chore: bug reports to include @vitest/* packages versions

[AriPerkkio]

Include @vitest/* packages in envinfo. Helps narrowing down issues like #2716 where users have different versions of vitest and other @vitest/* packages running.

[AriPerkkio]

The checklist could also include following:

- label: Check that the versions of `vitest` and other `@vitest/*` packages are identical
  required: true

Any thoughts?

[sheremet-va]

The checklist could also include following:

- label: Check that the versions of `vitest` and other `@vitest/*` packages are identical
  required: true

Any thoughts?

Maybe instead we should change peerDependencies from * to workspace:* (gets replaced with the current version, when published)?

[AriPerkkio]

Yes, peerDependencies is the correct way to solve these conflicting versions. Looks like the coverage packages are not included as optional peer dependencies in vitest package's package.json at all.

Should the @vitest/* packages also have vitest as required peer dependency? Currently they list vitest as regular production dependency instead.

[sheremet-va]

Should the @vitest/* packages also have vitest as required peer dependency? Currently they list vitest as regular production dependency instead.

I think only coverage packages depend on Vitest. Others can be used without Vitest at all (except for UI maybe).

[AriPerkkio]

I think the peer dependencies should be set as following, using 0.27.2 version as example:

{
  "name": "vitest",
  "version": "0.27.2",
  "peerDependencies": {
    "@vitest/coverage-c8": "0.27.2",
    "@vitest/coverage-istanbul": "0.27.2"
  },
  "peerDependenciesMeta": {
    "@vitest/coverage-c8": {
      "optional": true
    },
    "@vitest/coverage-istanbul": {
      "optional": true
    }
  }
}

{
  "name": "@vitest/coverage-c8",  "//": "Same for @vitest/coverage-istanbul package",
  "version": "0.27.2",
  "dependencies": {
    "//": "No vitest here. This package should only be used in a project that has vitest as dependency. Do not bring it with this package."
  },
  "peerDependencies": {
    "vitest": "0.27.2"
  },
  "peerDependenciesMeta": {
    "vitest": {
      "optional": false, "//": "Default value so can be omitted"
    }
  }
}

Currently the @vitest/coverage-x packages declare vitest as production dependency. They should not. Currently it's possible to cause conflicts where @vitest/coverage-x packages bring in their own vitest instead of utilizing the peer dependency:

$ npm i -D -E vitest@0.26.0
$ npm i -D -E @vitest/coverage-istanbul@0.27.2
$ npm ls vitest
repro@1.0.0 /Users/x/y/repro
├─┬ @vitest/coverage-istanbul@0.27.2
│ └── vitest@0.27.2
└── vitest@0.26.0

# Check from file system as well:
$ grep version node_modules/vitest/package.json 
  "version": "0.26.0",
$ grep version node_modules/@vitest/coverage-istanbul/package.json
  "version": "0.27.2",
$ grep version node_modules/@vitest/coverage-istanbul/node_modules/vitest/package.json
  "version": "0.27.2",

[sheremet-va]

I think the peer dependencies should be set as following, using 0.27.2 version as example:

I don't think versions should be hardcoded, * is enough. Also vitest is not an optional peer dependency for coverage providers.

[AriPerkkio]

Also vitest is not an optional peer dependency for coverage providers.

Yes exactly, this is what the peerDependenciesMeta.vitest.optional: false above means.

I don't think versions should be hardcoded, * is enough.

Using * is fine if the packages are indeed guaranteed to work with mixed versions. I think I'll just need to be more careful in future and not introduce breaking changes between vitest and @vitest/* packages like below:

• @vitest/coverage-* packages require vitest version which exports coverageConfigDefaults. They'll break if vitest is not updated as well:
  • 1503e15#diff-ae7036446bf57196f2cd8050393760894902442f7568db2938587837d0edb5a2R28
• @vitest/coverage-* packages expect vitest to pass reportContext as argument:
  • 8509628#diff-e1a511166fbfd52ea3880685b345f4d24b827c3c8c9a8639f2eb51478a0f6f36R16
  • This was made backwards compatible afterwards in 4cd1e5d

If the peerDependencies was used to enforce versions these conflicts would come up when updating the packages. But hardcoding the versions to exact version on every release would not be a solution, as those changes of peer dependency requirements would yet be another breaking change. Maybe let's leave them as is for now and just update this issue template.

[sheremet-va]

@AriPerkkio feel free to open PR with the versions then. I don't think we need to mention the same versions in an issue form, since it's not usually a requirement 🤔