test_ok2/TODO.txt

295 lines
9.4 KiB
Plaintext

Things to do for 1.0.0
=========================
- ease building of py lib, only depend on python interp
- separate py.magic.greenlet into its own project, and remove c-extension
- scale down "magic" impressions & improve py.test:
- documentation entry points for users
- documentation entry points for extenders
- advertise plugins instead of conftest!
- new config system
- small things: --nomagic -> --no-reassert
- deprecate py.magic
py.test
--------------
- compatilibity: honour/warn item.run() method of test items (so far
probably only execute() is warned about or the other way round)
- introduce plugin arch, port existing things to plugins:
- importorskip
- filelog
- chtmpdir per method
- apigen
- xfail
- acceptance/assertlines
- dist-testing?
- introduce setuptools-style version checking, at least
for py lib itself, maybe also for other packages:
py.checkversion("py>=1.0")
- generative tests: it is somewhat misleading for
the classical htmlviews that generated tests are identified
by numbers rather than by its parameters. unclear how
to fix this because we may not always be able to assume
that we can identify a generated tests by using its parameters
(it might not be hashable, doesn't have a sensical repr ...?)
- turn deprecation / apiwarnings into events, report them at the end?
- get APIGEN back to work
- get web reporter back to work
- introduce decorator "shouldfail" or "xfail"
as to mark a test as "expected to fail",
report specially if it surprisingly passes
- nightly test runs on multiple platforms
- review and refactor architecture of py.test with particular
respect to:
- writing (stacked) extensions / plugins (compared to Nose)
- porting existing extensions (htmlconftest / buildbot / PyPy's conftest's ...)
- fast and stable distributed testing
- reliable cross-platform testing
- improve py.test documentation to reflect new
event architecture
- review and optimize skip-handling (it can be quite slow in
certain situations because e.g. setup/teardown is fully performed
although we have "skip by keyword" and could detect this early)
py.execnet
--------------
- cross-python version (2.2/2.3-2.5/6) and cross-platform testing of
setup/teardown semantics
- optimize general setup and rsync timing?
py.apigen
----------------
- make it work again
see apigen_refactorings.txt
- check out CodeInvestigator
http://codeinvestigator.googlepages.com/main
or other code that collects data from running a program
(in our case running the tests)
Criticism and solutions
--------------------------------
"too big":
- too much code, you need entire py lib, hard to include into app
+ have a small pytest boostrap that loads pylib.zip from net
+ provide smaller script ala simpy
- lots of cmdline options, possibilities, documentation
rather unsorted
"needless differences between py.test and nosetests":
- py.test.skip
- raises
"tutorial structure missing", e.g.:
- "how to get started" in a minimal way, also how to use
existing conftests/plugins
- how to configure py.test
- how to write plugins/extensions
"too much magic"
- re-execution of assert expressions
+ rename "--nomagic" to something that turns off "superassertions"
+ hint at --tb=...
- get rid of py/magic directory
has a good ui but could be better
- support developer communication, e.g. py.test
--sendfailures=freenode-pypy
--sendfailures=pocoo # prints out paste.pocoo.url with traceback
- graphical interface, probably QT
- generally store test results and use them for subsequent calls
ld (review and shift to above)
=================================
refactorings
------------------
- refine doctests usage (particularly skips of doctests if
some imports/conditions are not satisfied)
- check if it works on win32
- refine error reporting (don't show python tracebacks)
- generalization of "host specifications" for execnet and
py.test --dist usages in particular (see also revision 37500 which
contained a draft for that). The goal is to have cross-platform
testing and dist-testing and other usages of py.execnet all
use a common syntax for specifiying connection methods and
be able to instantiate gateways/connections through it.
- unification of "gateway"/host setup and teardown, including
rsyncing, i.e. cross-platform and dist-testing.
- py.log: unify API, possibly deprecate duplicate ones,
base things on a Config object (hte latter almost a feature though)
(M988)
- see to teardown more eagerly
features
--------------
- (Harald Armin Massa): make py2exe work with py lib
- optimize file checking with --looponfailing (harald has code for win32)
- have a py.test scan/run database for results and test names
etc. (to allow quicker selection of tests and post-run
information on failures etc.) (M760)
- have config options from environment, command line or conftest's
have py/doc/config/"OPTNAME".txt for each option pypy-style
py.test --showconfig shows current configuration according
to envvars, cmdlineopts and conftests considered for your dir location.
py.test --help-conftest lists all possible environment envs
py.test --help-env lists all possible environment envs
- consider features of py.apigen (recheck closed "M1016")
- integrate rlcompleter2 (make it remotely workable)
and maybe integrate with "pdb" / pdbplus (M975)
- integrate native collecting of unittest.py tests from py.test
(along the PyPy lib-python tests) (M987)
- provide an automated conversion script helper for converting
unittest.py based tests to py.test ones. (M987)
- references from ReST docs to modules, functions and classes
of apigen generated html docs (M960)
- review svn-testing (and escape characters), consider
svn-bindings (M634)
- py.test.pdb - there is my hack for a while now, which integrates
rlcompleter2 with pdb. First of all it requires some strange changes
to rlcompleter itself, which has no tests. Long-term plan would be
to have pyrepl+rlcompleter2+pdb fixes integrated into pylib and
have it tested. This requires work though.
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
--- below neeeds more review ---
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
More random notes, goals
--------------------------------
- REDUCE "MAGICALNESS", from an IRC discussion with ronny:
- integration into IDEs
- python2.6/python3 compat?
- initpkg exports
- assert reinterpretation
- greenlet compiles at runtime only in "dev-mode"
- conftest's are "scary", hum, what about
nicefications:
looponfailing shoudl nicely signal tests that failed but now PASS tests
rename Node to itemtestloop
refactor config and session tests to go into test_distsession.py
raises DID NOT RAISE: report the return value
have node.shutdown perform out-of-band so that shutdowns happens more quickly
tracebacks of importerrors of test modules should start with the test module file
- fix hostmanage to care for setting PYTHONPATH properly
- BRANCH:
adding of options
- BRANCH:
TEMPDIR handling
syspath handling (notify on changes, restore for each test?)
- COMPLETE REPORTING FOR MERGE!
- remove ItemStart and CollectionStart
which are only needed for collectonly.
implement it some other way.
- pre-counting of test items
- move assert reinterp back to session?
- merge "--tb" and "--fulltrace" option, --tb=full
- implement --showouterr, don't show outerr by default
- domainpath?
- translate remote filenames to local filenames so that,
probably based on option
- review safe_repr
- move OutcomeRepr.where/exconly attr to ReprExceptionInfo
or substitute as LocationRepr?
- reprcrash rename message to exconly
- test terminal reporter
ACCEPTANCE test for eventlog writing
ACCEPTANCE test for acceptance custom reporting :)
ACCEPTANCE showing nice Collection Errors
ACCEPTANCE test for "py.test2" exit signals
ACCEPTANCE test for "py.test2 --traceconfig"
ACCEPTANCE test for nice reprsentation of failures during Collection
ACCEPTANCE test for nice reprsentation of failures during Generator Collection
ACCEPTANCE test for "py.test2" conftest containing syntax errors
ACCEPTANCE test for "py.test2" honouring conftest specifying "extrainfo"
ACCEPTANCE test for "py.test2" on a simple example project
- merge CollectionFinish and ItemTestReport
maybe with base class: BaseReport and some common attrs/methods
- merge terminal/remote and dist-testing
and make allocation of tests to hosts more dynamic
- time setup/teardown and the actual test runs separately
ACCEPTANCE:
py.test2 --
py.test2 --repeat=10
- expect failing tests
py.test2.expectfail(feature=138)
while killing a process:
Exception in thread receiver:
Traceback (most recent call last):
File "threading.py", line 460, in __bootstrap
self.run()
File "threading.py", line 440, in run
self.__target(*self.__args, **self.__kwargs)
File "/home/hpk/py/branch/event/py/execnet/gateway.py", line 140, in _thread_receiver
self._stopsend()
File "/home/hpk/py/branch/event/py/execnet/gateway.py", line 329, in _stopsend
self._send(None)
File "/home/hpk/py/branch/event/py/execnet/gateway.py", line 147, in _send
self._io.close_write()
File "/home/hpk/py/branch/event/py/execnet/inputoutput.py", line 106, in close_write
self.outfile.close()
IOError: [Errno 32] Broken pipe