I have a very large python project with a very large test suite. Recently we have decided to quantify the quality of our test-coverage.
I'm looking for a tool to automate the test coverage report generation. Ideally I'd like to have attractive, easy to read reports but I'd settle for less attractive reports if I could make it work quickly.
I've tried Nose, which is not good enough: It is incompatible with distribute / setuptools' namespace package feature. Unfortunately nose coverage will never work for us since we make abundant use of this feature. That's a real shame because Nose seems to work really nicely in Hudson (mostly)
As an alternative, I've heard that there's a way to do a Python coverage analysis in Eclipse, but I've not quite locked-down the perfect technique.
Any suggestions welcome!
FYI we use Python 2.4.4 on Windows XP 32bit
Have you tried using coverage.py? It underlies "nose coverage", but can be run perfectly well outside of nose if you need to.
If you run your tests with (hypothetically) python run_my_tests.py, then you can measure coverage with coverage run run_my_tests.py, then get HTML reports with coverage html.
From your description, I'm not sure what problem you had with nose, especially whether it was a nose issue, or a coverage.py issue. Provide some more details, and I'm sure we can work through them.
Ned has already mentioned his excellent coverage.py module.
If the problem you're having is something nose specific, you might want to consider using another test runner. I've used py.test along with the pytest_coverage plugin that lets you generate coverage statistics. It also has a pytest_nose plugin to help you migrate.
However, I don't understand exactly what the problem you're facing is. Can you elaborate a little on the "distribute / setuptools' namespace package feature" you mentioned? I'm curious to know what the problem is.
Related
I am in the process of trying to integrate code coverage into our development pipeline, and we are having a hard time using nosetest to accurately estimate our code coverage due to the fact that it also checks coverage for each of our imported libraries in our python packages. So we end up with code coverage percentages for things like import os, which is not allowing us to really see the data we want.
I have looked into using coverage.py, but that's in the very early stages.
I thought to ask if anyone else has had this issue and how they overcame it.
Thanks in advance!
Nosetests has an option to only produce coverage for named packages e.g.
--cover-package=foo --cover-package=bar
This is what I've done in the past.
However, I moved over to pytest. I liked this better because it produced better error message, including dictionary diffs.
We have been using Django for a long time. Some old code is not being used now. How can I find which code is not being used any more and remove them.
I used coverage.py with unit tests, which works fine and shows which part of code is never used, but the test covered is very low. Is there any way to use it with WSGI server to find which code have never served any web requests?
It depends on what you mean by unused code.
For unreachable dead code, like functions are were never called, classes that are never instantiated, you can use a pure static code analyzer to find them. Pylint is a good option. Bear in mind that this is not 100% accurate, false positive is possible:
# static analysis can't detect methods called this way
func = getattr(obj, "func_name")
func()
For code that are reachable, but never reached. You have to rely on tools like coverage.py, and improve your test coverage.
On a well tested project, coverage would be ideal but with some untested legacy code I don't think there is a magical tool.
You could write a big test loading all the pages and run coverage to get some indication.
Cowboy style:
If it's not some critical code and you're fairly sure it's unused (i.e. not handling payments, etc.). Comment it out, check that the tests pass, deploy and wait a week or so before removing it definitely (or putting it back if you got a notification).
As other answers indicate coverage.py accurately finds out which parts of the code are never executed, but coverage requires your code to be actually run to perform the analysis. Vulture on the other hand, runs static analysis for finding dead (unused code) for Python Programs. Also, if you run vulture on both your library and test suite, you might be able to find untested code.
Vulture is a standard PyPI package and can be installed using pip:
$ pip install vulture
Run vulture using the command:
$ vulture apps/ tests/ --exclude settings
Bear in mind that due to Python's dynamic nature, there may be some false positives, but they can be dealt with by the means of "Whitelists" - Please refer to this answer by Rahul for more information on how to use Vulture with django and tackle false positives.
I'm writing quite a few unit tests and using nosetests to run them. Nose certainly makes it nice and easy to run tests, but the output can be pretty cluttered at the best of times, and a downright mess at others, depending on warnings and errors.
I was wondering if there are any visual tools that wrap nose so that the feedback cleaner. A results grid showing test names, run times and indicating success/failure status with colours would be a huge visual aide. Better still, if it could split up the error messages and tracebacks on a case by case basis, it would really cut down on the amount of clutter when a large number of tests fail on a major change.
This is starting to read like a wishlist, but does anything even close to this exist?
There's a very similar question here.
It doesn't look like there's a standardized GUI for nosetests (as in, standardized for just the python interface). There seems to be a few GUI plugins for nosetests depending on which IDE you're using - a quick search brought up this link for Komodo.
However, there are many other GUI based unit testing frameworks out there for Python - have you tried PyUnit?
Well, what you can do is to use the jenkins integration server, we use it with py.test and phpunit (it is really simple to use it with any other testing framework), it provides us some fancy (and useful!) graphs containing
Number of tests that run
Number of tests passing/failing
The stack trace for the failed tests
Execution times for tests
Code coverage for the tests
You can also relate it to git/svn/mercurial...
Here you have an example of how to configure hudson (for the record, jenkins is a fork of hudson, so the instructions are still valid :), in order to provide the elements from your wishlist.
An easier option is to use a IDE as PyCharm, it has integration with nosetests, and will tell you which tests passed/failed and its times, however, jenkis stores historical data.
Good luck :)
Have you taken a look at the Nose plugins page? A quick glance revealed at least a couple that stated they could provide a GUI interface.
In C++ I have compiler that tell me if something wrong with my code after refactoring. How to make sure that Python code is at least correct after changes? There may be some stupid error like wrong function name etc. that pretty easy to find in compile time.
Thanks
Looks like PyChecker or pylint are what you're looking for
use editor / IDE that supports code highlighting. E.g., Notepad++ has word-highlighting feature that I find very useful.
use unit tests
stupid errors will be weeded out first, so I wouldn't worry to much about this type of errors. it's "smart" error you should be afraid of.
Use tools such as pylint or PyChecker.
Write unit tests.
Unit test. http://docs.python.org/library/unittest.html
If your tests are written at a reasonable level of granularity, it can be as fast to unit test as it is to run lint or a compiler.
Static analysis (as from the IDE, or from tools like pyLint and pyChecker) is a very quick and effective way to check simple errors, and enforce a common style.
Unit tests are a great way to ensure the code stands for its contract.
Code reviews and pair programming are one of the best ways to find errors of all sorts, and to spread knowledge in a team.
All of the options require some time, to setup and to execute. However, the gains are tremendous, and far higher than the investment.
Eclipse has a good python plugin for doing the syntax highlighting and debugging.
Pylint is almost doing what you are looking for.
You can also force the compilation of your python files. That will show some basic syntax error (it doesn't have all the capability of a c++ compiler)
I've read this article and decided to make an automated build system with pyDev and ant. It does the compilation of the python files and is running the unit tests. Next step is to integrate pylint to that process.
I hope it helps
As with other languages, you should use assertions liberally throughout your code. Use assertions when you must rely on the predicate to be true for the program to run, not as exception/error handling. An assertion should be used to check for irrecoverable errors and force the program to crash. More on assertions (and python error checking in general)
You may need this:
python -m py_compile script.py
You might also want to check out PEP8 as a style guide for Python Code.
I have already visited Preferred Python unit-testing framework. I am not just looking at Python Unit Testing Framework, but also code coverage with respect to unit tests. So far I have only come across coverage.py. Is there any better option?
An interesting option for me is to integrate cpython, unit testing of Python code and code coverage of Python code with Visual Studio 2008 through plugins (something similar to IronPython Studio). What can be done to achieve this? I look forward to suggestions.
We use this Django coverage integration, but instead of using the default coverage.py reporting, we generate some simple HTML:
Colorize Python source using the built-in tokenizer.
PyDev seems to allow code coverage from within Eclipse.
I've yet to find how to integrate that with my own (rather complex) build process, so I use Ned Batchelder's coverage.py at the command line.
There is also figleaf which I think is based on Ned Batchelder's coverage.py. We use nose as the driver for the testing. It all works pretty well. We write our unit tests using the built-in unittest and doctest modules.
NetBeans' new Python support has tightly integrated code coverage support - more info here.
If you want interactive code coverage, where you can see your coverage stats change in real time, take a look at Python Coverage Validator.
Testoob has a neat "--coverage" command-line option to generate a coverage report.