Glob fails to link if there's only one file

[jwpeddle]

When I try to use glob with only one file in the source directory, dotbot fails to create the link with "Nonexistent source for.."

- defaults:
    link:
      relink: true
      glob: true
      force: true
- link:
    ~/.config/broot: broot/*

.
├── broot
│   └── conf.hjson

❯ ./install 
Nonexistent source for ~/.config/broot : /home/jay/dotfiles/broot/*

==> Some tasks were not executed successfully

Adding a second file without changing anything else makes it function as expected. I would expect it to behave the same regardless of how many files are in the dir.

Why bother using glob if there's only one file? Because it means if I add a second file later, I don't need to also edit my install.conf.yaml. Almost always, I just want to symlink whatever is in the source directory. It took some squinting to even identify what was going wrong.

[eengstrom]

I can confirm this with a new test case I created, and that it worked as of the last tagged version (v.1.19.0), and failed after a patch (that I wrote) was merged into master. I will develop a fix.

[anishathalye]

Thanks for looking into it!

@jwpeddle for the moment, you could use Dotbot v1.19.0 (cd dotbot; git checkout v1.19.0; cd ..; git add dotbot; git commit, to make sure the ./install doesn't revert back to whichever version you had pinned).

[eengstrom]

I have a fix of a sort - line 73 of link.py goes from:

                elif len(glob_results) == 1 and destination[-1] != '/':

to

               elif len(glob_results) == 1 and (glob_results[0] == path):

... but I'm not sure the reason one would want to specify glob: true and yet not use any glob meta-chars. Essentially, the entire elif stanza seems like it hides errors in usage and tries to guess the/a right thing to do. The immediately preceding stanza does the same thing, but throws an error if the result is a directory.

Unless there is a clear use-case for allowing the use of glob: true and source path specifications without any glob meta-chars, I'd suggest always throwing an error.

I suppose the possible use case is when people want to put glob: true into a defaults block, but then again, the code is still hiding errors in use that aught to be corrected.

Looking for feedback.

[jwpeddle]

@jwpeddle for the moment, you could use Dotbot v1.19.0 (cd dotbot; git checkout v1.19.0; cd ..; git add dotbot; git commit, to make sure the ./install doesn't revert back to whichever version you had pinned).

Done, and confirmed working, thanks! 🎈

[niraami]

+1 on this one

but I'm not sure the reason one would want to specify glob: true and yet not use any glob meta-chars

Well, I guess the that's the fault of the question. IMO it's better to ask if specifying glob: true without using it could in any way lead to unexpected behaviour - I'm not familiar in the way dotbot handles globs, but I'd like to think that there isn't.
But then there is also the question, if using /* on a folder with a single file is really considered not using any glob meta-chars?

There are a few pro's I can think of right away:

1. If a folder once included multiple files, but you later deleted all but one, the install script is still going to function as expected
2. Consistency with globs elsewhere (bash, zsh, etc). As long as there is at least 1 file, you can use /* to rm, cp, mv, etc.

I don't really have any other examples, as I rarely utilize globs.

The way I found this issue is that I recently removed one of my Polybar scripts, and after re-installing my dotfiles on another machine, the whole scripts/ folder was suddenly gone. This was because in the end, I was left with only 1 file in scripts/ & this yaml syntax:

~/.config/polybar/scripts:
      glob: true
      path: config/polybar/scripts/*

Thus I got here. Well, after finally noticing what dotbot was even talking about with the "Some tasks were not executed successfully" - which leads onto my next point.

It took some squinting to even identify what was going wrong.

This is also very valid, there is no color/formatting difference between information & warnings/errors. I don't have a screenshot on hand, but I can get one if needed.

[LeuschkeTressa]

Came here after scratching my head for quiet a while.

This bug was reported more than two years ago. Is dotbot still being maintained?

If there is anyone with commit access still in the project, may I suggest to immediately add the 2-3 lines of code that would be needed to identify when the unexpected behaviour is triggered, and print a warning message - however short - to the console? That would be fair to the users in my opinion, respecting their time and possibly saving someone some frustration.

(Duplicate: #315)

./install --version
Dotbot version 1.19.1 (git b'53b3781fbb')

[anishathalye]

Thanks, should be fixed in 416f32f (released in 1.19.2).