"Erase" still needed in 4.2

[nedbat]

Originally reported by Michał Bultrowicz (Bitbucket: Butla, GitHub: Butla)

---

If I understand http://nedbatchelder.com/blog/201607/coveragepy_42.html correctly, then running "run" and then "combine" (I'm running parallel tests) should properly detect loss of test coverage, right? Because they shouldn't look at the old .coverage by default.

But at the moment I still need to run "erase" to get rid of the old results before running the tests.

Here's my full tox.ini: https://github.com/butla/PyDAS/blob/a8c76d3221447fb34af60f5d03b18df52286ed4c/tox.ini

---

• Bitbucket: https://bitbucket.org/ned/coveragepy/issue/510

[nedbat]

@butla Can you provide me with a reproducible scenario? I tried running your tests, and got many errors like this: docker.errors.DockerException: Error while fetching server API version: ('Connection aborted.', FileNotFoundError(2, 'No such file or directory'))

I see that you used to have two "coverage erase" lines and now you have only one?

[nedbat]

Original comment by Michał Bultrowicz (Bitbucket: Butla, GitHub: Butla)

---

@nedbat Well, the tests require Docker. I have Xubuntu 14.04 with Docker 1.6.2 installed.

If you don't want to setup Docker, you can use a Vagrant box with Docker (this of course requires Vagrant and Virtualbox):

#!bash

vagrant init williamyeh/ubuntu-trusty64-docker
vagrant up --provider virtualbox
vagrant ssh
sudo apt-get update
sudo apt-get install python3-pip
sudo pip3 install tox
git clone https://github.com/butla/pydas
cd pydas
tox

Is this OK or should I provide a simpler scenario?

About two "coverage erase" - yes, I needed two and I reported this behavior in #492, but it's fixed in 4.2.

[nedbat]

Original comment by Michał Bultrowicz (Bitbucket: Butla, GitHub: Butla)

---

Oh, right, you need to remove "coverage erase" from tox.ini before
rerunning the tests.

As I understand in 4.2 erase shouldn't be necessary right? Because
appending to existing .coverage is optional?

[nedbat]

Hmm, the problem is that "tox" and "pylint" each run with coverage measurement (because of coverage_pth), and they each read the .coverage file before running. This carries the .coverage data into a parallel .coverage.localhost.xxx file, which then gets combined by the combine step.

Seems like reading the .coverage file during auto-started coverage measurement is a bad thing.

[nedbat]

Original comment by Michał Bultrowicz (Bitbucket: Butla, GitHub: Butla)

---

Can .coverage be of any use in the parallel case? Everything get's stored in separate files, right?

[nedbat]

The recent auto_load changes fixed #510

→ <<cset fee2fd97520c (bb)>>

[nedbat]

This is fixed in the handful of commits ending with fee2fd97520c (bb)

[nedbat]

This fix was released as part of Coverage.py 4.3.

[nedbat]

Original comment by Michał Bultrowicz (Bitbucket: Butla, GitHub: Butla)

---

Great to hear that :)