This was documented before, but never enforced. Passing non-strings could
have strange side-effects and enforcing a string simplifies other
implementation.
To keep existing tests which emit RemovedInPytest4Warnings running, decided
to go with a command line option because:
* Is harder to integrate an ini option with tests which already use an ini file
* It also marks tests which need to be removed/updated in 4.1, when
RemovedInPytest4Warning and related functionality are removed.
Fix#3737
Optionally raise an exception when parametrize collects no arguments.
Provide the name of the test causing the failure in the exception
message.
See: #3849
We eventually want to replace py.io.TerminalWriter (exposed as
"_tw" in TerminalReporter) by something else, so deprecating access
to "_tw" and making it public as "writer" is contrary to that.
this allows a clear addition of parameterization parameters that carry along marks
instead of nesting multiple mark objects and destroying the possibility of creating
function valued parameters,
it just folders everything together into one object carrfying parameters, and the marks.
at the beginning of strings and even that is deprecated. Use "not" instead.
This should allow to pick parametrized tests where "-" appeared in the parameter.
--HG--
branch : issue557
When pytest is called with the '--markers' option, it should collect marker
information from the current folder, and they should get loaded and used
correctly before the '--markers' output is constructed.
When a MarkDecorator instance is called it does the following:
1. If called with a single class as its only positional argument and no
additional keyword arguments, it attaches itself to the class so it gets
applied automatically to all test cases found in that class.
2. If called with a single function as its only positional argument and no
additional keyword arguments, it attaches a MarkInfo object to the
function, containing all the arguments already stored internally in the
MarkDecorator.
3. When called in any other case, it performs a 'fake construction' call, i.e.
it returns a new MarkDecorator instance with the original MarkDecorator's
content updated with the arguments passed to this call.
When Python applies a function decorator it always passes the target class/
function to the decorator as its positional argument with no additional
positional or keyword arguments. However, when MarkDecorator was deciding
whether it was being called to decorate a target function/class (cases 1. & 2.
as documented above) or to return an updated MarkDecorator (case 3. as
documented above), it only checked that it received a single callable positional
argument and did not take into consideration whether additional keyword
arguments were being passed in as well.
With this change, it is now possible to create a pytest mark storing a function/
class parameter passed as its only positional argument and accompanied by one or
more additional keyword arguments. Before, it was only possible to do so if the
function/class parameter argument was accompanied by at least one other
positional argument.
Added a related unit test.
Updated MarkDecorator doc-string.
filtering with simple strings that are not valid python expressions.
Examples: "-k 1.3" matches all tests parametrized with 1.3.
"-k None" filters all tests that have "None" in their name
and conversely "-k 'not None'".
Previously these examples would raise syntax errors.
Also add a note to the docs about what is allowed.
@pytest.mark.some(lambda arg: ...)
def test_function():
would not work correctly because pytest assumes @pytest.mark.some
gets a function to be decorated already. We now at least detect if this
arg is an lambda and thus the example will work. Thanks Alex Gaynor
for bringing it up.
that argparse does not have Option objects -> added class Argument
Needed explicit call of MyOptionParser.format_epilog as argparse
does not have that. The parse_arg epilog argument wraps the text,
which is not the same (could be handled with a special formatter).
- parser.parse() now returns single argument (with positional args in
.file_or_dir)
- "file_or_dir" made a class variable Config._file_or_dir and used in help and tests
- added code for argcomplete (because of which this all started!)
addoption:
- if option type is a string ('int' or 'string', this converted to
int resp. str
- if option type is 'count' this is changed to the type of choices[0]
testing:
- added tests for Argument
- test_mark.test_keyword_extra split as ['-k', '-mykeyword'] generates argparse
error test split in two and one marked as fail
- testing hints, multiline and more strickt (for if someone moves format_epilog
to epilog argument of parse_args without Formatter)
- test for destination derived from long option with internal dash
- renamed second test_parseopt.test_parse() to test_parse2 as it was
not tested at all (the first was tested.)
--HG--
branch : argparse
Daniel Hahler
Ronny Pfannschmidt
Bruno Oliveira
Anthony Sottile
Ronny Pfannschmidt
Jeffrey Rackauckas
Andrea Cimatoribus
Ondřej Súkup
feuillemorte
Sviatoslav Abakumov
Leonard Lausen
Xuan Luong
Andras Tim
Vitaly Lashmanov
Michal Wajszczuk
palaviv
jakubo
Anatoly Bubenkov
holger krekel
Jurko Gospodnetić
Lukasz Balcerzak
Anthon van der Neut
Wouter van Ackooy