The singleton is not robust to initialization in several translation units between
shared library with hidden visibility. Removing
Fixing tests accordingly.
The diagnostic is issued when compiled with GCC/clang flag -Wundef.
Although in case BOOST_CLANG is not defined, ie. is not a macro
and BOOST_CLANG identifier is considered to be zero, the #ifdef
directive makes the intention clearer.
* topic/v38-doc-update:
Minor change log changes
colour-output default value fix
Updating runtime cla-help
Documentation added on test naming and sanitizing
Removing tabs
"--run_test=*" activates the disabled tests
* topic/12953-access-master_test_suite-in-datatest-cases:
Change log
Another example with different arity
Datasets based on std::initializer_list fixes
Delayed/lazy dataset construction
Handling sequences of size 0
# Conflicts:
# doc/closing_chapters/change_log.qbk
# test/Jamfile.v2
Since the previous changes, the elements of the std::initializer_list
get destroyed before the test tree initializes properly. The changes
are
- not relying on a moved std::initializer_list for storing the values
of the dataset. Those are moved to a std::vector instead if the
move constructor is noexcept as expected by std::vector,
- providing a template parameter pack for the same type of constructs
in place of the std::initializer_list
Some of the changes are also related to the fact that VS2013 does not
handle properly the std::initializer_list vs template parameter pack.
The dataset generators need to access runtime variables, which is
possible only after the test framework enters its setup.
The purpose of those changes is to instanciate the dataset and populate
the test tree during the initialization phase and not during the static
instanciation of the different test case:
- new delayed dataset type that is used for holding the parameters of a dataset
and its type. This dataset will be instanciated on demand (lazy construct)
- the test tree is now able to hold a generator until the init phase of the test
module. Once the init reached, the lazy datasets are instanciated and the test
tree populated with new tests.
- operations like zip do not require the size earlier than needed
