Modeline triggers E12 security errors

[psagers]

Create a .txt file somewhere with contents like this:

# vim: tw=80 lbr spell

Editing this file in GoNeovim 0.4.7 will cause three errors, one for each modeline setting.

Editing this file in GoNeovim (:e ~/path/to/test.txt) generates the following errors:

Error detected while processing OptionSet Autocommands for "*":
E12: Command not allowed from exrc/vimrc in current dir or tag search
E12: Command not allowed from exrc/vimrc in current dir or tag search
E12: Command not allowed from exrc/vimrc in current dir or tag search

Environment

• macOS 10.15.6
• Neovim 0.4.4 (homebrew)
• GoNeovim 0.4.7 (Github release)
• No init.vim

[cpkio]

It seems to be about # sign and txt ftype, not a goneovim issue.

[akiyosi]

@psagers @cpkio Hi, Thanks for the contribution. This was an issue with Goneovim. Goneovim needs to detect some Option changes and has its own autocmd to detect them internally. On the other hand, rpcnotify running on this autocmd seems to cause E12 in files with ro or modeline enabled.

The problem with modeline causing E12 has been fixed.

[akiyosi]

https://github.com/akiyosi/goneovim/commit/9949899bdf9a2128edadbbf30d0cc93d3e761c43#commitcomment-42190807

As for the above comment, I was wrong. I will look at ways around this problem.

[akiyosi]

@psagers Hi, The error itself could be suppressed by setting the function call with silent!. Please try it. https://github.com/akiyosi/goneovim/actions/runs/278584984

[psagers]

silent! call rpcnotify(0, "Gui", "gonvim_optionset") would presumably suppress the errors. Of course it would apply to more than just modelines and would also suppress any legitimate errors in the future. It might be nice to actually understand what gonvim_optionset is trying to do that Neovim considers to be security-sensitive.

Just a few thoughts. I haven't set up a goneovim build, I just noticed this problem when I was trying it.

[akiyosi]

Sorry, I had the wrong link. I fixed it to the correct link.

Thanks for the comment. Yes, I think we need to understand the harm of suppressing errors. Currently, I don't have any ideas on how to bypass other implementations, so I'd like to make the feature reduction related to this issue a limitation for a while.

[psagers]

Ah, that makes more sense. I can't seem to run the CI builds. It doesn't appear to be a code signing issue, but I'm not sure what else it would be. In any case, the change looks straightforward enough.

Thanks