git apply --indent-to-add deletes other files from the index

Commit Message

Ryan Hodges (rhodges) Oct. 26, 2021, 3:11 p.m. UTC

Hi all,
 
I’ve got a quick question about ‘git apply –intent-to-add’.  If I’ve got a patch that just adds one file to the tree:
 
[sjc-ads-2565:t.git]$ git diff
 
and I apply that patch with –intent-to-add:
 
[sjc-ads-2565:t.git]$ git apply --intent-to-add c.diff
 
The newly added file is tracked but other files in the tree get marked as deleted:
 
[sjc-ads-2565:t.git]$ git status
On branch master
Changes to be committed:
  (use “git restore –staged <file>…” to unstage)
                deleted:    a.c
                deleted:    b.c
 
Changes not staged for commit:
  (use “git add <file>…” to update what will be committed)
  (use “git restore <file>…” to discard changes in working directory)
                new file:   c.c
 
It looks like Git created a new index with only the newly added file in the patch.  However, I’d like Git to just add one entry to the index corresponding to the newly added file in the patch.  Is this a bug or am I completely misinterpreting the goal of ‘intent-to-add’.  I just started looking at the source but a quick message from the experts would be much appreciated. 
 
I’m currently testing with Git version 2.33.
 
Regards,
Ryan

Comments

Johannes Altmanninger Oct. 30, 2021, 8:39 p.m. UTC | #1

On Tue, Oct 26, 2021 at 03:11:36PM +0000, Ryan Hodges (rhodges) wrote:
> Hi all,
>  
> I’ve got a quick question about ‘git apply –intent-to-add’.  If I’ve got a patch that just adds one file to the tree:
>  
> [sjc-ads-2565:t.git]$ git diff
> diff --git a/c.c b/c.c
> new file mode 100644
> index 0000000..9daeafb
> --- /dev/null
> +++ b/c.c
> @@ -0,0 +1 @@
> +test
>  
> and I apply that patch with –intent-to-add:
>  
> [sjc-ads-2565:t.git]$ git apply --intent-to-add c.diff
>  
> The newly added file is tracked but other files in the tree get marked as deleted:
>  
> [sjc-ads-2565:t.git]$ git status
> On branch master
> Changes to be committed:
>   (use “git restore –staged <file>…” to unstage)
>                 deleted:    a.c

Yep, looks like a bug to me.
git apply should never change the status of files that are not mentioned in
the input patch.

>                 deleted:    b.c
>  
> Changes not staged for commit:
>   (use “git add <file>…” to update what will be committed)
>   (use “git restore <file>…” to discard changes in working directory)
>                 new file:   c.c
>  
> It looks like Git created a new index with only the newly added file in the patch.

Seems so.

> However, I’d like Git to just add one entry to the index corresponding
> to the newly added file in the patch.  Is this a bug or am I completely
> misinterpreting the goal of ‘intent-to-add’.

Yeah, I think your "git apply --intent-to-add c.diff" should behave exactly like

	echo test > c.c && git add --intent-to-add c.c

Ryan Hodges Oct. 30, 2021, 9:42 p.m. UTC | #2

Thank you. I was hoping to be the one that fixed this because it was a level of logic that matched my current knowledge level.  I appreciate you jumping in with a fix and also confirming this was unexpected behavior.  I was kind of surprised no one has reported this before.

Cheers,
Ryan






> On Oct 30, 2021, at 1:39 PM, Johannes Altmanninger <[e-mail]> wrote:
> 
> On Tue, Oct 26, 2021 at 03:11:36PM +0000, Ryan Hodges (rhodges) wrote:
>> Hi all,
>> 
>> I’ve got a quick question about ‘git apply –intent-to-add’.  If I’ve got a patch that just adds one file to the tree:
>> 
>> [sjc-ads-2565:t.git]$ git diff
>> diff --git a/c.c b/c.c
>> new file mode 100644
>> index 0000000..9daeafb
>> --- /dev/null
>> +++ b/c.c
>> @@ -0,0 +1 @@
>> +test
>> 
>> and I apply that patch with –intent-to-add:
>> 
>> [sjc-ads-2565:t.git]$ git apply --intent-to-add c.diff
>> 
>> The newly added file is tracked but other files in the tree get marked as deleted:
>> 
>> [sjc-ads-2565:t.git]$ git status
>> On branch master
>> Changes to be committed:
>>  (use “git restore –staged <file>…” to unstage)
>>                deleted:    a.c
> 
> Yep, looks like a bug to me.
> git apply should never change the status of files that are not mentioned in
> the input patch.
> 
>>                deleted:    b.c
>> 
>> Changes not staged for commit:
>>  (use “git add <file>…” to update what will be committed)
>>  (use “git restore <file>…” to discard changes in working directory)
>>                new file:   c.c
>> 
>> It looks like Git created a new index with only the newly added file in the patch.
> 
> Seems so.
> 
>> However, I’d like Git to just add one entry to the index corresponding
>> to the newly added file in the patch.  Is this a bug or am I completely
>> misinterpreting the goal of ‘intent-to-add’.
> 
> Yeah, I think your "git apply --intent-to-add c.diff" should behave exactly like
> 
> 	echo test > c.c && git add --intent-to-add c.c

Johannes Altmanninger Oct. 31, 2021, 6:43 a.m. UTC | #3

On Sat, Oct 30, 2021 at 02:42:42PM -0700, Ryan Hodges wrote:
> Thank you. I was hoping to be the one that fixed this because it was a level of logic that matched my current knowledge level.

Sorry I should have just confirmed the bug since you had already said to
look into it. (I usually try to send things when they are "done" from my
side to minimize roundtrips.)
I'm sure there are more low-hanging fruits but finding them is the hard part,
see also https://lore.kernel.org/git/xmqq7dl5z425.fsf@gitster.g/

> I was kind of surprised no one has reported this before.

I guess no one has used it since it was added in 2018.

Patch

diff --git a/c.c b/c.c
new file mode 100644
index 0000000..9daeafb
--- /dev/null
+++ b/c.c
@@ -0,0 +1 @@ 
+test