Sync fails when upstream force pushes Git tags

[antoineco]

• nvim --version: v0.9.0-dev-1285-g760b399f6
• git --version: 2.37.0
• Operating system/version: macOS 12.6
• Terminal name/version: kitty 0.26.4

• Problem description

Some projects update certain Git tags by force pushing them on a regular basis. Typical example: a nighly tag may change every day.

When this occurs anytime after packer has already been synced at least once, subsequent updates (sync) fail consistently with the message [rejected] ... would clobber existing tag. (see details below)

Steps to reproduce

1. return require'packer'.startup(function(use)
     use 'kyazdani42/nvim-tree.lua'
   end)

2. :PackerSync

3. Force push any existing tag / wait until upstream does.

4. :PackerSync

Actual behaviour

 ✗ Failed to update kyazdani42/nvim-tree.lua
  Errors:
    remote: Enumerating objects: 455, done.
...
    Resolving deltas: 100% (95/95), completed with 29 local objects.
    From https://github.com/kyazdani42/nvim-tree.lua
       875d38e..2b97063  master                   -> origin/*
     ! [rejected]        nightly                  -> *  (would clobber existing tag)
    875d38e

Expected behaviour

 ✓ Updated kyazdani42/nvim-tree.lua

Additional notes

Fetching from Git with the --force flag is the easiest solution that comes to mind: https://stackoverflow.com/q/58031165/4716370

My current workaround is to add this flag to config.git.subcommands.update while calling init():

require'packer'.init{
  git = {
    subcommands = {
      update = 'pull --ff-only --progress --rebase=false --force'
    }
  }
}

packer files

Not relevant or already provided above.

[overhacked]

My current workaround is to add this flag to config.git.subcommands.update while calling init():

require'packer'.init{
  git = {
    subcommands = {
      update = 'pull --ff-only --progress --rebase=false --force'
    }
  }
}

Turns out this isn't quite enough if plugin.commit or plugin.tag is specified, because of this condition: https://github.com/wbthomason/packer.nvim/blob/64ae65fea395d8dc461e3884688f340dd43950ba/lua/packer/plugin_types/git.lua#L221-L223

The workaround for me was:

require'packer'.init{
    git = {
      subcommands = {
        -- Fix for https://github.com/wbthomason/packer.nvim/issues/1100,
        -- which blocks updating the nvim-tree nightly tag
        update = 'pull --ff-only --progress --rebase=false --force',
        fetch = 'fetch --depth 999999 --progress --force',
      }
    },
}

[antoineco]

@overhacked good point, sorry if I didn't understand your comment on the linked PR.

[dave-kennedy]

Setting --force on all plugins might be excessive. Consider instead setting the --no-tags option for just the nvim-tree repository, e.g.:

$ cd ~/.local/share/nvim/site/pack/packer/start/nvim-tree.lua # or wherever it's installed on your machine
$ git config --add remote.origin.fetch '^refs/tags/*'
$ git config remote.origin.tagOpt --no-tags

I'm not sure why the negative refspec (^refs/tags/*) is required. Seems like --no-tags should be sufficient by itself. It isn't.

[overhacked]

I think you're right that --force is overkill and (maybe??) destructive if someone is hacking on a plugin by editing and committing inside of the nvim pack directory. I read through the git fetch man page, and it seems that fetch behavior for tags changed in Git version 2.20:

Until Git version 2.20, and unlike when pushing with git-push(1), any updates to refs/tags/* would be
accepted without + in the refspec (or --force). When fetching, we promiscuously considered all tag
updates from a remote to be forced fetches. Since Git version 2.20, fetching to update refs/tags/*
works the same way as when pushing. I.e. any updates will be rejected without + in the refspec (or
--force).

Unlike when pushing with git-push(1), any updates outside of refs/{tags,heads}/* will be accepted
without + in the refspec (or --force), whether that’s swapping e.g. a tree object for a blob, or a
commit for another commit that’s doesn’t have the previous commit as an ancestor etc.

This looks to me like packer should add a + in front of the refspec if the plugin use spec includes tag or branch. I'll open a PR for discussion.

[antoineco]

@overhacked please note that the issue isn't only occurring when a branch or tag is specified (I have no such thing in my config). It occurs any time upstream force pushes a tag, even one that's not used in the Packer config.

Regarding the suggestion to manually edit the Git config for the problematic repo, I don't think this really solves anything considering the main advantage of Packer's is to have a declarative configuration.

I didn't know about the + form of the refspec. It sounds reasonable if done right.

[overhacked]

I've done some more experimenting with --force and come to understand that it has zero impact on local branch development. The --force option when used with pull or fetch only impacts:

1. The configured remote.<name>.fetch refspec(s) (usually default: +refs/heads/*:refs/remotes/origin/*)
2. Tags fetched with refs/tags/*:refs/tags/*
3. Any explicit refspecs on the command line (not used by Packer)

Because the only local refs affected by (1) and (2) are refs/tags/* and refs/remotes/origin/*, there's no possibility of the local branch heads in refs/heads/* being clobbered. refs/remotes/... is by definition coming from the remote and refs/tags/... would only be an issue if you were running all over a local repo tagging commits with names that don't yet exist on origin but might someday. If you try to use a tag name that already has been fetched, you'd have to git tag -f locally.

tl; dr:fetch --force and pull --force have...

• Zero risk of clobbering local branches
• Benefit of always fetching remote tags and branches, even when maintainers push --force or tag --force their repos
• Tiny risk of clobbering local tags (would anyone do this in a Packer-synced repo?) if-and-only-if you tag locally and then, the remote maintainer uses the same tag name afterwards

I argue that Packer should change config_defaults.git.subcommands.update and ...fetch to include --force.