Issue with pnpmfileChecksum and frozen-lockfile about pnpm HOT 19 OPEN

(or add a new flag, I don't know which is better)

I think adding a flag makes more sense here since we're having these issues in cases where the pnpm options are not under our control.

Regardless, I believe I found a way to get our .pnpmfile.cjs file into the Render build. We have a monorepo and Render only uses a subset of our monorepo, so we're thinking about fixing this by moving our .pnpmfile.cjs to our package (which is called packages/app) and then doing a symlink from the root of our repo to the file in our package:

❯ ln -s packages/app/.pnpmfile.cjs .pnpmfile.cjs

That will allow us to both use pnpm i locally and pnpm i in Render without having to customize the pnpm options. So hopefully it will work. We're testing it now.

The next issue will come when we have more than one frontend package deployed on Render, as we'll then need to duplicate .pnpmfile.cjs in every package.

from pnpm.

overrides cannot remove dependencies.

I think in this scenario it would be OK to ignore the missing pnpmfile.

from pnpm.

I think that should be fine

from pnpm.

@KSXGitHub could the checksum of the .pnpmfile.cjs differ in different envs?

from pnpm.

@divmgl do you have the same .pnpmfile.cjs file both locally (when you create the lockfile) and in the docker, when you run pnpm install --frozen-lockfile?

from pnpm.

@zkochan checking now

from pnpm.

I guess if git changes the line endings, the checksum could differ

from pnpm.

!!!!! @zkochan we did not! that fixed it! thanks a million

from pnpm.

could the checksum of the .pnpmfile.cjs differ in different envs?

I don't think it could. And CRLFs are normalized into LFs by the hash function so it should be a problem either. But I don't remember the hash function trimming new lines, so if pnpmfile final newlines were trimmed, it could cause hash difference.

from pnpm.

we did not! that fixed it! thanks a million

Oh, it was because the pnpmfiles were different?

from pnpm.

Quick observation though: while this does fix things locally, this will break Render.com builds because the .pnpmfile.cjs file is not included.

@KSXGitHub yep, we had a custom .pnpmfile.cjs file that was not included in our Docker images.

from pnpm.

Maybe we could ignore the pnpmfileChecksum check if --frozen-lockfile is used and .pnpmfile.cjs is not present.

(or add a new flag, I don't know which is better)

from pnpm.

(or add a new flag, I don't know which is better)

I think adding a flag makes more sense here since we're having these issues in cases where the pnpm options are not under our control.

If the options are not under your control then how do you expect to use the said flag?

from pnpm.

If the options are not under your control then how do you expect to use the said flag?

Ah, that's true :(. We're having a different issue now we're the .pnpmfile.cjs is not being picked up by GHA, so we're looking into that now lol

from pnpm.

We're thinking about just removing our .pnpmfile.cjs file at this point. It exists to avoid installing Webpack (as we have dependencies that are using Webpack for some reason) but it's causing more headaches for us.

from pnpm.

It exists to avoid installing Webpack (as we have dependencies that are using Webpack for some reason) but it's causing more headaches for us.

Can you use overrides? Many of pnpmfile use cases can be replaced with pnpm.overrides.

from pnpm.

If webpack was an optional dependencies then there's ignoredOptionalDependencies.

from pnpm.

We could add a way to remove other types of dependencies using settings. It was already requested once.

from pnpm.

We could add a way to remove other types of dependencies using settings. It was already requested once.

Can and should it be integrated to overrides? For example: "webpack": "-" means removing webpack.

from pnpm.