On Fri, 21 Apr 2017 at 15:33 Victor Stinner <victor.stinner@gmail.com> wrote:

Ah, I found a workaround: Firefox on Android has a "[x] See the computer version" option which allows the merge!?

Victor

Le 22 avr. 2017 12:29 AM, "Victor Stinner" <victor.stinner@gmail.com> a écrit :

Hi,

I tried to merge a pull request on my phone, but I got the error:

"Pull requests that have a failing status can’t be merged on a phone."

Well that's annoying.

 

The GitHub PEP announced that it will be possible to merge a change from the beach. Well, it's doable but only if you bring a laptop, not a phone :-)

All tests pass except Codecov which is unstable. On a computer, I can merge such PR.

What is the status of Codecov? Is someone actively working on fixing it to make it more reliable.

https://github.com/python/core-workflow/issues/38 is tracking trying to fix the inconsistency problem and https://github.com/python/core-workflow/issues/18 is tracking getting a more complete coverage report.

I would like to see this fixed and it's on my workflow todo list, but obviously that list is not short so I don't know when I will get to it (especially since at some point I will need to take a workflow break and actually write code and review PRs again ;) .

 

I dislike code coverage in general, even more when it's run in a CI.

Can we change the merge policy to allow merge on a phone? Or can we fix Codecov?

We might be able to turn off the status check. The config file is https://github.com/python/cpython/blob/master/.codecov.yml and the docs are at https://docs.codecov.io/docs/codecov-yaml (I'm on vacation so I don't have time to look up the exact config change since I really shouldn't even be replying to this email ;) .

 

Note: the PR is

https://github.com/python/cpython/pull/1237#pullrequestreview-34044762

Codecov says "10% of diff hit (target: 100%)". The newly added code is tested on Windows on release build. Maybe Codecov only tests on Windows?

Codecov actually only runs under Travis, so it's only testing on Linux.

 

I dislike code coverage because there is a temptation to write artficial tests whereas the code is tested indirectly or the code is not important enough to *require* tests.

I personally disagree as code that isn't tested isn't checked that at least some semblance of backwards-compatibility is being kept. Now I'm not advocating we must have explicit tests for every line or code, but I think we should somehow exercise as much code as possible.