byte no
integer selected_int_kind(4) short no
integer selected_int_kind(9) int yes
integer selected_int_kind(18) long yes
real selected_real_kind(6) float yes
real selected_real_kind(15) double yes
real selected_real_kind(18) ext. double no
real selected_real_kind(30) quad no
complex selected_real_kind(6) float complex yes
complex selected_real_kind(15) double complex yes
complex selected_real_kind(18) ext. double complex no
complex selected_real_kind(30) quad complex no
logical selected_int_kind(2) byte no
logical selected_int_kind(4) short no
logical selected_int_kind(9) int yes
logical selected_int_kind(18) long no
type(c_ptr) void * no
type(c_funptr) (procedure) void * no

In the Default column one can see which data-types are enabled by default. The most commonly used data-types are enabled.

To enable the non-default data types you can do so with (Makefile scheme):

FYPPFLAGS += -DWITH_INT8=1 # for int kind(2)
FYPPFLAGS += -DWITH_INT16=1 # for int kind(4)
# Note that not all compilers support extended precisions
# If you experience compiler errors, this is likely the cause.
FYPPFLAGS += -DWITH_REAL80=1 # for real and complex kind(18)
FYPPFLAGS += -DWITH_REAL128=1 # for real and complex kind(30)
FYPPFLAGS += -DWITH_LOG8=1 # for logical kind(2)
FYPPFLAGS += -DWITH_LOG16=1 # for logical kind(4)
FYPPFLAGS += -DWITH_LOG64=1 # for logical kind(18)
FYPPFLAGS += -DWITH_ISO_C=1 # for enabling c_ptr and c_funptr

For cmake the variables are all prefixed with FDICT_, e.g. -DFDICT_FYPPFLAGS, to ensure there are no clashes with parent programs.

By default fdict generates the kind specifications from the selected_*_kind routines, however, if one wishes to use the iso_fortran_env module simply add FYPPFLAGS += -DWITH_ISO_ENV.

To control the maximum ranks in the interfaces one can add these:

# type(c_ptr), type(c_funptr) and character(len=1)
# are data types that are not affected by the MAXRANK variable

# globally define the maximum ranks of all but the above listed
FYPPFLAGS += -DMAXRANK=n

# integer(*) types maximum rank
FYPPFLAGS += -DMAXRANK_INT=n

# real(*) types maximum rank
FYPPFLAGS += -DMAXRANK_REAL=n

# complex(*) types maximum rank
FYPPFLAGS += -DMAXRANK_CMPLX=n

# logical(*) types maximum rank
FYPPFLAGS += -DMAXRANK_LOG=n

# type(c_ptr), type(c_funptr) types maximum rank
FYPPFLAGS += -DMAXRANK_ISO_C=n

variable

Using this module one gains access to a generic type variable which can contain any data format.

It is used like this:

use variable
integer :: a(3
type(variable_t) :: v
a = 2
call assign(v,a)
a = 3
call assign(a,v)

Also the variable contains an abbreviation for assigning pointers to not copy data, but retain data locality:

integer, target :: a(3)
type(variable_t) :: v
a = 2
call associate(v,a)
a = 3
! Now v contains a = 3

To delete a variable do:

use variable
type(variable_t) :: v
call delete(v)

However, when the variable is using pointers, instead the user can do

use variable
type(variable_t) :: v
! preferred
call nullify(v)
! or
call delete(v,dealloc=.false.)

which merely destroys the variable object and thus retains the data where it is. As with any other pointer arithmetic it is up to the programmer to ensure there is no memory leaks.

In some cases one does not know which data-type is being stored in a variable. Here it may be beneficial to lookup the type of data:

use variable
integer, target :: a(3)
type(variable_t) :: v
a(:) = 2
call associate(v,a)
if ( which(v) == which(a) ) then ! signal integer of 1D (i0 for scalar)
   call assign(a, v)
end if

! Another possibility is to *try* to get the value
logical :: success
integer, target :: i1(3)
real, target :: r1(3)

call assign(r1, v, success=success)
if ( .not. success ) then
    call assign(i1, v, success=success)
end if
... etc ...

However, it may be better to explicitly check the type using which. For consistency and API changes, it is encouraged to use which(<type>) to ensure that the data-types are as expected. I.e. which([real(real64) ::]) is the preferred way of forcing a data-type contained in a variable.

dictionary

Using type(variable_t) it becomes easy to create dictionaries in fortran.

Using this module we implement a dictionary which can contain any data format using a key:val based formalism. The underlying data structure is a linked list sorted according to hash-values of the keys. Hence searching for specific elements in the dictionary is extremely fast. Searching a dictionary with 100 keys 300000 times takes less than 0.04 seconds on a Haswell laptop. Concatenating dictionaries is also very fast.

Creating a dictionary is almost as easy as the Python equivalent:

use dictionary
type(dictionary_t) :: dict
dict = ('KEY'.kv.1)

To extend a dictionary one uses the concatenating format:

dict = dict // ('Hello'.kv.'world') // ('No'.kv.'world')

Again as is used by the type(variable_t) one can with benefit use .kvp. to create the dictionary value by pointers instead of copying the content. Hence doing:

real :: r(4)
dict = dict // ('reals'.kvp.r)
r = 4

will change the value in the dictionary. Note that one can easily create memory leaks with dictionaries:

use dictionary
type(dictionary_t) :: dict
dict = ('KEY'.kv.1)
dict = dict // ('KEY'.kv.2)
dict = ('KEY'.kv.3)

The 1st assignement is valid since the dictionary is empty. The 2nd assignment concatenates and does not produce any memory leaks. In that case the old key KEY is deleted and the new value 2 is inserted. The 3rd assignment produces a memory leak since the pointer to the original dictionary gets lost. Be sure to call call delete(dict) prior to single assignments.

There are various ways to access the data in a dictionary.

  1. Accessing specific keys may be exercised using

     use dictionary
     type(dictionary_t) :: dict
     type(variable_t) :: var
     integer :: i
     real :: r
     logical :: success
     dict = ('KEY'.kv.1)
     call assign(r, dict, 'KEY', success=success)
     if ( .not. success ) call assign(i, dict, 'KEY', success=success)
     call assign(var, dict, 'KEY')
    

    Since values in dictionaries are stored using variable_t we have to follow the limitations of that implementation. Therefore it may be better to always use a temporary variable_t to retrieve the values stored. This will remove a redundant lookup in the dictionary.

  2. Users may find the .key. and .value. operators which only acts on the first element of the dictionary (which may be a surprise). This is only useful for looping dictionaries.

     use dictionary
     type(dictionary_t) :: dict, dict_first
     type(variable_t) :: var
     character(DICTIONARY_KEY_LENGTH) :: key
     integer :: i
     real :: r
     logical :: success
     dict = ('KEY'.kv.1)
     dict = dict // ('KEY1'.kv.3)
    
     ! start looping
     dict_first = .first. dict
     do while ( .not. (.empty. dict_first) )
        ! now .key. and .value. could be used:
        key = .key. dict_first
        call assign(var, dict_first)
        ! Get next dictionary entry
        dict_first = .next. dict_first
     end while
    

Note that the dictionary can also contain any data type.

However, if it needs to do custom data-types the programmer needs to extend the code by supplying a few custom routines.

Intrinsically the dictionary can contain dictionaries by this:

use dictionary
type(dictionary_t) :: d1, d2
d1 = ('hello'.kv.'world')
d2 = ('hello'.kv.'world')
d1 = d1 // ('dict'.kvp.d2)

But it will be up to the user to know the key for data types other than integers, reals, complex numbers, characters and c_* extension types.

Note that the dictionary contained is passed by reference, and thus if you delete d2, you will have a dangling pointer in d1.

Contributions, issues and bugs

I would advice any users to contribute as much feedback and/or PRs to further maintain and expand this library.

Please do not hesitate to contribute!

If you find any bugs please form a bug report/issue.

If you have a fix please consider adding a pull request.

License

The fdict license is MPL-2.0, see the LICENSE file.

Thanks

A big thanks goes to Alberto Garcia for contributing ideas and giving me bug reports. Without him the interface would have been much more complex!