The new "--tb=auto" option (default) will only display long tracebacks
for the first and last entry. You can get the old behaviour of printing
all entries as long entries with "--tb=long". Also short entries by
default are now printed very similarly to "--tb=native" ones.
This is consistent with how callable() is called from the rest of pytest code
(see _pytest/nose.py & _pytest/python.py) plus, as a nice side-effect, it
makes pytest work correctly on Python 3.1.
--HG--
branch : python 3.1 fix
Without this if a session scoped fixture fails it's setup it will
be re-tried each time it is requested. Especially in case of
skip or failure exceptions this can be undesirable, but caching
makes sense for all exceptions.
If the type is not checked then an incomprehensible error will occur
later. This enforces the type and raies the same exception/msg as
CPython does in that case.
Docstring unmodified, just re-justified for pep8 compat.
Replace the verbose per-test reporting format of `file:line test_name RESULT`
with the node ID of the test, i.e. `file@line::class::method[param] RESULT`.
This patch does not update the examples in the docs; @hpk42 has a script
to regenerate those.
--HG--
branch : issue504
The result from the pytest_assertrepr_compare hook should not include
any newlines since that will confuse the mini-formatting language used
by assertion.util.format_explanation. So simply escape the included
newlines, this way hook writers do not have to worry about this at
all.
Fixes issue 453.
Such reference cycles unnecessarily cause Python interpreter not to garbage
collect the objects referenced in those cycles as soon they could be collected,
and in turn cause the tests to use more memory than is strictly necessary.
--HG--
branch : break_ExceptionInfo_reference_cycles
line arguments by using the result of parse_known_args rather than
the previous flaky heuristics. Thanks Marc Abramowitz for tests
and initial fixing approaches in this area.
Prevent error on exit if some code messes with sys.meta_path and removes the
assertionrewrite hook (CaptureMock seems to do this):
File "/Users/marca/dev/hg-repos/pytest/_pytest/assertion/__init__.py", line 64, in pytest_unconfigure
sys.meta_path.remove(hook)
ValueError: list.remove(x): x not in list
--HG--
branch : sys_meta_path_remove_hook_only_if_present
The `=` is better because it encourages folks to use the form that doesn't
suffer from issue #436 (https://bitbucket.org/hpk42/pytest/issue/436), which
can cause the arg to be treated as an "anchor" and used as the (unexpected)
path for searching for conftest.py files.
--HG--
branch : help_show_args_with_equals_instead_of_space
use the original stream.
- avoid resetting capture FDs/sys.stdout for each test by keeping capturing
always turned on and looking at snapshotted capturing data during runtest
and collection phases.
between working at the front and at the end of a hook call chain.
The idea is to make it easier for a plugin to "wrap" a certain hook
call and use context managers, in particular allow a major cleanup of
capturing.
- node.warn() for a node-specific warning
- config.warn() for a global non-node specific warning
Each warning is accompanied by a "warning number" so that we can later
introduce mechanisms for surpressing them.
Each warning will trigger a call to pytest_report_warn(number, node, message)
which is by default implemented by the TerminalReporter which introduces
a new option "-rw" to show details about warnings.
If the compared text was in bytes and not actually valid text
(i.e. could not be encoded to text/unicode using the default encoding)
then the assertrepr would fail with an EncodingError. This ensures
that the internal string is always valid unicode, converting any bytes
safely to valid unicode. This is done using repr() which then needs
post-processing to fix the encompassing quotes and un-escape newlines.
This fixes issue 429.
Made it clearer that clearing such references is not mandatory and is only an
optional step which may help the Python interpreter speed up its garbage
collection.
--HG--
branch : document_ExceptionInfo_ref_cycle
pytest.raises() returns an ExceptionInfo object which, if a local reference is
made to it, forms a reference cycle:
ExceptionInfo
--> exception
--> stack frame raising the exception
--> current stack frame
--> current local variables
--> Exception Info
Such a reference cycle would then prevent any local variables in the current
stack frame, or any of its child stack frames involved in the same reference
cycle, from being garbage collected until the next reference cycle garbage
collection phase. This unnecessarily increases the program's memory footprint
and potentially slows it down.
This situation is based on a similar one described in the official 'try'
statement Python documentation for locally stored exception references.
--HG--
branch : document_ExceptionInfo_ref_cycle
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.
The only remaining 'py.test' references are:
* those referring to the 'py.test' executable
* those in code explicitly testing py.test/pytest module compatibility
* those in old CHANGES documentation
* those in documentation generated based on external data
* those in seemingly unfinished & unmaintained Japanese documentation
Minor stylistic changes and typo corrections made to documentation next to
several applied py.test --> pytest content changes.
The need for comparing two separately instantiated nodes seems to be historic
(related to an already-gone mode of pytest-xdist which would re-collect nodes)
and not actually needed anymore.
in a test invocation will have a corresponding FixtureDef instance.
also fixes issue246 (again).
simplify parametrized fixture teardown by making it work lazy:
during the setup of a parametrized fixture instance any previously
setup instance which was setup with a different param is torn down
before setting up the new one.
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.
This will make more sense if multiple fixtures depend on each other.
It would be better if all exceptions could be shown however.
Also depend on python 2.5+ exception hierarchy and use sys module
directly.
Starting with Python 3.3, NamespacePath passed to importlib hooks
seem to have lost the ability to be accessed by index.
We wrap the index access in a try..except and wrap the path in a
list if it happens.
Fixes#383.
@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.
The _pytest.config module itself is no longer a plugin but the actual
config instance is plugin-registered as ``pytestconfig``.
This allows to put most pytest specific logic to _pytest.config instead
of in the core pluginmanager.
implemented by a slight internal refactoring and the introduction
of a new hook ``pytest_exception_interact`` hook.
- fix issue341: introduce new experimental hook for IDEs/terminals to
intercept debugging: ``pytest_exception_interact(node, call, report)``.
monkeypatch location from target (can be class/module/function or
string which is taken as importable python path)
examples:
monkeypatch.replace(os.path.abspath, lambda x: "")
monkeypatch.replace("requests.get", ...)
```
timeit result for 10000 iterations of expanding '/d' (lowered the count in the code afterwards)
# 2.7.5 3.3.2
# FilesCompleter 75.1109 69.2116
# FastFilesCompleter 0.7383 1.0760
```
- does not display prefix dir (like bash, not like compgen), py.test /usr/<TAB> does not show /usr/bin/ but bin/
don't use autouse fixtures for now because it would cause a proliferation
and overhead for the execution of every test. Rather introduce a
node.addfinalizer(fin) to attach a finalizer to the respective node
and call it from node.setup() functions if the setup phase succeeded
(i.e. there is no setup function or it finished successfully)
Mathieu Agopian for fixing. The patch moves handling setUpClass
into a new autofixture. (XXX impl-decide if rather adding addfinalizer()
API to node's would have a similar effect)
When an autouse fixture in a plugin was encountered None was stored as nodeid
where it used to be ''. This broke the lookup of autouse fixtures later on.
This also adds another test for the normal fixture ordering which was slightly
wrong: a fixture without location was always added at the front of the fixture
list rather then at the end of the fixtures without location but before the
fixtures with location.
- separate out most argcomplete related stuff in new file _argcomplete.py
(could probably be in the py library)
- allow positional arguments to be interspaced with optional arguments
( + test in test_parseopt.py )
- removed double argument in tox.ini
- add documentation on installing argcomplete (>=0.5.7 as needed for
Python 3), might need improving/incorporation in index.
This does not work on 2.5 yet. I have patches for argcomplete
(with/print()/"".format) but I am not sure they will be accepted.
Agreed with hpk not to push for that.
Removing argcomplete and leaving completion code active now works by early
exit, so <TAB> no longer re-runs the programs without parameters
(which took long for py.test)
test calls bash with a script that redirects filedescriptor 8 (as used by
argcomplete), so the result can be tested.
--HG--
branch : argcomplete
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
Conftests are plugins with a location attached to them while other
plugins do not have a location. When ordering fixturedefs those from
plugins without a location need to be listed first.
- When not specifying ids, let None and bools use their native string form (like str, int, float) rather than obfuscated form used for objects
- When specifying ids, explicitly raise a ValueError if a different number of ids are specified compared to the test cases
- Add tests for both these items.
address some comments by @hpk42 on 0b9d82e :
- move tests into their own class, rename
- add test showing metafunc.parametrize called in pytest_generate_tests rather than as decorator
- add test and fix single-argname case
- convert two loops into one in parametrize()
also
- renamed 'input' to 'n', since 'input' is a built-in
pytest-rerunfailures plugins) by re-initializing and removing
request/funcargs information in runtestprotocol() - which is a slightly
odd place to add funcarg-related functionality but it allows all
pytest_runtest_setup/teardown hooks to properly see a valid
request/funcarg content on test items.
When the conftest.py files are looked for intially they got loaded
starting from the subdir ending at the parent dir(s). Later on during
collection any conftest.py files are loaded starting from the parent
dir ending at the subdir. Due to how extending fixtures works the
latter is correct as otherwise the wrong fixture will be available.
So this changes the initial conftest loading to start at the root and
go towards the subdir.
This does also affect the order of other hooks, hence the order of the
reporting being different in testing/test_terminal.py.
holger krekel
Jurko Gospodnetić
Benjamin Peterson
Marc Abramowitz
Christian Henz
christian@christian-linux.sarrazin.local
Sébastien Fievet
WooParadog
Brianna Laugher
Floris Bruynooghe
Andy Freeland
Anatoly Bubenkov
Daniel Grana
Chris Gilling
schlamar
Ronny Pfannschmidt
Simon Liedtke
Lukasz Balcerzak
Virgil Dupras
James Lan
Jason R. Coombs
Andreas Zeidler
Anthon van der Neut
Ronny Pfannschmidt
Andreas Pelme
Mathieu Agopian
holger krekel
Katarzyna Jachim
Christian Theune
variedthoughts
Brianna Laugher
Erik Bray
Wouter van Ackooy
Brianna Laugher
Danilo de Jesus da Silva Bellini
maho
Jaap Broekhuizen