Aborting 'rebase main feat' removes unversioned files

Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

Looks like a bug in git rebase main feat.

To reproduce:
git init
git commit -m 'init' --allow-empty
git checkout -b feat
echo 123 > readme.txt
git add readme.txt
git commit -m 'txt=123'
git checkout main
echo 012 > readme.txt
git rebase main feat
git rebase --abort

Expected result:
readme.txt contains 012

Actual result:
readme.txt contains 123

According to the docs, git rebase main feat is a shorthand for git checkout feat followed by git rebase main. I have checked that doing checkout and rebase separately instead of using the shorthand does not have the same issue.

Cheers!

Fedor

Re: Aborting 'rebase main feat' removes unversioned files

[Bagas Sanjaya]

On 04/09/21 03.33, Fedor Biryukov wrote:
> Looks like a bug in git rebase main feat.
> 
> To reproduce:
> git init
> git commit -m 'init' --allow-empty
> git checkout -b feat
> echo 123 > readme.txt
> git add readme.txt
> git commit -m 'txt=123'
> git checkout main
> echo 012 > readme.txt
> git rebase main feat
> git rebase --abort
> 

Did you forget committing?

> Expected result:
> readme.txt contains 012
> 
> Actual result:
> readme.txt contains 123
> 
> According to the docs, git rebase main feat is a shorthand for git checkout feat followed by git rebase main. I have checked that doing checkout and rebase separately instead of using the shorthand does not have the same issue.

I think this is non-issue (behavior as intended).

So when you say `git rebase main feat`, Git will rebase your commits in 
feat on top of main. If any conflicts occur and you abort rebasing (`git 
rebase --abort`), your feat branch just looks like before rebasing.

-- 
An old man doll... just what I always wanted! - Clara

Re: Aborting 'rebase main feat' removes unversioned files

[Jeff King]

On Sat, Sep 04, 2021 at 01:57:11PM +0700, Bagas Sanjaya wrote:

> On 04/09/21 03.33, Fedor Biryukov wrote:
> > Looks like a bug in git rebase main feat.
> > 
> > To reproduce:
> > git init
> > git commit -m 'init' --allow-empty
> > git checkout -b feat
> > echo 123 > readme.txt
> > git add readme.txt
> > git commit -m 'txt=123'
> > git checkout main
> > echo 012 > readme.txt
> > git rebase main feat
> > git rebase --abort
> > 
> 
> Did you forget committing?

I don't think so.

The point is that "readme.txt" is not a tracked file on the main branch,
and thus Git should consider it precious.

I don't think the "rebase --abort" is the problem here, though. It's the
command before:

  git rebase main feat

The "feat" branch is already ahead of "main" (which has no new commits),
and so it just says:

  Current branch feat is up to date.

and leaves us on the "feat" branch. But in doing so, it overwrites the
precious untracked content in the working tree.

The "git rebase --abort" command then does nothing, because there's no
rebase in progress.

-Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

There is no way this could be the intended behavior, but the good news
is that I cannot reproduce it...
Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
And it does not occur in the latest git version: git version 2.33.0.windows.2.

-Fedor

On Sat, Sep 4, 2021 at 11:48 AM Jeff King <peff@peff.net> wrote:
>
> On Sat, Sep 04, 2021 at 01:57:11PM +0700, Bagas Sanjaya wrote:
>
> > On 04/09/21 03.33, Fedor Biryukov wrote:
> > > Looks like a bug in git rebase main feat.
> > >
> > > To reproduce:
> > > git init
> > > git commit -m 'init' --allow-empty
> > > git checkout -b feat
> > > echo 123 > readme.txt
> > > git add readme.txt
> > > git commit -m 'txt=123'
> > > git checkout main
> > > echo 012 > readme.txt
> > > git rebase main feat
> > > git rebase --abort
> > >
> >
> > Did you forget committing?
>
> I don't think so.
>
> The point is that "readme.txt" is not a tracked file on the main branch,
> and thus Git should consider it precious.
>
> I don't think the "rebase --abort" is the problem here, though. It's the
> command before:
>
>   git rebase main feat
>
> The "feat" branch is already ahead of "main" (which has no new commits),
> and so it just says:
>
>   Current branch feat is up to date.
>
> and leaves us on the "feat" branch. But in doing so, it overwrites the
> precious untracked content in the working tree.
>
> The "git rebase --abort" command then does nothing, because there's no
> rebase in progress.
>
> -Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

Just tested it on my MacBook. To my surprise, the file got deleted by
'git rebase main feat':

550$ git --version
git version 2.33.0

551$ git init -b main
Initialized empty Git repository in
/Users/ted/workspace/git-abort-bug/test/.git/

552$ git commit -m 'init' --allow-empty
[main (root-commit) ede7bdf] init

553$ git checkout -b feat
Switched to a new branch 'feat'

554$ echo 123 > readme.txt

555$ git add readme.txt

556$ git commit -m 'txt=123'
[feat 29ac3de] txt=123
 1 file changed, 1 insertion(+)
 create mode 100644 readme.txt

557$ git checkout main
Switched to branch 'main'

558$ echo 012 > readme.txt

559$ git rebase main feat
Current branch feat is up to date.

560$ git rebase --abort
fatal: No rebase in progress?

On Sat, Sep 4, 2021 at 11:51 AM Fedor Biryukov <fedor.birjukov@gmail.com> wrote:
>
> There is no way this could be the intended behavior, but the good news
> is that I cannot reproduce it...
> Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> And it does not occur in the latest git version: git version 2.33.0.windows.2.
>
> -Fedor
>
> On Sat, Sep 4, 2021 at 11:48 AM Jeff King <peff@peff.net> wrote:
> >
> > On Sat, Sep 04, 2021 at 01:57:11PM +0700, Bagas Sanjaya wrote:
> >
> > > On 04/09/21 03.33, Fedor Biryukov wrote:
> > > > Looks like a bug in git rebase main feat.
> > > >
> > > > To reproduce:
> > > > git init
> > > > git commit -m 'init' --allow-empty
> > > > git checkout -b feat
> > > > echo 123 > readme.txt
> > > > git add readme.txt
> > > > git commit -m 'txt=123'
> > > > git checkout main
> > > > echo 012 > readme.txt
> > > > git rebase main feat
> > > > git rebase --abort
> > > >
> > >
> > > Did you forget committing?
> >
> > I don't think so.
> >
> > The point is that "readme.txt" is not a tracked file on the main branch,
> > and thus Git should consider it precious.
> >
> > I don't think the "rebase --abort" is the problem here, though. It's the
> > command before:
> >
> >   git rebase main feat
> >
> > The "feat" branch is already ahead of "main" (which has no new commits),
> > and so it just says:
> >
> >   Current branch feat is up to date.
> >
> > and leaves us on the "feat" branch. But in doing so, it overwrites the
> > precious untracked content in the working tree.
> >
> > The "git rebase --abort" command then does nothing, because there's no
> > rebase in progress.
> >
> > -Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

Here's the output from Windows, where everything works as expected.

PS> git --version
git version 2.33.0.windows.2
PS> rm -re -fo *
PS> git init -b main
Initialized empty Git repository in C:/Users/ted/test/.git/
PS> git commit -m 'init' --allow-empty
[main (root-commit) ea4422d] init
PS> git checkout -b feat
Switched to a new branch 'feat'
PS> echo 123 > readme.txt
PS> git add readme.txt
PS> git commit -m 'txt=123'
[feat 82fe34b] txt=123
 1 file changed, 0 insertions(+), 0 deletions(-)
 create mode 100644 readme.txt
PS> git checkout main
Switched to branch 'main'
PS> echo 012 > readme.txt
PS> git status
On branch main
Untracked files:
  (use "git add <file>..." to include in what will be committed)
        readme.txt

nothing added to commit but untracked files present (use "git add" to track)
PS> git rebase main feat
error: The following untracked working tree files would be overwritten
by checkout:
        readme.txt
Please move or remove them before you switch branches.
Aborting
error: could not detach HEAD
PS> git rebase --abort
fatal: No rebase in progress?

On Sat, Sep 4, 2021 at 11:58 AM Fedor Biryukov <fedor.birjukov@gmail.com> wrote:
>
> Just tested it on my MacBook. To my surprise, the file got deleted by
> 'git rebase main feat':
>
> 550$ git --version
> git version 2.33.0
>
> 551$ git init -b main
> Initialized empty Git repository in
> /Users/ted/workspace/git-abort-bug/test/.git/
>
> 552$ git commit -m 'init' --allow-empty
> [main (root-commit) ede7bdf] init
>
> 553$ git checkout -b feat
> Switched to a new branch 'feat'
>
> 554$ echo 123 > readme.txt
>
> 555$ git add readme.txt
>
> 556$ git commit -m 'txt=123'
> [feat 29ac3de] txt=123
>  1 file changed, 1 insertion(+)
>  create mode 100644 readme.txt
>
> 557$ git checkout main
> Switched to branch 'main'
>
> 558$ echo 012 > readme.txt
>
> 559$ git rebase main feat
> Current branch feat is up to date.
>
> 560$ git rebase --abort
> fatal: No rebase in progress?
>
> On Sat, Sep 4, 2021 at 11:51 AM Fedor Biryukov <fedor.birjukov@gmail.com> wrote:
> >
> > There is no way this could be the intended behavior, but the good news
> > is that I cannot reproduce it...
> > Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> > And it does not occur in the latest git version: git version 2.33.0.windows.2.
> >
> > -Fedor
> >
> > On Sat, Sep 4, 2021 at 11:48 AM Jeff King <peff@peff.net> wrote:
> > >
> > > On Sat, Sep 04, 2021 at 01:57:11PM +0700, Bagas Sanjaya wrote:
> > >
> > > > On 04/09/21 03.33, Fedor Biryukov wrote:
> > > > > Looks like a bug in git rebase main feat.
> > > > >
> > > > > To reproduce:
> > > > > git init
> > > > > git commit -m 'init' --allow-empty
> > > > > git checkout -b feat
> > > > > echo 123 > readme.txt
> > > > > git add readme.txt
> > > > > git commit -m 'txt=123'
> > > > > git checkout main
> > > > > echo 012 > readme.txt
> > > > > git rebase main feat
> > > > > git rebase --abort
> > > > >
> > > >
> > > > Did you forget committing?
> > >
> > > I don't think so.
> > >
> > > The point is that "readme.txt" is not a tracked file on the main branch,
> > > and thus Git should consider it precious.
> > >
> > > I don't think the "rebase --abort" is the problem here, though. It's the
> > > command before:
> > >
> > >   git rebase main feat
> > >
> > > The "feat" branch is already ahead of "main" (which has no new commits),
> > > and so it just says:
> > >
> > >   Current branch feat is up to date.
> > >
> > > and leaves us on the "feat" branch. But in doing so, it overwrites the
> > > precious untracked content in the working tree.
> > >
> > > The "git rebase --abort" command then does nothing, because there's no
> > > rebase in progress.
> > >
> > > -Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Jeff King]

On Sat, Sep 04, 2021 at 12:03:49PM +0200, Fedor Biryukov wrote:

> Here's the output from Windows, where everything works as expected.
> [...]
> PS> git rebase main feat
> error: The following untracked working tree files would be overwritten
> by checkout:
>         readme.txt
> Please move or remove them before you switch branches.
> Aborting
> error: could not detach HEAD

Interesting that it behaves differently than the mac and linux versions.
There are some changes between Git for Windows and regular upstream Git,
but looking over the diff between the v2.33.0 releases, I don't see
anything that might cause this discrepancy.

If the problem used to occur on v2.31 and now doesn't, it might be
interesting to bisect it.

-Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

Actually it's the same on all operating systems. I just forgot to add
a new base commit in main in the reproduction scenario - there was
nothing to rebase...
Here is the correct scenario along with the output.

+ git init -b main repo
Initialized empty Git repository in
/Users/ted/workspace/git-abort-bug/repo/.git/
+ cd repo
+ git commit -m base --allow-empty
[main (root-commit) 3299694] base
+ git checkout -b feat
Switched to a new branch 'feat'
+ echo feat >readme.txt
+ git add readme.txt
+ git commit -m txt=feat
[feat c9dc9b4] txt=feat
 1 file changed, 1 insertion(+)
 create mode 100644 readme.txt
+ git checkout main
Switched to branch 'main'
+ git commit -m new-base --allow-empty
[main c5c292c] new-base
+ echo precious > readme.txt
+ git rebase main feat
error: The following untracked working tree files would be overwritten by merge:
readme.txt
Please move or remove them before you merge.
Aborting
hint: Could not execute the todo command
hint:
hint:     pick c9dc9b4d5aca461f1a1bedd3b99b5c8533d5ef10 txt=feat
hint:
hint: It has been rescheduled; To edit the command before continuing, please
hint: edit the todo list first:
hint:
hint:     git rebase --edit-todo
hint:     git rebase --continue
Could not apply c9dc9b4... txt=feat
+ cat readme.txt
precious
+ git rebase --abort
+ cat readme.txt
feat

-Fedor

On Sat, Sep 4, 2021 at 12:24 PM Jeff King <peff@peff.net> wrote:
>
> On Sat, Sep 04, 2021 at 12:03:49PM +0200, Fedor Biryukov wrote:
>
> > Here's the output from Windows, where everything works as expected.
> > [...]
> > PS> git rebase main feat
> > error: The following untracked working tree files would be overwritten
> > by checkout:
> >         readme.txt
> > Please move or remove them before you switch branches.
> > Aborting
> > error: could not detach HEAD
>
> Interesting that it behaves differently than the mac and linux versions.
> There are some changes between Git for Windows and regular upstream Git,
> but looking over the diff between the v2.33.0 releases, I don't see
> anything that might cause this discrepancy.
>
> If the problem used to occur on v2.31 and now doesn't, it might be
> interesting to bisect it.
>
> -Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Jeff King]

On Sat, Sep 04, 2021 at 11:51:19AM +0200, Fedor Biryukov wrote:

> There is no way this could be the intended behavior, but the good news
> is that I cannot reproduce it...
> Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> And it does not occur in the latest git version: git version 2.33.0.windows.2.

I think it is a bug, and it seems to reproduce easily for me (with both
the current tip of master, and with v2.33.0). Here's the recipe you
showed, with a little debugging at the end:

  set -x
  git init repo
  cd repo
  git commit -m base --allow-empty
  git checkout -b feat
  echo feat >readme.txt
  git add readme.txt
  git commit -m txt=feat
  git checkout main
  echo precious >readme.txt

  cat readme.txt
  git checkout feat
  cat readme.txt
  git rebase main feat
  cat readme.txt

This produces:

  + cat readme.txt
  precious
  + git checkout feat
  error: The following untracked working tree files would be overwritten by checkout:
  	readme.txt
  Please move or remove them before you switch branches.
  Aborting
  + cat readme.txt
  precious
  + git rebase main feat
  Current branch feat is up to date.
  + cat readme.txt
  feat

So git-checkout was not willing to overwrite the untracked content, but
rebase was happy to obliterate it.

It does the right thing in very old versions. Bisecting, it looks like
the problem arrived in 5541bd5b8f (rebase: default to using the builtin
rebase, 2018-08-08). So the bug is in the conversion from the legacy
shell script to C (which makes sense; the shell version was just calling
"git checkout", which we know does the right thing).

-Peff

Re: Aborting 'rebase main feat' removes unversioned files

[Elijah Newren]

On Sat, Sep 4, 2021 at 3:19 AM Jeff King <peff@peff.net> wrote:
>
> On Sat, Sep 04, 2021 at 11:51:19AM +0200, Fedor Biryukov wrote:
>
> > There is no way this could be the intended behavior, but the good news
> > is that I cannot reproduce it...
> > Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> > And it does not occur in the latest git version: git version 2.33.0.windows.2.
>
> I think it is a bug, and it seems to reproduce easily for me (with both
> the current tip of master, and with v2.33.0). Here's the recipe you
> showed, with a little debugging at the end:
>
>   set -x
>   git init repo
>   cd repo
>   git commit -m base --allow-empty
>   git checkout -b feat
>   echo feat >readme.txt
>   git add readme.txt
>   git commit -m txt=feat
>   git checkout main
>   echo precious >readme.txt
>
>   cat readme.txt
>   git checkout feat
>   cat readme.txt
>   git rebase main feat
>   cat readme.txt
>
> This produces:
>
>   + cat readme.txt
>   precious
>   + git checkout feat
>   error: The following untracked working tree files would be overwritten by checkout:
>         readme.txt
>   Please move or remove them before you switch branches.
>   Aborting
>   + cat readme.txt
>   precious
>   + git rebase main feat
>   Current branch feat is up to date.
>   + cat readme.txt
>   feat
>
> So git-checkout was not willing to overwrite the untracked content, but
> rebase was happy to obliterate it.
>
> It does the right thing in very old versions. Bisecting, it looks like
> the problem arrived in 5541bd5b8f (rebase: default to using the builtin
> rebase, 2018-08-08). So the bug is in the conversion from the legacy
> shell script to C (which makes sense; the shell version was just calling
> "git checkout", which we know does the right thing).
>
> -Peff

Turns out this is quite a mess.  It's also related to the "don't
remove empty working directories" discussion we had earlier this
week[1], because we assumed all relevant codepaths correctly avoided
deleting untracked files and directories in the way.  But they don't.
And rebase isn't the only offender, because this is buried in
unpack_trees.  In fact, it traces back to (and before)
    fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
working tree files.", 2006-05-17)
which has additional commentary over at
https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
It appears that before this time, git happily nuked untracked files
and considered them expendable, in basically all cases.  However, this
patch continued considering them as expendable whenever opts->reset
was true.  There wasn't much comment about it at the time for the
reasoning of how opts->reset was handled, though trying to read
between the lines perhaps Junio was trying to limit the backward
compatibility concerns of introducing new errors to fewer code paths?
Anyway, Junio did mention `read-tree --reset` explicitly, but this
opts->reset usage also occurs in am, checkout, reset -- and anything
that calls the reset_head() function including: rebase, stash,
sequencer.c, and add-patch.c.

So, then...should we preserve untracked (and non-ignored) files in all
these cases?  This rebase case seems clear, but others might be less
clear.  For example, should "git reset --hard" nuke untracked files
(what if it's a directory of untracked files getting nuked just to
place a single file in the location of the directory)?  The
documentation isn't explicit, but after reading it I would assume that
untracked files should be preserved.  Since we've had bugs in "git
reset --hard" before, such as requiring two invocations in order to
clear out unmerged entries (see [2] and [3]), that also suggests that
this is just another bug in the same area.  But the bug has been
around so long that people might be expecting it; our testsuite has
several cases that incidentally do.  Granted, it's better to throw an
error and require explicit extra steps than to nuke potentially
important work, but some folks might be unhappy with a change here.
Similarly with "git checkout -f".

And stash.c, which operates in that edge case area has tests with
files nuked from the cache without nuking it from the working tree
(causing the working tree file to be considered untracked), and then
attempts to have multiple tests operate on that kind of case.  Those
cases look a bit buggy to me for other reasons (I'm still digging),
but those bugs are kind of hidden by the untracked file nuking bugs,
so fixing the latter requires fixing the former.  And stash.c is a
mess of shelling out to subcommands.  Ick.

I have some partial patches, but don't know if I'll have anything to
post until I figure out the stash mess...

[1] https://lore.kernel.org/git/YS8eEtwQvF7TaLCb@coredump.intra.peff.net/
[2] 25c200a700 ("t1015: demonstrate directory/file conflict recovery
failures", 2018-07-31)
[3] ad3762042a ("read-cache: fix directory/file conflict handling in
read_index_unmerged()", 2018-07-31)

Re: Aborting 'rebase main feat' removes unversioned files

[Ævar Arnfjörð Bjarmason]

On Sat, Sep 04 2021, Elijah Newren wrote:

> On Sat, Sep 4, 2021 at 3:19 AM Jeff King <peff@peff.net> wrote:
>>
>> On Sat, Sep 04, 2021 at 11:51:19AM +0200, Fedor Biryukov wrote:
>>
>> > There is no way this could be the intended behavior, but the good news
>> > is that I cannot reproduce it...
>> > Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
>> > And it does not occur in the latest git version: git version 2.33.0.windows.2.
>>
>> I think it is a bug, and it seems to reproduce easily for me (with both
>> the current tip of master, and with v2.33.0). Here's the recipe you
>> showed, with a little debugging at the end:
>>
>>   set -x
>>   git init repo
>>   cd repo
>>   git commit -m base --allow-empty
>>   git checkout -b feat
>>   echo feat >readme.txt
>>   git add readme.txt
>>   git commit -m txt=feat
>>   git checkout main
>>   echo precious >readme.txt
>>
>>   cat readme.txt
>>   git checkout feat
>>   cat readme.txt
>>   git rebase main feat
>>   cat readme.txt
>>
>> This produces:
>>
>>   + cat readme.txt
>>   precious
>>   + git checkout feat
>>   error: The following untracked working tree files would be overwritten by checkout:
>>         readme.txt
>>   Please move or remove them before you switch branches.
>>   Aborting
>>   + cat readme.txt
>>   precious
>>   + git rebase main feat
>>   Current branch feat is up to date.
>>   + cat readme.txt
>>   feat
>>
>> So git-checkout was not willing to overwrite the untracked content, but
>> rebase was happy to obliterate it.
>>
>> It does the right thing in very old versions. Bisecting, it looks like
>> the problem arrived in 5541bd5b8f (rebase: default to using the builtin
>> rebase, 2018-08-08). So the bug is in the conversion from the legacy
>> shell script to C (which makes sense; the shell version was just calling
>> "git checkout", which we know does the right thing).
>>
>> -Peff
>
> Turns out this is quite a mess.  It's also related to the "don't
> remove empty working directories" discussion we had earlier this
> week[1], because we assumed all relevant codepaths correctly avoided
> deleting untracked files and directories in the way.  But they don't.
> And rebase isn't the only offender, because this is buried in
> unpack_trees.  In fact, it traces back to (and before)
>     fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
> working tree files.", 2006-05-17)
> which has additional commentary over at
> https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
> It appears that before this time, git happily nuked untracked files
> and considered them expendable, in basically all cases.  However, this
> patch continued considering them as expendable whenever opts->reset
> was true.  There wasn't much comment about it at the time for the
> reasoning of how opts->reset was handled, though trying to read
> between the lines perhaps Junio was trying to limit the backward
> compatibility concerns of introducing new errors to fewer code paths?
> Anyway, Junio did mention `read-tree --reset` explicitly, but this
> opts->reset usage also occurs in am, checkout, reset -- and anything
> that calls the reset_head() function including: rebase, stash,
> sequencer.c, and add-patch.c.
>
> So, then...should we preserve untracked (and non-ignored) files in all
> these cases?  This rebase case seems clear, but others might be less
> clear.  For example, should "git reset --hard" nuke untracked files
> (what if it's a directory of untracked files getting nuked just to
> place a single file in the location of the directory)?  The
> documentation isn't explicit, but after reading it I would assume that
> untracked files should be preserved.  Since we've had bugs in "git
> reset --hard" before, such as requiring two invocations in order to
> clear out unmerged entries (see [2] and [3]), that also suggests that
> this is just another bug in the same area.  But the bug has been
> around so long that people might be expecting it; our testsuite has
> several cases that incidentally do.  Granted, it's better to throw an
> error and require explicit extra steps than to nuke potentially
> important work, but some folks might be unhappy with a change here.
> Similarly with "git checkout -f".
>
> And stash.c, which operates in that edge case area has tests with
> files nuked from the cache without nuking it from the working tree
> (causing the working tree file to be considered untracked), and then
> attempts to have multiple tests operate on that kind of case.  Those
> cases look a bit buggy to me for other reasons (I'm still digging),
> but those bugs are kind of hidden by the untracked file nuking bugs,
> so fixing the latter requires fixing the former.  And stash.c is a
> mess of shelling out to subcommands.  Ick.
>
> I have some partial patches, but don't know if I'll have anything to
> post until I figure out the stash mess...

I'd just like to applaud this effort, and also suggest that the most
useful result of any such findings would be for us to produce some new
test in t/ showing these various cases of nuking/clobbering and other
"non-precious" edge cases in this logic. See[1] and its linked [2] for
references to some of the past discussions around these cases.

1. https://lore.kernel.org/git/87a6q9kacx.fsf@evledraar.gmail.com/
2. https://lore.kernel.org/git/87ftsi68ke.fsf@evledraar.gmail.com/

> [1] https://lore.kernel.org/git/YS8eEtwQvF7TaLCb@coredump.intra.peff.net/
> [2] 25c200a700 ("t1015: demonstrate directory/file conflict recovery
> failures", 2018-07-31)
> [3] ad3762042a ("read-cache: fix directory/file conflict handling in
> read_index_unmerged()", 2018-07-31)

Re: Aborting 'rebase main feat' removes unversioned files

[Fedor Biryukov]

Had a brief look at the code.
builtin/rebase.c is using rebase_head helper function defined in
rebase.c/h to do a checkout.
builtin/checkout.c does not use the rebase_head helper function.
I would expect both to use the same code defined in checkout.c/h, but
there is no such file.

On Sun, Sep 5, 2021 at 9:44 AM Ævar Arnfjörð Bjarmason <avarab@gmail.com> wrote:
>
>
> On Sat, Sep 04 2021, Elijah Newren wrote:
>
> > On Sat, Sep 4, 2021 at 3:19 AM Jeff King <peff@peff.net> wrote:
> >>
> >> On Sat, Sep 04, 2021 at 11:51:19AM +0200, Fedor Biryukov wrote:
> >>
> >> > There is no way this could be the intended behavior, but the good news
> >> > is that I cannot reproduce it...
> >> > Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> >> > And it does not occur in the latest git version: git version 2.33.0.windows.2.
> >>
> >> I think it is a bug, and it seems to reproduce easily for me (with both
> >> the current tip of master, and with v2.33.0). Here's the recipe you
> >> showed, with a little debugging at the end:
> >>
> >>   set -x
> >>   git init repo
> >>   cd repo
> >>   git commit -m base --allow-empty
> >>   git checkout -b feat
> >>   echo feat >readme.txt
> >>   git add readme.txt
> >>   git commit -m txt=feat
> >>   git checkout main
> >>   echo precious >readme.txt
> >>
> >>   cat readme.txt
> >>   git checkout feat
> >>   cat readme.txt
> >>   git rebase main feat
> >>   cat readme.txt
> >>
> >> This produces:
> >>
> >>   + cat readme.txt
> >>   precious
> >>   + git checkout feat
> >>   error: The following untracked working tree files would be overwritten by checkout:
> >>         readme.txt
> >>   Please move or remove them before you switch branches.
> >>   Aborting
> >>   + cat readme.txt
> >>   precious
> >>   + git rebase main feat
> >>   Current branch feat is up to date.
> >>   + cat readme.txt
> >>   feat
> >>
> >> So git-checkout was not willing to overwrite the untracked content, but
> >> rebase was happy to obliterate it.
> >>
> >> It does the right thing in very old versions. Bisecting, it looks like
> >> the problem arrived in 5541bd5b8f (rebase: default to using the builtin
> >> rebase, 2018-08-08). So the bug is in the conversion from the legacy
> >> shell script to C (which makes sense; the shell version was just calling
> >> "git checkout", which we know does the right thing).
> >>
> >> -Peff
> >
> > Turns out this is quite a mess.  It's also related to the "don't
> > remove empty working directories" discussion we had earlier this
> > week[1], because we assumed all relevant codepaths correctly avoided
> > deleting untracked files and directories in the way.  But they don't.
> > And rebase isn't the only offender, because this is buried in
> > unpack_trees.  In fact, it traces back to (and before)
> >     fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
> > working tree files.", 2006-05-17)
> > which has additional commentary over at
> > https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
> > It appears that before this time, git happily nuked untracked files
> > and considered them expendable, in basically all cases.  However, this
> > patch continued considering them as expendable whenever opts->reset
> > was true.  There wasn't much comment about it at the time for the
> > reasoning of how opts->reset was handled, though trying to read
> > between the lines perhaps Junio was trying to limit the backward
> > compatibility concerns of introducing new errors to fewer code paths?
> > Anyway, Junio did mention `read-tree --reset` explicitly, but this
> > opts->reset usage also occurs in am, checkout, reset -- and anything
> > that calls the reset_head() function including: rebase, stash,
> > sequencer.c, and add-patch.c.
> >
> > So, then...should we preserve untracked (and non-ignored) files in all
> > these cases?  This rebase case seems clear, but others might be less
> > clear.  For example, should "git reset --hard" nuke untracked files
> > (what if it's a directory of untracked files getting nuked just to
> > place a single file in the location of the directory)?  The
> > documentation isn't explicit, but after reading it I would assume that
> > untracked files should be preserved.  Since we've had bugs in "git
> > reset --hard" before, such as requiring two invocations in order to
> > clear out unmerged entries (see [2] and [3]), that also suggests that
> > this is just another bug in the same area.  But the bug has been
> > around so long that people might be expecting it; our testsuite has
> > several cases that incidentally do.  Granted, it's better to throw an
> > error and require explicit extra steps than to nuke potentially
> > important work, but some folks might be unhappy with a change here.
> > Similarly with "git checkout -f".
> >
> > And stash.c, which operates in that edge case area has tests with
> > files nuked from the cache without nuking it from the working tree
> > (causing the working tree file to be considered untracked), and then
> > attempts to have multiple tests operate on that kind of case.  Those
> > cases look a bit buggy to me for other reasons (I'm still digging),
> > but those bugs are kind of hidden by the untracked file nuking bugs,
> > so fixing the latter requires fixing the former.  And stash.c is a
> > mess of shelling out to subcommands.  Ick.
> >
> > I have some partial patches, but don't know if I'll have anything to
> > post until I figure out the stash mess...
>
> I'd just like to applaud this effort, and also suggest that the most
> useful result of any such findings would be for us to produce some new
> test in t/ showing these various cases of nuking/clobbering and other
> "non-precious" edge cases in this logic. See[1] and its linked [2] for
> references to some of the past discussions around these cases.
>
> 1. https://lore.kernel.org/git/87a6q9kacx.fsf@evledraar.gmail.com/
> 2. https://lore.kernel.org/git/87ftsi68ke.fsf@evledraar.gmail.com/
>
> > [1] https://lore.kernel.org/git/YS8eEtwQvF7TaLCb@coredump.intra.peff.net/
> > [2] 25c200a700 ("t1015: demonstrate directory/file conflict recovery
> > failures", 2018-07-31)
> > [3] ad3762042a ("read-cache: fix directory/file conflict handling in
> > read_index_unmerged()", 2018-07-31)
>

Re: Aborting 'rebase main feat' removes unversioned files

[Elijah Newren]

On Sun, Sep 5, 2021 at 12:44 AM Ævar Arnfjörð Bjarmason
<avarab@gmail.com> wrote:
>
> On Sat, Sep 04 2021, Elijah Newren wrote:
>
> > On Sat, Sep 4, 2021 at 3:19 AM Jeff King <peff@peff.net> wrote:
> >>
> >> On Sat, Sep 04, 2021 at 11:51:19AM +0200, Fedor Biryukov wrote:
> >>
> >> > There is no way this could be the intended behavior, but the good news
> >> > is that I cannot reproduce it...
> >> > Looks like it occurs only in one git version (2.31.0.windows.1, IIRC).
> >> > And it does not occur in the latest git version: git version 2.33.0.windows.2.
> >>
> >> I think it is a bug, and it seems to reproduce easily for me (with both
> >> the current tip of master, and with v2.33.0). Here's the recipe you
> >> showed, with a little debugging at the end:
> >>
> >>   set -x
> >>   git init repo
> >>   cd repo
> >>   git commit -m base --allow-empty
> >>   git checkout -b feat
> >>   echo feat >readme.txt
> >>   git add readme.txt
> >>   git commit -m txt=feat
> >>   git checkout main
> >>   echo precious >readme.txt
> >>
> >>   cat readme.txt
> >>   git checkout feat
> >>   cat readme.txt
> >>   git rebase main feat
> >>   cat readme.txt
> >>
> >> This produces:
> >>
> >>   + cat readme.txt
> >>   precious
> >>   + git checkout feat
> >>   error: The following untracked working tree files would be overwritten by checkout:
> >>         readme.txt
> >>   Please move or remove them before you switch branches.
> >>   Aborting
> >>   + cat readme.txt
> >>   precious
> >>   + git rebase main feat
> >>   Current branch feat is up to date.
> >>   + cat readme.txt
> >>   feat
> >>
> >> So git-checkout was not willing to overwrite the untracked content, but
> >> rebase was happy to obliterate it.
> >>
> >> It does the right thing in very old versions. Bisecting, it looks like
> >> the problem arrived in 5541bd5b8f (rebase: default to using the builtin
> >> rebase, 2018-08-08). So the bug is in the conversion from the legacy
> >> shell script to C (which makes sense; the shell version was just calling
> >> "git checkout", which we know does the right thing).
> >>
> >> -Peff
> >
> > Turns out this is quite a mess.  It's also related to the "don't
> > remove empty working directories" discussion we had earlier this
> > week[1], because we assumed all relevant codepaths correctly avoided
> > deleting untracked files and directories in the way.  But they don't.
> > And rebase isn't the only offender, because this is buried in
> > unpack_trees.  In fact, it traces back to (and before)
> >     fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
> > working tree files.", 2006-05-17)
> > which has additional commentary over at
> > https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
> > It appears that before this time, git happily nuked untracked files
> > and considered them expendable, in basically all cases.  However, this
> > patch continued considering them as expendable whenever opts->reset
> > was true.  There wasn't much comment about it at the time for the
> > reasoning of how opts->reset was handled, though trying to read
> > between the lines perhaps Junio was trying to limit the backward
> > compatibility concerns of introducing new errors to fewer code paths?
> > Anyway, Junio did mention `read-tree --reset` explicitly, but this
> > opts->reset usage also occurs in am, checkout, reset -- and anything
> > that calls the reset_head() function including: rebase, stash,
> > sequencer.c, and add-patch.c.
> >
> > So, then...should we preserve untracked (and non-ignored) files in all
> > these cases?  This rebase case seems clear, but others might be less
> > clear.  For example, should "git reset --hard" nuke untracked files
> > (what if it's a directory of untracked files getting nuked just to
> > place a single file in the location of the directory)?  The
> > documentation isn't explicit, but after reading it I would assume that
> > untracked files should be preserved.  Since we've had bugs in "git
> > reset --hard" before, such as requiring two invocations in order to
> > clear out unmerged entries (see [2] and [3]), that also suggests that
> > this is just another bug in the same area.  But the bug has been
> > around so long that people might be expecting it; our testsuite has
> > several cases that incidentally do.  Granted, it's better to throw an
> > error and require explicit extra steps than to nuke potentially
> > important work, but some folks might be unhappy with a change here.
> > Similarly with "git checkout -f".
> >
> > And stash.c, which operates in that edge case area has tests with
> > files nuked from the cache without nuking it from the working tree
> > (causing the working tree file to be considered untracked), and then
> > attempts to have multiple tests operate on that kind of case.  Those
> > cases look a bit buggy to me for other reasons (I'm still digging),
> > but those bugs are kind of hidden by the untracked file nuking bugs,
> > so fixing the latter requires fixing the former.  And stash.c is a
> > mess of shelling out to subcommands.  Ick.
> >
> > I have some partial patches, but don't know if I'll have anything to
> > post until I figure out the stash mess...
>
> I'd just like to applaud this effort, and also suggest that the most
> useful result of any such findings would be for us to produce some new
> test in t/ showing these various cases of nuking/clobbering and other
> "non-precious" edge cases in this logic. See[1] and its linked [2] for
> references to some of the past discussions around these cases.

Yeah, I discovered the problems by making some testcases for the
avoid-removing-current-working-directory case, and then when I was
surprised by some behaviors I saw I started adding tests for the
(mistaken?) nuking of untracked files...and then saw this bug report.
While I've also been trying to add fixes for the issues I've found,
making the testcases (trying to figure out which commands had bugs and
how to trigger them) was definitely the hardest part of the effort so
far, and likely the most valuable.

> 1. https://lore.kernel.org/git/87a6q9kacx.fsf@evledraar.gmail.com/
> 2. https://lore.kernel.org/git/87ftsi68ke.fsf@evledraar.gmail.com/

The first item, preserving ignored files, is related but different (I
was specifically discussing non-ignored untracked files above).
However on the topic of precious ignored files...both checkout and
merge have a --[no-]overwriite-ignore flag to select this behavior.
Unfortunately, the merge command only passes that flag along to the
fast-forwarding code path and ignores it for all other merge backends.
But merge-ort could trivially be made to handle it (search for the
line of code that reads "if (1/* FIXME: opts->overwrite_ignore*/)" and
note opts->overwrite_ignore isn't a flag that exists, yet), and then
it'd just need to be threaded through merge/rebase/cherry-pick/revert
toplevel commands down to that line of code.  Also, my preliminary
patches touch on the ignore handling for other codepaths and make it
clear how one could support such a flag for those other cases; I'll
leave some comments in the cover letter when I submit it.  Anyway, if
we were to get those other codepaths all hooked up, perhaps we could
add a core.overwrite_ignored option defaulting to true and giving
people a place to just configure this once.  I would much rather a
global option than attempting to provide another class of file that
users have to configure (ignored but precious), especially as we have
had plenty of bugs and problems just dealing with the two classes of
'untracked' and 'ignored' already.

Re: Aborting 'rebase main feat' removes unversioned files

[Junio C Hamano]

Elijah Newren <newren@gmail.com> writes:

> unpack_trees.  In fact, it traces back to (and before)
>     fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
> working tree files.", 2006-05-17)
> which has additional commentary over at
> https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
> It appears that before this time, git happily nuked untracked files
> and considered them expendable, in basically all cases.  However, this
> patch continued considering them as expendable whenever opts->reset
> was true.

Thanks for digging.  Yes, the 'reset' bit was treated as the license
to kill untracked working tree files and directories that get in the
way in order to carry out the unpack_trees operation.

> So, then...should we preserve untracked (and non-ignored) files in all
> these cases?  This rebase case seems clear, but others might be less
> clear....

In short, the guiding principle ought to be that "checkout --force"
and anything that is given "force" as a stronger override should be
allowed to do whatever minimally necessary to match the end result
in the working tree to what the command wants to show in the absense
of these untracked paths.  And without being forced, untracked and
unignored paths are precious and should cause commands to fail, if
they need to be touched for the commands to complete what they are
asked to do [*].

"reset --hard HEAD" is an oddball.

Naïvely, because it is often used as the way to tell Git to "no
matter what, match the working tree to HEAD", even though it does
not have an explicit "--force" on the command line, it feels that it
also should be allowed to do whatever necessary to the working tree
files.  And historically, that is what we wanted to implement.  If
we suddenly made it "safer", I am sure a lot of existing things will
break.

But unfortunately, "--hard" means a bit more than that in the
context of "reset", in that we want to reset in a way that is
different from "--mixed" (reset the index only without touching the
working tree) and "--soft" (do not touch the index or the working
tree), more specifically, with "--hard", we want to reset both the
index and the working tree to match the given committish (often
"HEAD").  From that point of view, "reset --hard" that tries to
preserve untracked and unignored paths, and "reset --force --hard"
that does whatever necessary to untracked and unignored paths to
match the working tree files, when they reset the index and the
working tree to the named committish, may have made sense.  If we
were designing the feature without any existing users, it is no
brainer to imagine that our design would: (1) call the three 'reset'
modes as "both", "index-only" and "neither", instead of "hard",
"mixed" and "soft", and (2) require "--force" to touch untracked and
unignored paths.

And I think that may be a reasonable longer-term goal, but since we
have existing users and scripts, we cannot go there overnight without
devising a migration path.

Thanks.


[Footnote]

* We sometimes talk about adding a precious category of paths that
  are "ignored", but this discussion is orthogonal to that.  Things
  that are not tracked and not ignored ought to be precious in
  principle.

Re: Aborting 'rebase main feat' removes unversioned files

[Elijah Newren]

On Sun, Sep 5, 2021 at 3:31 PM Junio C Hamano <gitster@pobox.com> wrote:
>
> Elijah Newren <newren@gmail.com> writes:
>
> > unpack_trees.  In fact, it traces back to (and before)
> >     fcc387db9b ("read-tree -m -u: do not overwrite or remove untracked
> > working tree files.", 2006-05-17)
> > which has additional commentary over at
> > https://lore.kernel.org/git/7v8xp1jc9h.fsf_-_@assigned-by-dhcp.cox.net/.
> > It appears that before this time, git happily nuked untracked files
> > and considered them expendable, in basically all cases.  However, this
> > patch continued considering them as expendable whenever opts->reset
> > was true.
>
> Thanks for digging.  Yes, the 'reset' bit was treated as the license
> to kill untracked working tree files and directories that get in the
> way in order to carry out the unpack_trees operation.
>
> > So, then...should we preserve untracked (and non-ignored) files in all
> > these cases?  This rebase case seems clear, but others might be less
> > clear....
>
> In short, the guiding principle ought to be that "checkout --force"
> and anything that is given "force" as a stronger override should be
> allowed to do whatever minimally necessary to match the end result
> in the working tree to what the command wants to show in the absense
> of these untracked paths.  And without being forced, untracked and
> unignored paths are precious and should cause commands to fail, if
> they need to be touched for the commands to complete what they are
> asked to do [*].
>
> "reset --hard HEAD" is an oddball.
>
> Naïvely, because it is often used as the way to tell Git to "no
> matter what, match the working tree to HEAD", even though it does
> not have an explicit "--force" on the command line, it feels that it
> also should be allowed to do whatever necessary to the working tree
> files.  And historically, that is what we wanted to implement.  If
> we suddenly made it "safer", I am sure a lot of existing things will
> break.
>
> But unfortunately, "--hard" means a bit more than that in the
> context of "reset", in that we want to reset in a way that is
> different from "--mixed" (reset the index only without touching the
> working tree) and "--soft" (do not touch the index or the working
> tree), more specifically, with "--hard", we want to reset both the
> index and the working tree to match the given committish (often
> "HEAD").  From that point of view, "reset --hard" that tries to
> preserve untracked and unignored paths, and "reset --force --hard"
> that does whatever necessary to untracked and unignored paths to
> match the working tree files, when they reset the index and the
> working tree to the named committish, may have made sense.  If we
> were designing the feature without any existing users, it is no
> brainer to imagine that our design would: (1) call the three 'reset'
> modes as "both", "index-only" and "neither", instead of "hard",
> "mixed" and "soft", and (2) require "--force" to touch untracked and
> unignored paths.
>
> And I think that may be a reasonable longer-term goal, but since we
> have existing users and scripts, we cannot go there overnight without
> devising a migration path.

Sounds reasonable.  I'm presuming that `read-tree --reset` is also an
oddball, but as plumbing it's better to keep its behavior as-is, i.e.
let it nuke untracked files/directories too.  Let me know if you would
prefer otherwise.

I have a bunch of relatively small patches
(https://github.com/git/git/pull/1085, if you want a preview), which
fix most of the problems I found.  I'm going to split it up into five
patch series, the first three of which are independent and shorter
than the remaining two.