Option to use relative file paths in report

[teake]

Is your feature request related to a problem? Please describe.
The new relative_files config options toggle whether relative paths are stored in the data files, and also how they're reported in XML reports. However, file paths are still absolute in the CLI report (i.e. the one shown with coverage report).

Describe the solution you'd like
A possibility to have the CLI report respect relative_files and show relative paths, either by default or through a toggle.

Describe alternatives you've considered
Using sed to remove the prefix in reported file paths.

Additional context
Related issue: #948.

[teake]

Also: thanks for the great package!

[warsaw]

I think this issue describes the problem I am having too. I reached for relative_files but it didn't solve my problem. For $reasons, our build system wants to run coverage (through pytest) in a different directory, against the the installed site-packages location of our code inside a venv. While I'm not super happy about that, I can't change it. But this leads to ugly and useless absolute paths in the coverage console report:

---------- coverage: platform darwin, python 3.7.7-final-0 -----------
Name                                                                                                                                                         Stmts   Miss Branch BrPart  Cover   Missing
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
/Users/bwarsaw/projects/xx/xxxx-xxxxxxx_trunk/build/xxxx-xxxxxxx/environments/satellites/project/lib/python3.7/site-packages/xxxxxxxx/xxxxxxx/__init__.py        0      0      0      0   100%
/Users/bwarsaw/projects/xx/xxxx-xxxxxxx_trunk/build/xxxx-xxxxxxx/environments/satellites/project/lib/python3.7/site-packages/xxxxxxxx/xxxxxxx/constants.py      19      0      0      0   100%
/Users/bwarsaw/projects/xx/xxxx-xxxxxxx_trunk/build/xxxx-xxxxxxx/environments/satellites/project/lib/python3.7/site-packages/xxxxxxxx/xxxxxxx/exception.py       5      0      0      0   100%
/Users/bwarsaw/projects/xx/xxxx-xxxxxxx_trunk/build/xxxx-xxxxxxx/environments/satellites/project/lib/python3.7/site-packages/xxxxxxxx/xxxxxxx/xxxxxxx.py        67     10     36      7    82%   89->90, 90, 101->102, 102, 115->121, 134->135, 135, 137->161, 141->142, 142, 144->145, 145, 161-162, 168-171
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
TOTAL                                                                                                                                                           91     10     36      7    85%

I would love to find a way to chop off everything left of and including the site-packages component.

[cnzhujie]

This problem also happened in the html report

[nedbat]

See also #277.

[nedbat]

Is it enough to use the relative file as stored?

[teake]

I think so. But where are they stored? I cannot find any relative paths in the data file. I'm running 6.1.1 on Python 3.10.0 via

$ MY_DIR=$(pwd)
$ cd /tmp
$ PYTHONPATH=$MY_DIR coverage run --rcfile=$MY_DIR/.coveragerc --source=$MY_DIR -m pytest $MY_DIR/tests

with relative_files = True in .coveragerc. Am I overlooking something?

[ofek]

Would this account for src-layout structures? I would like src/pkg to become just pkg.

[adam-azarchs]

As others have already noted, this applies to the html report type as well as the command line report. It also applies to the lcov report, which is important for integrating into bazel (to avoid needing this hack). I do believe from my experimentation that the relative file as stored is good enough there.

[alexeagle]

I spent some time studying the situation for lcov, as relates to Bazel. In fact coveragepy already tries to make the SF: line in lcov report to be a relative path, calling relative_filename() here
https://github.com/nedbat/coveragepy/blob/master/coverage/lcovreport.py#L180

However they are relative to the working directory at the time the cov.start() initializes, here
https://github.com/nedbat/coveragepy/blob/master/coverage/files.py#L41
and Bazel actions run under a "sandbox" folder as the working directory, which contains symlinks to the real source tree. The canonical filename of the source files is passed through os.path.realpath via this call
https://github.com/nedbat/coveragepy/blob/master/coverage/files.py#L86

So, like issues with many tools under Bazel, it's the symlink handling that's going wrong in producing the desired report, possibly not the same issue as the OP here.

[jab]

possibly not the same issue as the OP here

related: #1674 (the issue I opened last time I ran into something like aspect-build/rules_py#524)

[nedbat]

@alexeagle have you tried using a [paths] configuration to map files from one location to another? https://coverage.readthedocs.io/en/7.6.11/config.html#paths

[alexeagle]

Thanks @nedbat - no I haven't tried that. I'm not sure when/if I'll get time but it looks promising.