HowTo guide for developers¶
This page is intended as a quick reference to solve some of the problems that are commonly encountered when developing within the Abinit project.
How to generate the configure script via makemake¶
Abinit uses the standard configure && make approach to build from source. Note, however, that the developmental version does not contain the configure script. To generate the configure script and the other files required by the build system, you need to execute makemake:
cd ~abinit ./config/scripts/makemake
makemake requires a recent version of the python interpreter as well as m4, autoconf, and automake. If these tools are not installed on your machine, you need to compile them from source or use your preferred package manager to install them. I usually use the conda package manager and:
conda install m4 autoconf automake -c conda-forge
Remember to run makemake every time you add/remove a Fortran file or a new directory or you change parts the buildsystem i.e. the files in ~abinit/config.
How to build Abinit¶
Developers are invited to build the executables inside a build directory i.e. a directory that is separated
from the source tree in order to keep the source directory as clean as possible and allow for multiple builds.
I usually use the naming scheme:
_build_[compiler_name] for the build directory and an external file
(e.g. gcc.ac) storing the configuration options that can be passed to configure via the –with-config-file option:
mkdir _build_gcc cd _build_gcc ../configure --with-config-file=gcc.ac make -j8 # use 8 processes to compile
Note that the name of the options in the config-file is in normalized form that is:
- Remove the initial
--from the name of the option
--with-mpi-prefix in normalized form becomes
Examples of configuration files for clusters can be found in the abiconfig package.
A detailed description of the configuration options supported by the build system is given in this guide by Marc:
Once the build is completed, it is a good idea to check whether the executable works as expected by running the tests in the v1 directory with:
cd tests ../../tests/runtests.py v1 -j4
As usual, use:
to list the available options. A more detailed discussion is given in this page.
Remember to run the tests as frequently as possible while developing new features in order to spot possible regressions or incompatibilities. Trust me, you can save a lot of time if you run runtests.py systematically!
How to browse the source files¶
The HTML documentation generated by Robodoc is available at this page.
If you need a tool to navigate the Abinit code inside the editor, I would sugest exuberant-ctags.
To generate a tags file containing the list of procedures, modules, datatypes for all files inside ~abinit/src, use:
cd ~abinit/src ctags -R
Now it is possible to open the file containing the declaration of the dataset_type Fortran datatype directly from the terminal with:
vi -t dataset_type
Inside the editor, you can go directly to a tag definition by entering the following in vim command mode:
Finally, one can use the abisrc.py script in the
How to debug with gdb¶
Load the executable in the GNU debugger using the syntax:
Run the code with the gdb run command and redirect the standard input with:
(gdb) run < run.files
Wait for the error e.g. SIGSEGV, then print the backtrace with:
Remeber to compile the code with the
-g option. Avoid debugging code compiled with -O3.
In some tricky cases, you may need to resort to -O0 or use Fortran
For a more complete introduction to gdb, we suggest this youtube tutorial:
How to add a new Fortran file¶
- Create the F90 module and
- Register the F90 file in the
abinit.srcfile (avoid duplicated names in the public API, abisrc.py will complain about that)
makemakein the source directory
makein the build directory (possibly
make clean && make)
How to add a new Abinit input variable¶
This section documents the procedure required to add a new Abinit input variable. To make things as simple as possible, we ignore the (more complicated) case of dimensions such as nkpt or nsym whose value may depend on the dataset. To add a new variables follow the below steps:
Add the new variable to dataset_type. Remember that the name cannot end with a digit as this enters into conflict with the multidataset syntax.
The default value of the new input variable can be specified in two different ways:
- in the declaration of the Fortran type if the size is known at compile time and the initial value does not depend on other variables.
- in the indefo routine if the value must be computed at runtime.
Add the name of the new variable to chkvars.
Add a new section to dtset_copy to copy the new variable (use alloc_copy if allocatable).
If you need an allocatable entity, remember to deallocate memory in dtset_free.
Read the variable in the invars2 (if it is not a basic dimension).
Change one of the outvars routines (outvar_a_h, outvar_i_n, outvar_o_z) to print the variable according to the first letter of the new variable
The logic for checking the consistency of input variables goes to chkinp. Use the routines chkint_eq, chkint_ne, chkint_ge, chkint_le, chkdpr.
Add the documentation of the new variable to
~abinit/abimkdocs/variables_CODE.py. following the instructions given in this section.
make clean && make -j8
since you broke the ABI of a public datastructure and all the object files depending on this datastructure must be recompiled (if you are developing a library, you should release a new major version!)
No, it’s not a typo, ABIs and APIs are two different concepts! From this answer on stackoverflow
If you expand, say, a 16-bit data structure field into a 32-bit field, then already-compiled code that uses that data structure will not be accessing that field (or any following it) correctly. Accessing data structure members gets converted into memory addresses and offsets during compilation and if the data structure changes, then these offsets will not point to what the code is expecting them to point to and the results are unpredictable at best.
For the treatment of dimensions see invars0, invars1m
How to add a new test in the test suite?¶
The following information complements the testsuite documentation.
In order to introduce a test, one needs to:
Provide a new input file in e.g. tests/v8/Input (or modify an existing one, possibly in another directory, but please do not suppress the existing capability testing!)
Provide a new reference file in the corresponding tests/v8/Refs
Insert the test in the list of tests to be done, by adding its name in tests/v8/__init__.py.
Document the test by adding a commented section inside the input file (edit an existing input file to follow its style)
Declare the pseudopotentials, the flow of actions, the files to be analyzed, the tolerances for the test, inside this documentation. For the tolerances, start with
tolnlines = 0, tolabs = 0.000e+00, tolrel = 0.000e+00
The different fields control how strictly the test results will be analysed:
- the maximum floating point difference found must be less than tolabs,
- the maximum relative difference less than tolrel, for each line individually,
- tolnlines is the maximal number of lines found to be different between the output and reference file (within another tolerance that can be tuned by the option opt=-medium, opt=-easy or opt=-ridiculous, see the added section in other input files).
If this procedure fails, contact the code maintainers in order adjust the test case. This might mean modifying the tolerances files. Unless you are really expert, let the maintainer do the final adjustment.
Please, try to keep preferably the total time per test case to less than 10 seconds. 30 seconds is a maximum for most test cases. Going being this needs exceptional reasons.
Supposing now that you have introduced a new test case. It can be used, with the other tests of the ABINIT test suite, through different channels:
these tests can be triggered on the test farm Web Page.
locally (on your machine), the whole set of sequential tests, or particular tests or series of tests, can be triggered by issuing, in the ABINIT/tests directory, the command ./runtests.py. (see the many capabilities of this scripts by issuing ./runtests.py –help). Other sets of calculations can be triggered, as described by issuing make help. The result will appear in a new subdirectory Test_suite
Last but not least: are you sure that your modifications do not deteriorate the performance of the code in the regime where your modifications are not used? You should inspect your modifications for both memory use and CPU time.
In computer science, code coverage is a measure used to describe the degree to which the source code of a program is tested by a particular test suite. A program with high code coverage has been more thoroughly tested and has a lower chance of containing software bugs than a program with low code coverage. Many different metrics can be used to calculate code coverage; some of the most basic are the percent of program subroutines and the percent of program statements called during execution of the test suite. We aim that the test suite covers all the functionalities of ABINIT.
How to trigger a coverage report?
There is one slave dedicated to on-demand execution of branches by the developers that produces a code coverage report, at present, higgs_gnu_7.3_cov. It can be launched by the general on-demand interface (contact Jean-Michel or Xavier if you do not yet have access to it). Code coverage reports from recent runs of the tests are available here. If you see parts of the code which are not well tested, please contribute to improving coverage by writing new tests!
How does it work?
ABINIT is built with special options such that every function that is executed in the program
is mapped back to the function points in the source code.
.gcno file is generated when the source file is compiled with the GCC -ftest-coverage option.
It contains information to reconstruct the basic block graphs and assign source line numbers to blocks.
More info are available in the Gvoc page.
A .gcda file is generated when a program containing object files built with the GCC -fprofile-arcs option is executed.
A separate .gcda file is created for each object file compiled with this option.
It contains arc transition counts, and some summary information.
Finally, we use lcov to analyze the .gcda files for generating a html report
How to fix Robodoc warnings¶
The following section is aimed at helping developers to solve problems related to robodoc error messages. It comes from the file doc/developers/robodoc.doc.txt
A robodoc section must always start with a “begin marker” and end with a “end marker”. In ABINIT, the begin marker is usually:
while the end marker is always
A common problem when generating .html for ABINIT thanks to Robodoc, is the lack of an end marker for each begin marker. The solution is simple: find the end of the robodoc section (often, either the end of file, or just before the next begin marker), and insert
How to make the parents script happy¶
The script parents relies heavily on the Robodoc-type structuration of the source file. In particular, a source file should be made of sets of robodoc blocks, possibly separated by comments or blank lines.
In most of the source files, one block relates to a subroutine, typically:
!!****f* ABINIT/name_of_subroutine !! NAME !! name_of_subroutine <...> subroutine name_of_subroutine <...> !!***
The most common source of errors comes from the fact that the parent.pl script expects the name_of_subroutine to be coherent in the three above occurrences. In this case, you should get something like:
Error: Robodoc subroutine NAME missing in <dir/name_of_routine.F90>
For example, the following will generate an error
!!****f* ABINIT/blabla !! NAME !! blabla <...> subroutine blabla_int <...> !!***
It will be easy to correct this piece of code:
!!****f* ABINIT/blabla_int !! NAME !! blabla_int <...> subroutine blabla_int <...> !!***
Unfortunately, in some ABINIT source files, the number of such blocks can be large (bigger than 10), and the above-mentioned error message does not mention the line where the error was found, neither the subroutine for which there has been a mismatch. How to find the place where the error occurred?
In the full list of messages obtained when executing
one can find the list of block in this source file for which make parent succeeded.
The suggestion is thus to examine the list of blocks or subroutines (e.g. grep ‘end subroutine’), and proceed by elimination: the one that failed is the only one who did not succeed. Sorry if this procedure is not very nice.
Q2 How to debug if a section of the source file disappeared?
Answer: Well, it is likely that you have forgotten the keyword
after the section CHILDREN.
How to make the abiauty script happy¶
in the src directory, one can get several types of messages. However, they are not always very indicative of what should be corrected. Here follows a set of hints to fix the problem identified by abiauty.
The goal of abiauty is to indent correctly the lines inside the F90 source files of ABINIT. In order to do this, the F90 source files must respect some rules. It will be quite long to explain them. You can get a feeling by simply looking at existing files!
The script abiauty relies heavily on the structuration of the source file. In particular, after a section where variables are declared, abiauty expects the following line:
namely, a ‘!’, then a blank, then 73 stars - copy it from here, it is the simplest.
In the above mentioned case, such a line was missing at the line 138 of the file 32_util/interpol3d.F90, preventing abiauty to understand the structure of the file.
More explicitly, here is the old section (that does not follow the abirules):
real(dp) :: d1,d2,d3 d1=one/nr1
And here is the new section (that follows the abirules):
real(dp) :: d1,d2,d3 ! ************************************************************************* d1=one/nr1
Q1: How to debug ABINIT when he following diagnostic is obtained
ERROR(66_paw/qijb_bk.F90): found end statement at line 183 for '' subroutine '' instead of '' subroutine qijb_bk ''
Simply, you have not mentioned the name of the routine at the “end subroutine statement”. So, in the routine name_of_routine.F90, replace:
end subroutine name_of_routine
Q2: How to debug ABINIT when the following diagnostics are obtained
ERROR(72_response/nstpaw3.F90): found end statement at line 652 for '' if '' instead of '' do construct '' ERROR(42_parser/instrng.F90): found end statement at line 316 for '' subroutine instrng '' instead of '' do construct ''
There are some fundamental limitations of the abiauty script, moreover accompanied with a diagnostic that is of no help.
(1) abiauty cannot treat an “if” statement, followed by a conditional that is split on two lines, with moreover, a comment at the end of the first line. E.g. it cannot treat:
if( conditional1 .and. & ! Here is a comment, that cannot be treated correctly by abiauty & conditional2 ) then
You can choose between different possibilities avoid this problem:
! Here is a comment BEFORE the if if( conditional1 .and. & & conditional2 ) then
if( conditional1 .and. conditional2 ) then ! Here is a comment for the whole line
if( conditional1 .and. & & conditional2 ) then ! Here is a comment for the split line
(2) in certain cases, abiauty cannot treat a “call” that appears immediately after a “if”. E.g it cannot treat:
if (condition) call name_of_subroutine
You have to replace it by
if (condition) then call name_of_subroutine endif
(3) abiauty cannot treat a “enddo” that appears on the same line than another command thanks to a “;”. E.g. it cannot treate:
do ii=1,nn ; mm=mm+ii ; enddo
You have to replace it by
do ii=1,nn mm=mm+ii enddo
The indication of the line where the script realizes that there is a problem, is not the indication of the line at which the problem occurs. However, the abiauty script identifies also, earlier, that there is a problem with this line. Typically, the two error messages appear:
ERROR(42_parser/instrng.F90): semicolon found at line 102 with do statement ERROR(42_parser/instrng.F90): found end statement at line 103 for '' if '' instead of '' do construct ''
ERROR(42_parser/instrng.F90): semicolon found at line 101 with do statement ERROR(42_parser/instrng.F90): found end statement at line 316 for '' subroutine instrng '' instead of '' do construct ''
Suggestion: examine first the ‘semicolon found at line 102 with do statement’ error message.
Q3: How to debug ABINIT when the following diagnostic is obtained:
ERROR(95_drive/timana.F90): semicolon found at line 853 with case(1) statement
Following the abirules, one should not have the following syntax:
case(xx) ; instruction
The instruction should instead appear in the line following the case condition: