A script to visualize coverage reports via HTML
This package contains tools to work with Python coverage data.
coveragereport produces HTML reports from coverage data, with syntax-highlighted source code and per-package aggregate numbers.
coveragediff compares two sets of coverage reports and reports regressions, that is, increases in the number of untested lines of code.
- Using coveragereport
- Using coveragediff
- Getting coverage data
- Frequently Asked Questions
$ coveragereport --help Usage: coveragereport [options] [inputpath [outputdir]] Converts coverage reports to HTML. If the input path is omitted, it defaults to coverage or .coverage, whichever exists. If the output directory is omitted, it defaults to inputpath + /report or ./coverage-reports, depending on whether the input path points to a directory or a file. Options: -h, --help show this help message and exit -q, --quiet be quiet -v, --verbose be verbose (default) --strip-prefix=PREFIX strip base directory from filenames loaded from .coverage --path-alias=PATH=LOCALPATH define path mappings for filenames loaded from .coverage
Example use with zope.testrunner:
$ bin/test --coverage=coverage $ coveragereport $ ln -s mypackage.html coverage/report/index.html $ xdg-open coverage/report/index.html $ xdg-open coverage/report/all.html
Example use with nose and coverage.py:
$ nosetests --with-coverage --cover-erase $ coveragereport --strip-prefix=/full/path/to/source/ $ ln -s mypackage.html coverage-reports/index.html $ xdg-open coverage-reports/index.html $ xdg-open coverage-reports/all.html
You need enscript installed and available in your $PATH if you want syntax highlighting.
Usage: coveragediff [options] olddir newdir Options: -h, --help show this help message and exit --include=REGEX only consider files matching REGEX --exclude=REGEX ignore files matching REGEX --email=ADDR send the report to a given email address (only if regressions were found) --from=ADDR set the email sender address --subject=SUBJECT set the email subject --web-url=BASEURL include hyperlinks to HTML-ized coverage reports at a given URL
Usage example with zope.testrunner:
$ bin/test --coverage=coverage $ vi src/... $ mv coverage coverage.old $ bin/test --coverage=coverage $ coveragediff coverage.old coverage
You cannot use coveragediff with coverage.py data. More on that below.
$ coveragediff coverage.old coverage my.package.module: 36 new lines of untested code my.package.newmodule: new file with 15 lines of untested code (out of 23)
Output with clickable links:
$ coveragediff coverage.old coverage --web-url=http://example.com/coverage my.package.module: 36 new lines of untested code See http://example.com/coverage/my.package.module.html my.package.newmodule: new file with 15 lines of untested code (out of 23) See http://example.com/coverage/my.package.newmodule.html
Output via email, convenient for continuous integration:
$ coveragediff coverage.old coverage --web-url=http://example.com/coverage \ --email 'Developers <email@example.com>' \ --from 'Buildbot <firstname.lastname@example.org>'
That last example doesn’t produce any output, but sends an email (via SMTP to localhost:25).
zope.testrunner can produce a directory full of files named dotted.package.name.cover that contain source code annotated with coverage information. To get them, use
Both coveragereport and coveragediff accept this as inputs.
coverage.py can produce a .coverage file containing (incomplete) coverage information. To get it, use
coverage run bin/testrunner
coveragereport can take the .coverage file as an input, but it also needs access to the matching source files. And you have to manually specify the absolute pathname prefix of your source tree so that the report know how to translate filenames into dotted package names. Also, it’s not enough to have absolute pathnames, you need to supply the canonical absolute pathname (with no symlink segments), such as returned by os.path.realpath. This is very inconvenient. Sorry.
coveragediff is unable to compare two .coverage files and report regressions. One reason for that is the incompleteness of the data format (it line numbers of executed statements, but doesn’t say which lines contain code and which ones are blank/comments/continuation lines/excluded source lines). The other reason is simpler: nobody wrote the code. ;)
Unfortunately coverage annotate does not produce files compatible with coveragereport/coveragediff. This could also be remedied if somebody wrote a patch.
If you want to use a .coverage file produced on another machine or simply in a different working directory, you will need to tell coveragereport how to adjust the absolute filenames so that the sources can be found. Use the --path-alias option for that. Alternatively you could use coverage combine to manipulate the .coverage file itself, as described in the documentation.
Some people prefer the look of the reports produced by z3c.coverage. Some people find per-package coverage summaries or the tree-like navigation convenient.
coverage.py is much faster, but using it (and hooking it up to z3c.coverage) is perhaps less convenient. E.g. if you use zc.buildout 1.5.x with zc.recipe.testrunner, you will be unable to use coverage run bin/test because of mystic semi-broken site isolation magic of the former.
Does asking myself count?
- Support (and require) coverage.py >= 4.0.
- Fix incorrect highlighting of missed lines when processing coverage.py data (broken since 2.0.2).
- Dropped Python 2.6 support.
- Added Python 3.4, 3.5 and 3.6 support.
- Standardize namespace __init__.
- Bug: coveragereport now also accepts non-ASCII chars in source files (also fix the case when enscript is not available).
- Bug: coveragereport now also accepts non-ASCII chars in source files.
- Added Python 3.3 and PyPy 1.9 support.
- Dropped Python 2.4 and 2.5 support.
- Nicer PyPI description. Doctests are tests, not docs.
- coveragereport now accepts --path-alias.
- coveragereport: new color step between yellow (90%) and green (100%), a yellowish-green (95%).
- coveragereport now accepts --help, --verbose and --quiet options, with verbose being on by default.
- coveragereport now can handle .coverage files produced by http://pypi.python.org/pypi/coverage
- Bugfix: sorting by numbered of uncovered lines was broken in the all.html report.
- Rename the coverage script to coveragereport, to avoid name clashes with Ned Batchelder’s excellent coverage.py.
- Bug: Doctest did not normalize the whitespace in coveragediff.txt. For some reason it passes while testing independently, but when running all tests, it failed.
- Bug: When a package path contained anywhere the word “test”, it was ignored from the coverage report. The intended behavior, however, was to ignore files that relate to setting up tests.
- Bug: Sort the results of os.listdir() in README.txt to avoid non-deterministic failures.
- Bug: The logic for ignoring unit and functional test modules also used to ignore modules and packages called testing.
- Change “Unit test coverage” to “Test coverage” in the title – it works perfectly fine for functional tests too.
- Bug: When the package was released, the test which tests the availability of an SVN revision number failed. Made the test more reliable.
- Feature: The main() coverage report function now accepts the arguments of the script as a function argument, making it easier to configure the script from buildout.
- Feature: When the report directory does not exist, the report generator creates it for you.
- Feature: Eat your own dog food by creating a buildout that can create coverage reports.
- Bug: Improved the test coverage to 100%.
- Bug: Fixed meta-data.
- First public release.
- Feature: Added the --web option to coveragediff.
- Feature: Added a test suite.
- Feature: Added coveragediff.py.
- Initial release of coveragereport.py.