`~/.gitignore` is ignored

[ijlyttle]

I realize that I can't provide a reproducible example; I need to tell a story (sorry).

I recently upgraded my Mac to Big Sur. Of course, one of the first things I did was install.packages("usethis"), then usethis::git_vaccinate().

I kept on seeing .DS_Store wanting to be committed, which was odd, so I kept adding it to local .gitignore. Finally, this morning, I said "enough" and started digging into it.

Of course, my ~/.gitignore file looked OK, so a little investigation brought me to this SO answer, and sure enough, I did not have git's core.excludesfile set.

So I set core.excludesfile, and .DS_Store "went away".

I see that in edit_git_ignore(), if there is no global gitignore file: you set core.excludesfile, then call git_vaccinate().

However, if you just call git_vaccinate(), I don't see that core.excludesfile gets set. I think this is how I got here.

As well git_vaccinated() does not check that core.excludesfile is set.

I have a few ideas on some solutions, and will be happy to contribute, but I figure I'd let you react first.

Thanks!

[malcolmbarrett]

Ah, by a lucky coincidence, I just got a new computer and ran git_vaccinated() a couple of days ago, and it resulted in the same problem for me. Consider it reproduced!

[ijlyttle]

I'm glad (?) it is reproducible!

My thought is that git_ignore_path() is used by edit_git_ignore(), git_vaccinate() and git_vaccinated().

Perhaps git_ignore_path() could contain the logic for core.excludesfile, moving some of the logic out of edit_git_ignore():

If scope == "user", it could:

• consult core.excludesfile

• if core.excludesfile not set:

  • check if files "~/.gitignore" or "~/.gitignore_global" exist (in that order):
    • if either files exists, use that value.
    • if neither file exists, use (and create) "~/.gitignore", tell user via ui_done().
  • set core.excludesfile, tell user via ui_done()

• return value of core.excludesfile

Happy to start a PR along these lines!

(edit to add that git_vaccinated() should be able to call git_ignore_path() in read-only mode.)

[jennybc]

This all seems reasonable to me 👍