git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH] Documentation/git-checkout: make doc. of checkout <tree-ish> clearer

From
Philip Oakley <philipoakley@iee.org>
Date
Apr 17, 2017, 16:04 UTC
Message-ID
<5EBADDE444D141918F6873BE8456E026@PhilipOakley>
In-Reply-To
<1492380399.19991.13.camel@gmail.com>
From: "Christoph Michelbach" <michelbach94@gmail.com>
Show 61 quoted lines
> On Sun, 2017-04-16 at 22:25 +0100, Philip Oakley wrote:
>> From: "Christoph Michelbach" <michelbach94@gmail.com>
>> > It's: git checkout [-p|--patch] [<tree-ish>] [--] <pathspec>...
>> The one I quoted is direct from the Synopsis, which does indicate
>> there are
>> potentially more aspects to resolve, such as the influence of using
>> the
>> [-p|--patch] options.
>
> Oh, you are right. I didn't even notice the one in the synopsis doesn't
> match the one further down. The one in the synopsis is wrong because
> after removing the optional parameters, it's the same as the first one
> in the synopsis, yet we expect very different behavior from them.
>
>
>> It maybe that the paragraph / sentence that needs adjusting is;
>>
>> 'git checkout' with <paths> or `--patch` is used to restore modified
>> or
>> deleted paths to their original contents from the index or replace
>> paths
>> with the contents from a named <tree-ish> (most often a commit-ish).
>>
>> and split it at the "or replace paths" option to pick out your
>> specific
>> case.
>
> This one is confusing, too: Paths can lead to folders, yet folders whose
> contents have been modified are not restored to their original contents
> when executing that command. Only files are.
>
> After reading the documentation and having never used the command
> before, one would expect
>
>    #!/bin/bash
>    rm -Rf repo
>    mkdir repo
>    cd repo
>    git init &> /dev/null
>    mkdir folder
>    echo a > folder/a
>    git add -A
>    git commit -m "Commit 1." &> /dev/null
>    echo b > folder/b
>    git add -A
>    git commit -m "Commit 2." &> /dev/null
>    echo c > folder/c
>    git add -A
>    git commit -m "Commit 3." &> /dev/null
>    git checkout `git log --pretty=format:%H | tail -1` folder
>    ls folder
>
> to print `a`. However, it prints `a b c` because all of the files inside
> `folder` which have been modified or deleted since (here: none) are
> reset to their original state after the first commit, but `folder`
> itself isn't. Yet, the only path which was passed to the command in
> question is `folder`.
>
> In my opinion, this command needs improved documentation (and the
> synopsis needs to be fixed).
>
I think this example is of a different kind of misunderstanding.

We are at commit 3, with a b c in the working directory and index, and then we ask to checkout certain specific files from the first commit 1, which contains the file a. That old file is extracted and replaces the file from commit 3.

As the file a is identical there is no change and we still have the original b and c files from commit c.

I'd guess that the misunderstanding is that you maybe thought that the whole directory would be reset to it's old state and the files b and c deleted, rather than just the named files present in that old commit being extracted. If we'd created and added a file d just before the checkout, what should have happened to d, and why?

>

Comming back to the mental model issue with the implicit staging of checked out paths, I suspect this is a because we have both a snapshot and a diff based mental model. Normally the distinction is natural. We have the snapshot from the last commit (or branch checkout) in the index, and then we have the two steps for additional changes we personally made, and then added added, that determine the diff(s).

However in this (git checkout <treeish> -- <paths>) case we don't get that two step option. There is IIUC no 'git copyout <treeish> -- <paths>' which simply copies older files into the current worktree without adding it to the index/staging area.

The confounding of the, both optional, [--patch] and [<paths>] in the same description doesn't make it any easier. Splitting their synopses and descriptions may be the way forward.

I also haven't used the --patch option directly so there may be more issues beneath the surface.

> --
> Christoph
>
>

-- Philip

Previous: Christoph MichelbachNext: Junio C Hamano
Message 8 of 23 in “Documentation/git-checkout: make doc. of checkout <tree-ish> clearer”
  1. Documentation/git-checkout: make doc. of checkout <tree-ish> clearerChristoph Michelbach, Apr 15, 2017
  2. Philip OakleyApr 15, 2017
  3. Christoph MichelbachApr 16, 2017
  4. Philip OakleyApr 16, 2017
  5. Christoph MichelbachApr 16, 2017
  6. Philip OakleyApr 16, 2017
  7. Christoph MichelbachApr 16, 2017
  8. Philip OakleyApr 17, 2017
  9. Junio C HamanoApr 18, 2017
  10. Christoph MichelbachApr 18, 2017
  11. Philip OakleyApr 19, 2017
  12. Philip OakleyApr 17, 2017
  13. Junio C HamanoApr 18, 2017
  14. Christoph MichelbachApr 18, 2017
  15. Junio C HamanoApr 19, 2017
  16. Christoph MichelbachApr 22, 2017
  17. Junio C HamanoApr 24, 2017
  18. Christoph MichelbachApr 24, 2017
  19. Christoph MichelbachApr 24, 2017
  20. Junio C HamanoApr 25, 2017
  21. Christoph MichelbachApr 25, 2017
  22. Philip OakleyApr 19, 2017
  23. Christoph MichelbachApr 18, 2017

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.