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

From: Jonathan Gilbert <logic@deltaq.org>
Date: Nov 3, 2019, 04:41 UTC
Message-ID: <CAPSOpYsBRf4tCXP-wod2h+g_=NZ-Qr0KOSZw3M=96d5FUFpFzw@mail.gmail.com>
In-Reply-To: <20191103011225.b6emr2ldskktokgd@yadavpratyush.com>

On Sat, Nov 2, 2019, 8:12 PM Pratyush Yadav, <me@yadavpratyush.com> wrote:

Show 21 quoted lines

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

But, since this is in this GUI interface, you can clearly see which are which and select only the files you want to affect. If you have so many files that you have to select indiscriminately, then the command-line is probably a better choice anyway. In any case, my proposed change prompts for each part of the change, so you _can_ just select everything, press ^J, and then say "Yes" to only one of the prompts.

Show 7 quoted lines

> > 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.

[..]

> 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?

You are entirely correct, this was just a massive brain fart. Every time I wrote `git revert` in my head I was actually thinking of exactly what Git Gui does, reverting working copy changes by checking out the file. I should have written "reverting using `git checkout`". My apologies!

In my defence, I have over the past few days found myself digging into code hosted in SVN repositories, and `svn revert` does exactly what `git checkout` does to an unstaged modified file. :-)

Jonathan Gilbert

On Sat, Nov 2, 2019 at 8:12 PM Pratyush Yadav <me@yadavpratyush.com> wrote:

Show 55 quoted lines

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