[rush] Fix rush pnpm patch commit

[chengcyber]

Summary

As @elliot-nelson pointed out at #3554 (comment), rush-pnpm patch-commit fails when running it from the root of monorepo, this PR makes rush-pnpm patch-commit works from the repo root folder.

Details

The reason rush-pnpm patch-commit fails is because pnpm patch-commit internally calls a logic which tries to read package.json file under procss.cwd(). While the actual top-level package.json in Rush.js monorepo is common/temp/package.json. So when ${cwd}/package.json doesn't exists and -C or --dir CLI parameter is not specified, Rush internally specifies --dir common/temp parameter for rush-pnpm patch-commit

How it was tested

Tested it with this repo: https://github.com/chengcyber/rush-pnpm-patch-demo

[D4N14L]

Unsure if this check is necessary. There are many Rush projects that have a package.json file at the root of the repo, generally to provide npm scripts at the root level. Additionally, Rush always uses the common/temp folder for the install directory for PNPM

[iclanton]

Probably the most expected behavior would be to always pass this when we're doing an operation that expects to be run in the simulated workspace root.

[chengcyber]

The tricky part is the the installation just expects a "package.json" file under the working directory. After that, the lockfile directory are used to correctly install deps for all packages... where lockfile directory is set to "common/temp" before.

[iclanton]

If there is a package.json at the repo root, how does pnpm find the common/temp lockfile without the --dir/-C parameter?

[chengcyber]

In pnpm, there are two "dir" settings, 1. dir 2. lockfileDir. lockfileDir are used to specify where is the pnpm-lock.yaml, which is common/temp/pnpm-lock.yaml.

Related code:

    const workspaceFolder: string = rushConfiguration.commonTempFolder;
    const pnpmEnvironmentMap: EnvironmentMap = new EnvironmentMap(process.env);
    pnpmEnvironmentMap.set('NPM_CONFIG_WORKSPACE_DIR', workspaceFolder);

[D4N14L]

Is -C just the short form for --dir?

[iclanton]

Seems like we should error in this case and tell people not to pass --dir or -C if they're running patch-commit.

[elliot-nelson]

I suppose it doesn't harm people to pass -C if they want to do something very specific, but if it doesn't exist, we should add the flag to point at our temp folder.

The same logic applies to commands like "rush-pnpm why", which doesn't work right unless in temp folder.

[chengcyber]

Is -C just the short form for --dir?

Yes.

I suppose it doesn't harm people to pass -C if they want to do something very specific, but if it doesn't exist, we should add the flag to point at our temp folder.

Yes. The user story would be

1. Run rush-pnpm patch-commit <path> under the repo root works ✅
2. Run rush-pnpm patch-commit <path> under a project folder works ✅
3. Run rush-pnpm patch-commit <path> --dir <specified_dir> works ✅

[chengcyber]

Updates: Feedback at #3554 (comment), when re-running rush-pnpm patch-commit <path> for the same dependency, the patches folder are not synced to "common/pnpm-patches" folder. I updated the code to handle this case as well in this PR.

[iclanton]

If there is a package.json at the repo root, how does pnpm find the common/temp lockfile without the --dir/-C parameter?