As discussed in the mailing list, unfortunately this might break APIs
due to the subtle differences between new and old-style classes (see #2398).
This reverts commit d4afa1554b from PR #2179.
This addresses ref https://github.com/pytest-dev/pytest/issues/1954.
The current truncation for assertion explanations does not deal with long lines
properly:
- Previously if lines were too long it would display a "-n more lines"
message.
- 999e7c6541 introduced a bug where long lines can
cause index errors if there are < 10 lines.
Extract the truncation logic into its own file and ensure it can deal with
long lines properly.
Part two of https://github.com/pytest-dev/pytest/issues/1512. Update the format
of the truncation message to help make it clear that pytest truncates the
entire assertion output when verbosity < 2.
The assertion reinterpretation is an old backwards compatibility
mode which was no longer being maintained on feature-parity with
the assertion rewriting mode. It was also responsible for some
dubious patching of builtins and test with side-effects would
suddenly start passing. Since re-writing has been the default for
a long time and plugins are now also re-written it is time to
retire reinterpretation.
Hook up the PEP 302 import hook very early in pytest startup so
that it gets installed before setuptools-installed plugins are
imported. Also iterate over all installed plugins and mark them
for rewriting. If an installed plugin is already imported then
a warning is issued, we can not break since that might break
existing plugins and the fallback will still be gracefull to
plain asserts.
Some existing tests are failing in this commit because of the new
warning triggered by inline pytest runs due to the hypothesis
plugin already being imported. The tests will be fixed in the next
commit.
When you don't get enough information with a test running on a CI, it's quite
frustrating, for various reasons:
- It's more likely to be a flaky test, so you might not be able to reproduce
the failure.
- Passing -vv is quite bothersome (creating a temporary commit and reverting
it)
For those reasons, if something goes wrong on CI, it's good to have as much
information as possible.
for the already existing cleanup logic of the config object.
This simplifies lifecycle management as we don't keep two
layers of shutdown functions and also simplifies the pluginmanager
interface.
also add some docstrings.
--HG--
branch : plugin_no_pytest
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.
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
Before this was only done at the time the assertion plugin was loaded.
This lead to counter-intuitive behaviour where two subdirectories with
a pytest_assertrepr_compare hook in their conftest.py would not work,
only one would ever be used.
This defers assiging the _pytest.assertion.util._reprcompare function
until the item is loaded (pytest_runtest_setup) so that it can use the
hookrelay of the test item to find the appropriate
pytest_assertrepr_compare hook for the item.
This fixes issue #77.
Use load_module on the import hook to load the rewritten module. This allows the
removal of the complicated code related to copying pyc files in and out of the
cache location. It also plays more nicely with parallel py.test processes like
the ones found in xdist.