Monorepo with git-hosted package.

[kjs3]

Do you want to request a feature or report a bug?

Feature (I think). It's possible this is considered a bug.

What is the current behavior?

Private (all?) Github modules that exist as a package in a monorepo are installed locally to the package that depends on them.

monorepo/
  node_modules/
    package-1/ -> symlink
  packages/
    package-1/
      package.json
        {
          version: "1.0.1"
        }
    package-2/
      node_modules/
        package-1/ => THIS IS THE PROBLEM
      package.json
        {
          dependencies: {
            package-1: "git+https://token:x-oauth-basic@github.com/user/package-1.git#v1.0.1"
          }
        }

package-1 is hoisted and installed at the root but is also installed in package-2 where it is listed as a dependency.

What is the expected behavior?

monorepo/
  node_modules/
    package-1/ -> symlink
  packages/
    package-1/ => Published to stand-alone git repo in some way. Not Yarn's problem.
      package.json
        {
          version: "1.0.1"
        }
    package-2/
      node_modules/
        package-1-dir-is-not-here
      package.json
        {
          dependencies: {
            package-1: "git+https://token:x-oauth-basic@github.com/user/package-1.git#v1.0.1"
          }
        }

As long as package-1/package.json.version matches the commit-ish in package-2's dependency url then it should not be installed in package-2's node_modules/

Lerna gained this behavior here but I'd rather Yarn handled my deps.

[rally25rs]

This is actually sort of a bug where your git dependency is actually being retrieved from github, not using the workspace package, which is why it can't be hoisted to the top level. This could be fixed by #6012 if it gets merged, but I'm not checking for a commit-ish hash at the end of the URL, I just took the simplistic approach of matching by name only if the "version" is not semver (is a git url instead).

[rally25rs]

tagged as [needs-discussion] because in #6027 @arcanis had some concerns about supporting this behavior.

[kjs3]

@rally25rs I think not checking the commit-ish would be fine. Especially as a first step.