Thank you for taking the time to contribute!
If this is your first contribution, we invite you to read our list of guidelines, common to all AdaCore repositories.
Contributions to the src/gnat
subfolder are most suitable to be submitted
here. For any other contributions, consider submitting them directly to the
Why3 project.
Welcome, contributor, to the Why3 verification platform!
If you wish to contribute, open an issue or otherwise interact with our Gitlab, you will need to have an INRIA account. External users can file bug reports using our mailing list. We're sorry for the inconvenience.
To build Why3 locally you will need a functional installation of OCaml (at least 4.08), menhir
, num
and autoconf
. You can set up your developer build using the following commands:
autoconf
./configure --enable-local # stores the built binaries under ./bin
make
Note: there can be issues around Num, recall and document those.
Building the Why3 documentation requires an installation of Sphinx, as well as the sphinxcontrib-bibtex
package. If your package manager does not have up to date versions of these packages, or they otherwise fail to install, you can install them using pip3
as follows:
pip3 install sphinx sphinxcontrib-bibtex
You will need to re-configure Why3 to enable documentation building. Once it has been configured, documentation can be built by running make doc
.
To execute the Why3 tests run:
./bench/bench
You may run specific classes of tests through the -only
flag, the full listing of classes can be obtained from -help
.
For example to run the 'good file' tests, use:
./bench/bench -only goodfiles
Every new commit pushed on the gitlab repository invokes an continuous
integration check. The description on how to upgrade this CI procedure
is specifically done in misc/ci.md
To add a new directory of tests add a new line in bench/bench
. The suite should be added to the relevant block, which is determined by the kind of assertion it performs. For example, if you wished to add a set of files which should successfully prove, add a line like goods path/to/tests
to the appropriate block.
Certain tests, those in examples/
, examples/bts
, examples/programs
and examples/check-builtin
are replayed every night on Moloch. To ensure accuracy and avoid flakiness, the sessions for these tests must be created on Moloch as well.
Every night, the latest master is tested against a series of standard
configurations, on Moloch (moloch.lri.fr
). This ensures that
regressions are caught across a wide array of different prover
versions and families. The results of Nightly CI are emailed to a set
of selected developers emails each morning.
If you require access to Moloch in order to update sessions or test a new prover, ask one of the maintainers for help.
Why3 uses a Pull Request development model. Contributions should be made to separate branches and a Merge Request should be opened in Gitlab, explaining the contents and purpose of its contents.
While code review is not strictly enforced it is highly encouraged.
Every commit should:
- Compile, and pass CI.
- Be stripped of trailing whitespace.
- Use appropriate indentation.
- Be limited to 80 columns.
- Have a clear and relevant commit message
- Be as functionally isolated as possible
The module src/core/model_parser
contains the main data structure
for counterexamples.
Parser functions for counterexamples must be declared in driver files
(e.g. declaring model_parser "smtv2"
somewhere). Such functions
should first be registered using Model_parser.register_model_parser
.
The latter function takes as input a so-called raw_model_parser
(e.g. Smtv2_model_parser.parse
). This raw_model_parser
is an
input of the internal function Model_parser.model_parser
, which is
the main function for counterexamples generation. A
raw_model_parser
is a function that processes a solver textual
output and produces a preliminary structure for counterexamples, which
is completed by Model_parser.model_parser
.
So, as mentioned above, an example of a raw_model_parser
, for SMTv2
solvers, is Smtv2_model_parser.parse
, which makes use of another
data type Smtv2_model_defs.function_def
. This specific
Smtv2_model_parser.parse
function proceeds in 2 main phases:
- first transforms the textual output into an S-expression,
- then the latter S-expression is transformed into a
Model_parser.model_element list
.
The raw_model_parser
functions are using the variable names as
printed by the printer. It is only the Model_parser.build_model_rec
function that recovers the original Why3 name from the
printing_info
. The type printing_info
is declared in module
src/core/printer
. Task printers are taking as arguments a record of
type Printer.printer_args
. The field printing_info
is supposed to
be modified in place by printers.
Profiling execution of Why3 can be performed out-of-the-box using
under Linux using perf
. A typical usage is
perf record --call-graph=dwarf -- why3 command <options> <arguments>
perf report
Note that perf may complain you don't have enough priviledges. A typical configuration change required is, for the session only, execute
sysctl kernel.perf_event_paranoid=2
or, to make this setting permanent, add the line
kernel.perf_event_paranoid = 2
to sys/sysctl.conf
.