From: Julia Evans <julia@jvns.ca>
Date: Thu, 28 Aug 2025 13:26:54 GMT
Subject: Re: [PATCH 5/5] doc: git-checkout: clarify restoring files section
Message-ID: <80fba886-4693-4f31-b64f-438391edc3da@app.fastmail.com>
In-Reply-To: <xmqqiki9hfg2.fsf@gitster.g>

On Tue, Aug 26, 2025, at 6:43 PM, Junio C Hamano wrote:
> "Julia Evans via GitGitGadget" <gitgitgadget@gmail.com> writes:
>
>> diff --git a/Documentation/git-checkout.adoc b/Documentation/git-checkout.adoc
>> index b343d292b30b..9f2b86ac5368 100644
>> --- a/Documentation/git-checkout.adoc
>> +++ b/Documentation/git-checkout.adoc
>> @@ -82,20 +82,21 @@ Omitting _<branch>_ detaches `HEAD` at the tip of the current branch.
>>  `git checkout [-f|--ours|--theirs|-m|--conflict=<style>] [<tree-ish>] [--] <pathspec>...`::
>>  `git checkout [-f|--ours|--theirs|-m|--conflict=<style>] [<tree-ish>] --pathspec-from-file=<file> [--pathspec-file-nul]`::
>>  
>> -	Overwrite the contents of the files that match the pathspec.
>> -	When the _<tree-ish>_ (most often a commit) is not given,
>> -	overwrite working tree with the contents in the index.
>> -	When the _<tree-ish>_ is given, overwrite both the index and
>> -	the working tree with the contents at the _<tree-ish>_.
>> +	Restore another version of the file(s) that match the pathspec.
>
> The same comment about "files and directories from a different
> version" applies here.
>
> Also, I am not so sure about the claim that the verb "restore" is
> better understood than "overwrite" due to the presence of "git
> restore".  If you are working on, say, an older maintenance track,
> and want to borrow what is in a few files from the development
> branch, you may say "git restore --source=develop files..." go grab
> them down to your current working tree.  But at least to me,
> "restore" has a connotation to get back what you used to have, but
> the contents in these files you are grabbing had not existed on the
> older maintenance track you are working on, ever, and "restore" had
> always made me go "Huh?".
>
> I am not suggesting to change the command name "restore"; I am
> suggesting the operation that command does using a verb that is
> different from "restore" (in an ancient thread, we tried to explain
> "checkout" without using "checkout" as the primary verb to describe
> what it does, it is the same thing).
>
>>  +
>> -The index may contain unmerged entries because of a previous failed merge.
>> -By default, if you try to check out such an entry from the index, the
>> -checkout operation will fail and nothing will be checked out.
>> -Using `-f` will ignore these unmerged entries.  The contents from a
>
>> +If you specify a commit or tree to restore from (for example `git
>> +checkout main file.txt`), this will restore the version of the file(s)
>> +from that commit or tree. This overwrites the file in the working
>> +directory and stages the file's contents.
>
> OK.  I suspect the most common is to "restore" from HEAD, and the
> "Huh?"ness of using the verb "restore" goes away.  It clearly is
> getting you back to where you were back when your working tree was
> in sync with HEAD.  Perhaps use "git checkout HEAD file.txt" for the
> example?
>
>> +If you do not specify where to restore from (for example `git checkout
>> +file.txt`), this will replace the file(s) with the version from the index.
>
> That is a very negative way to state it.  It is not "do not specify
> where".  If you do not give tree, you are actively specifying that
> you want things from the index.
>
>> +If you check out a file with an unresolved merge
>> +conflict, the checkout operation will fail and no changes will be made.
>
> This is confusing in a way different from the original.  During a
> conflicted merge, you will see a few stages until they are resolved.
>
>  (1) The working tree file has conflict markers and text from
>      multiple variants.  The index has higher-stage cache entry for
>      such a path.
>
>  (2) The user edits the working tree file to resolve the conflicts.
>      Once all the conflict markers are removed, some people may say
>      "the merge conflict has been resolved".  To Git, the path is
>      still unmerged.
>
>  (3) The user tells the index what the resolution is, with commands
>      like "git add", "git rm", and the like.  The higher-stage cache
>      entries in the index for the path are moved and replaced with a
>      single stage-0 entry.  To Git, the path is now merged.
>
> So, "a file with an unresolved merge conflict" would not mean what
> you wanted to say for those who consider that dealing with the
> working tree files is enough to declare victory and consider the
> conflict has been resolved.
>
>> +Using `-f` will ignore the merge conflict.  The contents from a
>
> This changes the meaning to most people from what the original
> meant.  If you have a file F with merge conflicts in the earlier
> part but the later part merged cleanly, does "git checkout -f F"
> ignore the conflicted part and overwrite the rest of the file
> somehow?
>
> If you wanted to avoid the term "unmerged cache entries", you can
> say "unmerged paths".
>
> Taking all of the above into consideration, perhaps...
>
>     When you are in the middle of a conflicted merge (or
>     cherry-pick, "stash pop", etc.) and haven't told Git what the
>     resolution for these conflicted paths are with "git add" and
>     friends, using "git checkout" to check out such an unmerged path
>     out of the index would fail and the command exits with non-zero
>     exit status.
>
>     When the `-f` option is given, these unmerged paths are left
>     untouched, instead of triggering an error.  For all other
>     (i.e. merged) paths that match the <pathspec>, the working tree
>     files are overwritten by the version recorded in the index.
>
> or something along that line, but with a readability enhancement
> like you have been doing in your series ;-).

Thanks, this section of the man page is not easy and this is very
useful feedback. Will work on improving all of these points.