Explicitly ignore ERR_PNPM_ERR_PNPM_TARBALL_INTEGRITY
#15927
Replies: 9 comments
|
I wonder if this works for you:
|
|
I've tried the following steps:
and then if I try and add them back, it works properly. This isn't a productive workaround though since it takes some amount of time to manually delete all trace of these packages only to just re-add them again |
Is this step necessary? Can you also omit this? Is |
|
Why do you need to publish to verdaccio? You can install from a local tarball. I used something like that in the past and created this tool: https://github.com/zkochan/package-preview |
No, it's not enough. This will still result in a
I'm using Verdaccio precisely to avoid having to use hacky tools like package-preview which I think fundamentally will never work as well as something like Verdaccio |
|
How did you solve it? I also encountered this problem. |
I don't care if you use I am not sure disabling this check in pnpm is a good idea even as an opt-in. It will be a security vulnerability. Providing a list of packages for which to ignore the check could be safer but I am not sure this is a frequent enough use case. You could just publish your packages with hashes, when you do local testing. Like
What is the issue with bumping versions? Do you commit the lockfile created by e2e tests? Looks like you need to improve the automation. Maybe use a release tool that will bump the versions automatically. |
|
I also encountered the same problem, feeling terrible. I also used Verdaccio. |
|
Ran into this as well. I had a bug fix in a fork that I needed immediately, so I vendored the package as a local tarball via The lockfile ends up storing the checksum from the npm registry rather than recomputing it from the local tarball, causing I really don't want to add |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Contribution
Describe the user story
Using a tool like
Verdaccioto manage packages locally helps a lot for local testing. My flow often looks like this:and I iterate on this loop until things compile and the code feels good.
Note: I do NOT want to have to bump the version of A every single time I iterate. It gets messy, causes conflicts if I push my branch, and in general it's a lot of busy work
However, if I keep the exact same version number, I get
ERR_PNPM_TARBALL_INTEGRITYwhich causes pnpm install to take a minute to install everything as it has to exhaust the retry countDescribe the solution you'd like
I want an option to just force the installation of the new package even if the sha512 differs without having to wait for all the retries
I tried
--fix-lockand--force, and neither of them seem to give me what I wantDescribe the drawbacks of your solution
This option could hide a case where a package actually differs in its hash due to a bug, although unlikely
Describe alternatives you've considered
https://github.com/orgs/verdaccio/discussions/4866
All reactions