Re: [PATCH 0/2] git-gui: revert untracked files by deleting them

From: Pratyush Yadav <me@yadavpratyush.com>
Date: Nov 3, 2019, 01:12 UTC
Message-ID: <20191103011225.b6emr2ldskktokgd@yadavpratyush.com>
In-Reply-To: <CAPSOpYufF8B0sBUUYqBx5YTxH+02qRom+tz4a0_2iBcNOvvMXA@mail.gmail.com>

On 30/10/19 12:16PM, Jonathan Gilbert wrote:

Show 11 quoted lines

> On Wed, Oct 30, 2019 at 4:09 AM Bert Wesarg
> bert.wesarg-at-googlemail.com |GitHub Public/Example Allow|
> <xlwsizdz58ciy7t@sneakemail.com> wrote:
> > in Git speak, that operation is called 'clean' (see 'git clean') why
> > should we overload the 'revert' operation here?
> 
> It's less about overloading the 'revert' operation as overloading the
> UI action which is currently called "Revert". I think it would be a
> worse experience to have to activate a different option to remove
> unwanted files as to remove unwanted changes. Maybe the UI option
> could be renamed "Revert & Clean" or something?

I disagree. There are valid workflows where you want to remove all changes to tracked files, but leave untracked ones alone. As an example, say you wrote a small script to fix some textual things, like your variable re-name patch. Now you run a diff before you commit those changes just to be sure, and notice that your script was overzealous and made some changes it shouldn't have. So, you clean up all tracked files, and give your script a fresh start. Here, you don't want to delete your script.

And in the other direction, say you want to delete all untracked files 
but have unstaged changes in your tracked files. Combining "Revert" and 
"Clean" does not give you an option to only delete untracked files. So 
you now either have to stash your changes, or run `git clean` from the 
command line.

> As a side note, `git clean untracked-file` won't do anything with a
> default configuration, you have to explicitly `-f` it. Not sure if
> that's relevant, but it does feel like a higher barrier to entry than
> `git revert`.

`git revert` is different from our "Revert", though I admit the naming is quite confusing. `git revert` creates a new commit that "reverses" the changes made in an earlier commit(s). The important point to note here is that `git revert` is used when you publish some commits, and then realise later they had some bugs. Now you can't just drop those commits because that would re-write the history, and it would change all the commit hashes since that commit. So, you use `git revert` to create a new commit that _textually_ reverses those changes. The buggy commit still exists in the tree, but its changes don't.

In contrast, git-gui's "Revert" works on unstaged changes. It does not create a new commit. In fact, our revert does something similar to `git checkout -- <file>` (it uses `git checkout-index` to be precise).

So I don't think you should, or _can_, use `git revert` for what you want to do. And so, I don't see why it is being factored in with this discussion. Am I missing something?

-- 
Regards,
Pratyush Yadav