Travis CI: macOS does not include Python anymore

[imatlopez]

Checklist

• commit message follows commit guidelines

Description of change

macOS stopped bundling python, this attempts to bring back travis tests on macOS

[cclauss]

Thanks for doing this but please use pyenv install Python on Travis CI. It is way faster than brew and supports all Python versions including Python 3.8 which brew is still struggling to support.

Try pyenv install --list to see all the versions available.

[cclauss]

Travis is dropping matrix: for jobs: so let's lose the matrix around line 4...

addons: homebrew: update: true packages: - npm - pyenv jobs: 

matrix:

This will ensure that all macOS jobs have access to the latest npm and pyenv.

[imatlopez]

@cclauss got it working, cleaning up. For pyenv to work the PATH needs to be changed and for python modules to work they need the python -m prefix

(there is no better way to test this is there?)

[imatlopez]

@cclauss ready for review :)

[rvagg]

I've force pushed out the HEAD commit on master, c392a17, so it's in the list of commits here too to be reviewed, edited (needs test: as the prefix) and have metadata attached.

[imatlopez]

I've force pushed out the HEAD commit on master, c392a17, so it's in the list of commits here too to be reviewed, edited (needs test: as the prefix) and have metadata attached.

reworded commit names, is this what you meant?

[rvagg]

reworded commit names, is this what you meant?

No, sorry, it's not your commit, we'll deal with it when merged but if you rebase onto master then you'll be without it. I think it comes from work @cclauss was doing for #1980 but got pushed to master prematurely?

[imatlopez]

Do you know why the python 3.5 test is failing? Can it be removed?

[rvagg]

nope, will wait for @cclauss' input on that

[cclauss]

Thanks for doing this! Well done.

[rvagg]

@cclauss got any suggestions on the 3.5 failure here?

[cclauss]

On the 3.5 tests, those seem to be diff tests and my sense is that all the same data is there but not necessarily in the right order. Perhaps those tests need to look at smaller chunks or sort things before looking at them.

I will study that output in the next few hours...

Update: The diff contains:

E - 'AdditionalOptions': '/more /J', E ? --- E + 'AdditionalOptions': '/more', 

So the /J switch has been removed in both tests.

[imatlopez]

What is that flag for? Is it related to using python -m pytest vs pytest? I know one difference is that the first one sets the current working directory as the PYTHONPATH while the second one does not.

[joaocgreis]

The 3.5 failure is not related to this PR, so I think this should land when ready.

I saw the 3.5 failure when working on #1961, but it went away on a second run. That points to something like a difference in Travis runners or a broken cache.

@targos (since I saw you enabled Travis here, but other org owners can also do this) can you clear the Travis cache? Thanks!

[targos]

@joaocgreis sure, what should I remove (I don't know anything about Travis caches):

[joaocgreis]

@targos please delete all repository caches (top right corner). Thanks!

I believe nodejs/node uses the cache to actually make the build possible because of timing issues, but here it should be ok to just delete everything.

[targos]

@joaocgreis it's done

[richardlau]

I restarted https://travis-ci.com/nodejs/node-gyp/jobs/261487549 after the cache was cleared but it still failed.

[imatlopez]

@rvagg anything pending for merging? I saw #1982 and it looks promising but I don't understand the distinction between travis and gh actions.

[rvagg]

d6a7e0e - @cclauss' commit
345c70e - the rest, I had a bit of trouble coming up with a summary but I think I got the gist if ot in there

[rvagg]

still dealing with linux 3.5 failures https://travis-ci.com/nodejs/node-gyp/builds/141142728, fwiw

[cclauss]

See #1995 for more on Python 3.5 on Linux failures...