Fix build on Windows

[raphinesse]

This fixes the package installation and test running on Windows to expose failing tests related to #1643.

More details in the commit messages.

[coveralls]

Coverage decreased (-4.1%) to 93.717% when pulling ec23455 on raphinesse:fix-windows-build into 45f0860 on benmosher:master.

[coveralls]

Coverage remained the same at 97.838% when pulling 8905007 on raphinesse:fix-windows-build into 2beec94 on benmosher:master.

[raphinesse]

Apparently using nyc@15 does not support node@<8. Given that those Node.js versions are EOL, this does not seem to be a big issue to me. However, that fact would have to be addressed with further changes.

[ljharb]

EOL is irrelevant; we'll support what our supported eslint versions work on.

[raphinesse]

Playing around with the node or npm version in the CI config might serve to work around the issue of tests not running on Appveyor but it won't resolve the issue for someone cloning the repo to their local Windows machine.

[ljharb]

That is true. I'll keep playing with alternatives.

[ljharb]

k, looks like adding in-publish, and downgrading npm on node 12, has exposed 50 failing tests, that seem legitimate.

[raphinesse]

Regarding your changes to the prepublish script: Is it intended that the build script is neither run on local npm install without any arguments nor on npm pack? You can achieve that by using the built-in prepublishOnly w/out using any additional dependencies. I would consider that less hacky.

[ljharb]

prepublishOnly doesn't work on all npm versions, nor does prepare; I'll only ever use prepublish on any project.

[raphinesse]

prepublishOnly doesn't work on all npm versions, nor does prepare; I'll only ever use prepublish on any project.

Ah, I see. against the backdrop of trying to support as many npm versions as possible, it makes sense, of course.

What about the other part of my question?

Is it intended that the build script is neither run on local npm install without any arguments nor on npm pack?

[armano2]

there seems to be more issues on windows

https://github.com/armano2/test-eslint-monorpo/actions/runs/36409848

[ljharb]

@raphinesse no; that's what i added not-in-publish; that way the build script only runs with npm publish, and not with npm install.

@armano2 the current appveyor failures in this PR seem to be legitimate issues on windows, yes.

[fisker]

I think #1651 is the reason those tests fails

[ljharb]

@fisker i rebased #1651 on top of this, and its tests are still failing (albeit with 1 failure instead of 50). Almost there :-)

[ljharb]

please prefer the if-based check; this won't work on node 4, for example.

[raphinesse]

Node 4 is commented out in the Appveyor config. I usually only consider live code. But this branch is exploratory anyhow. I do not find the npm version pinning done here to be a satisfactory solution to the nyc problem (see #1648 (comment)).

[ljharb]

yes, but the intention is to uncomment it.

Fair that it's exploratory, but we were down to 1 failing test before you both started force pushing everywhere :-)

[ljharb]

As for local dev, the first step is to get CI working - all local dev by maintainers is done on a Mac.

Once CI is working, we can try to fix local Windows dev, for the rare contributor that uses it.

[raphinesse]

yes, but the intention is to uncomment it.

Understood

Fair that it's exploratory, but we were down to 1 failing test before you both started force pushing everywhere :-)

Yes, but tests not running at all on Node 8 and 10.

[ljharb]

Again, first step is to get CI passing somehow.

We can uncomment node 10, 8, 6, 4, etc on appveyor in a future PR.

[raphinesse]

As for local dev, the first step is to get CI working - all local dev by maintainers is done on a Mac.

Understood

[raphinesse]

@ljharb I'm done force-pushing here. Can confirm that npm version pinning resolves the nyc problem on all Node versions currently running on AppVeyor.

[ljharb]

Thanks for all the legwork here! I'll merge this via #1651.

[ljharb]

It does still have to stay open until it’s merged tho :-)