[RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Sidhant Sharma]

Hi,

I made an attempt at writing warning messages for the commands the beginners
will be warned against, and would like to request your comments and feedback
on them. I tried to keep them simple, beginner friendly and educative of the
outcome of those commands.
I'd also like to ask if `git rebase` should be kept in the
list as it may not be bad in most cases, though I prepared a
message for that anyway.


Thanks and regards,
Sidhant Sharma


The current list of graylisted commands is as follows:
$ git rebase
$ git reset --hard
$ git clean -f
$ git gc --prune=now --aggressive
$ git push -f <branch>
$ git push remote [+/:]<branch>
$ git branch -D

Warning messages:

$ ggit rebase

[WARNING] You are about to rebase your commits in <topic-branch> onto the
$BASE_BRANCH, which will essentially replay the work done in $TOPIC_BRANCH
since last merge onto $BASE_BRANCH.
For instance,
Current state:

    o---o---A---B  $BASE_BRANCH
         \
          X---Y  $TOPIC_BRANCH

State after rebasing:

    o---o---A---B---X'---Y'  $BASE_BRANCH
         \
          X---Y  $TOPIC_BRANCH

where X' and Y' are the commits making changes identical to those made by X and
Y respectively.
Rebasing is not usually problematic except in cases when you are rebasing
commits that do not exist in your repository.


$ ggit reset --hard

Resetting to <destination-commit-hash>
[WARNING] You are about to hard reset the current HEAD (master) by <n> commit(s).
This will take you back to commit <destination-commit-hash>, and discard all
changes make thereafter. For instance,
Current state:

    o---o---A---B---C---D---E  $CURRENT_BRANCH

After resetting 3 commits:

    o---o---A---B  $CURRENT_BRANCH

The commits C, D and E and the changes made by them will be lost.
Note that if you make commits on top of B, you would have rewritten the history
and would have trouble restoring it easily.
You can undo this action by resetting the HEAD to the last commit you were on,
which can be seen by running `git reflog`. The first entry (HEAD{1}) points to
the current HEAD location, second entry (HEAD{1}) points to the last position of
your HEAD and so on.
If you want to reset while retaining the changes made since, use --soft instead
of --hard. This will reset without discarding the changes made in the previous
commits.


$ ggit push --force

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are about to purge commits from the <URL of origin> master branch.
For instance,
Current state:

        o---o---o---A---B  master on $ORIGIN_URL
             \
              X---Y---Z  your master

State after forced push:

        o---o---o---X---Y---Z  master on $ORIGIN_URL and your master

Commit A and B will be gone. If other people have worked on top of A or B then
they won't be able to merge their changes easily.


$ ggit push <remote> :<branch> ( ggit push <remote> --delete <branch> )

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are about delete a remote branch, which will result in the loss
of commits made on that branch. This may cause a problem if other people are
working on the branch you are deleting, as they would not be able to push or
merge their changes easily.
You can undo this by pushing the same branch to the remote with upstream flag,
that is, by running:
`$ ggit push -u $REMOTE $BRANCH`
This will work unless you have deleted the branch locally as well. In that case,
you need to restore it first.


$ ggit push <remote> +<branch>
( ggit push <remote> +<basebranch>:<targetbranch> )

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are attempting to push changes from $BASE_BRANCH to $TARGET_BRANCH
while allowing non-fast-forward updates. This can leave unreferenced commits
dangling in the origin repository.
For instance,
Current state:

        o---o---A---B  $REMOTE/$TARGET_BRANCH
             \
              X---Y---Z  your $BASE_BRANCH

State after forced push:

        o---o---A---B    (unnamed branch)
             \
              X---Y---Z  $REMOTE/$TARGET_BRANCH and your $BASE_BRANCH

Commits A and B would no longer belong to a branch with a symbolic name, and so
would be unreachable. Also, people who have based their work on A or B would
have to either merge or rebase after you have pushed.
If you wish to keep both the changesets (A,B and X,Y,Z), you may want to use
`merge` or `rebase`.


$ ggit branch -D
( ggit branch -d -f )

Deleting branch $TARGET_BRANCH
[WARNING] You are about to force delete the $TARGET_BRANCH branch. This will
cause git to delete the branch even if it has not been merged, which may result
in loss of commits. Unless you are confident the branch has been merged, use
`-d` or `--delete` without `--force` which will warn you if the branch has not
been merged.
You can restore a forcefully deleted branch by running:
`$ git branch <branch-name> <sha1>`
where <branch-name> is the name of the branch you wish to restore and <sha1> is
the last commit made to the branch you want to restore. You can find this
information by running `git reflog`.
To undo deletion of this branch, you can run:
`$ ggit branch $TARGET_BRANCH $LAST_COMMIT_SHA`


$ ggit clean -f [<path>]

Cleaning $PATH
[WARNING] You are about to clean $PATH, which will permanently delete all files
and directories present at the given path. Note that it would not be possible to
revert this action, and all files not tracked by git will be list permanently.
If you only wish to see what will be deleted on running `ggit clean`, use the
`-n` or `--dry-run` option.


$ ggit gc --prune=now --aggressive
(or $ ggit  gc --prune=all --aggressive)

Running garbage collection
[WARNING] You are about to prune all objects, regardless of their age or
reachability. The `--aggressive` flag tells git to aggressively optimize the
repository at the cost of taking more time to complete the operation. By setting
`--prune` option to `now` or `all`, you ask git to optimize all objects, and not
just the unreachable ones. Unless the repository is quiescent, you will lose
newly created objects that haven’t been anchored with the refs and end up
corrupting your repository.
It is also not possible to revert the actions taken by `git gc`.


Accepting input:

Are you sure you want to continue? [Y/n]

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Junio C Hamano]

Sidhant Sharma <tigerkid001@gmail.com> writes:

> $ ggit rebase
>
> [WARNING] You are about to rebase your commits in <topic-branch> onto the
> $BASE_BRANCH, which will essentially replay the work done in $TOPIC_BRANCH
> since last merge onto $BASE_BRANCH.
> For instance,
> Current state:
>
>     o---o---A---B  $BASE_BRANCH
>          \
>           X---Y  $TOPIC_BRANCH
>
> State after rebasing:
>
>     o---o---A---B---X'---Y'  $BASE_BRANCH
>          \
>           X---Y  $TOPIC_BRANCH
>
> where X' and Y' are the commits making changes identical to those made by X and
> Y respectively.

The topology may be correct, but the branch labels are both wrong,
no?  The tip of the base branch will stay at B, and the tip of the
topic will point at Y'.

> Rebasing is not usually problematic except in cases when you are rebasing
> commits that do not exist in your repository.

This cannot be correct, as you fundamentally cannot work on (not
limited to rebasing) commits that do not exist in your repository.

> $ ggit reset --hard
>
> Resetting to <destination-commit-hash>
> [WARNING] You are about to hard reset the current HEAD (master) by <n> commit(s).

If I were on B and did "git reset --hard Y", i.e.

     o---o---A---B  $CURRENT_BRANCH
          \
           X---Y  $CURRENT_BRANCH_AFTER_RESETTING

does the phrasing "about to reset by <n> commit(s):" make any sense?

> This will take you back to commit <destination-commit-hash>, and discard all
> changes make thereafter. For instance,
> Current state:
>
>     o---o---A---B---C---D---E  $CURRENT_BRANCH
>
> After resetting 3 commits:
>
>     o---o---A---B  $CURRENT_BRANCH

The above two examples make me wonder if these should be static
text.  "ggit rebase" and "ggit reset" have full information of the
concrete branch names, commit object names and the actual topology
of the history, so it should be able to give a description more
tailored to the user's situation.  Instead of giving a fictional
drawing with "For instance, Current state:", it should be able to
draw the actual before-and-after picture based on where the end-user
actually is.  I see _some_ attempts (e.g. with "<n>", mention of
"(master)" and $BASE_BRANCH, you may have meant that they will be
replaced with actual values), but I suspect that telling some truth
(i.e. use of the real branch names) while showing pictures that do
not match the reality (i.e. if the topology and the description are
done as fixed text) would only confuse the users.

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Jacob Keller]

On Fri, Mar 25, 2016 at 10:38 AM, Junio C Hamano <gitster@pobox.com> wrote:
> The above two examples make me wonder if these should be static
> text.  "ggit rebase" and "ggit reset" have full information of the
> concrete branch names, commit object names and the actual topology
> of the history, so it should be able to give a description more
> tailored to the user's situation.  Instead of giving a fictional
> drawing with "For instance, Current state:", it should be able to
> draw the actual before-and-after picture based on where the end-user
> actually is.  I see _some_ attempts (e.g. with "<n>", mention of
> "(master)" and $BASE_BRANCH, you may have meant that they will be
> replaced with actual values), but I suspect that telling some truth
> (i.e. use of the real branch names) while showing pictures that do
> not match the reality (i.e. if the topology and the description are
> done as fixed text) would only confuse the users.

If possible, I would suggest aiming for generating the actual topology
that the user is seeing, customized so that it gives relevenat
information, rather than static examples. It may be that it is not
possible or the effort is too large for such a project. If the latter
is the case, then using only static text is better than trying to use
some but not all the available information, as Junio points out above.

Regards,
Jake

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Matthieu Moy]

Jacob Keller <jacob.keller@gmail.com> writes:

> If possible, I would suggest aiming for generating the actual topology
> that the user is seeing, customized so that it gives relevenat
> information, rather than static examples.

Using the real topology in a useful way is actually pretty hard. It's
quite easy to throw the output of "git log --graph --oneline ..." to the
user, but as soon as the rebase deals with more than a handfull of
commits, we'd want to simplify the history to show something
understandable to the user (which by definition should be a beginner if
he uses ggit), like replacing long sequences of commits with "..." or
so. That is hard to get right.

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Jacob Keller]

On Sat, Mar 26, 2016 at 8:12 AM, Matthieu Moy
<Matthieu.Moy@grenoble-inp.fr> wrote:
> Jacob Keller <jacob.keller@gmail.com> writes:
>
>> If possible, I would suggest aiming for generating the actual topology
>> that the user is seeing, customized so that it gives relevenat
>> information, rather than static examples.
>
> Using the real topology in a useful way is actually pretty hard. It's
> quite easy to throw the output of "git log --graph --oneline ..." to the
> user, but as soon as the rebase deals with more than a handfull of
> commits, we'd want to simplify the history to show something
> understandable to the user (which by definition should be a beginner if
> he uses ggit), like replacing long sequences of commits with "..." or
> so. That is hard to get right.
>

Yes, in which case we should go Junio's route of not using anything
from the real topology.

Thanks,
Jake

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands

[Sidhant Sharma]

On Friday 25 March 2016 11:08 PM, Junio C Hamano wrote:
> Sidhant Sharma <tigerkid001@gmail.com> writes:
>
>> $ ggit rebase
>>
>> [WARNING] You are about to rebase your commits in <topic-branch> onto the
>> $BASE_BRANCH, which will essentially replay the work done in $TOPIC_BRANCH
>> since last merge onto $BASE_BRANCH.
>> For instance,
>> Current state:
>>
>>     o---o---A---B  $BASE_BRANCH
>>          \
>>           X---Y  $TOPIC_BRANCH
>>
>> State after rebasing:
>>
>>     o---o---A---B---X'---Y'  $BASE_BRANCH
>>          \
>>           X---Y  $TOPIC_BRANCH
>>
>> where X' and Y' are the commits making changes identical to those made by X and
>> Y respectively.
> The topology may be correct, but the branch labels are both wrong,
> no?  The tip of the base branch will stay at B, and the tip of the
> topic will point at Y'.
Thanks for pointing that out, will correct that.
>> Rebasing is not usually problematic except in cases when you are rebasing
>> commits that do not exist in your repository.
> This cannot be correct, as you fundamentally cannot work on (not
> limited to rebasing) commits that do not exist in your repository.
>
Actually I meant to refer to the commits that exist outside the local
repo (eg. on the remote), like it says in the 'The Perils of Rebasing'
section of the Git Branching and Rebasing documentation [1]. I'll rephrase
it to make it clearer.
>> $ ggit reset --hard
>>
>> Resetting to <destination-commit-hash>
>> [WARNING] You are about to hard reset the current HEAD (master) by <n> commit(s).
> If I were on B and did "git reset --hard Y", i.e.
>
>      o---o---A---B  $CURRENT_BRANCH
>           \
>            X---Y  $CURRENT_BRANCH_AFTER_RESETTING
>
> does the phrasing "about to reset by <n> commit(s):" make any sense?
>
>> This will take you back to commit <destination-commit-hash>, and discard all
>> changes make thereafter. For instance,
>> Current state:
>>
>>     o---o---A---B---C---D---E  $CURRENT_BRANCH
>>
>> After resetting 3 commits:
>>
>>     o---o---A---B  $CURRENT_BRANCH
> The above two examples make me wonder if these should be static
> text.  "ggit rebase" and "ggit reset" have full information of the
> concrete branch names, commit object names and the actual topology
> of the history, so it should be able to give a description more
> tailored to the user's situation.  Instead of giving a fictional
> drawing with "For instance, Current state:", it should be able to
> draw the actual before-and-after picture based on where the end-user
> actually is.  I see _some_ attempts (e.g. with "<n>", mention of
> "(master)" and $BASE_BRANCH, you may have meant that they will be
> replaced with actual values), but I suspect that telling some truth
> (i.e. use of the real branch names) while showing pictures that do
> not match the reality (i.e. if the topology and the description are
> done as fixed text) would only confuse the users.


On Sunday 27 March 2016 01:06 PM, Jacob Keller wrote:
> On Sat, Mar 26, 2016 at 8:12 AM, Matthieu Moy
> <Matthieu.Moy@grenoble-inp.fr> wrote:
>> Jacob Keller <jacob.keller@gmail.com> writes:
>>
>>> If possible, I would suggest aiming for generating the actual topology
>>> that the user is seeing, customized so that it gives relevenat
>>> information, rather than static examples.
>> Using the real topology in a useful way is actually pretty hard. It's
>> quite easy to throw the output of "git log --graph --oneline ..." to the
>> user, but as soon as the rebase deals with more than a handfull of
>> commits, we'd want to simplify the history to show something
>> understandable to the user (which by definition should be a beginner if
>> he uses ggit), like replacing long sequences of commits with "..." or
>> so. That is hard to get right.
>>
> Yes, in which case we should go Junio's route of not using anything
> from the real topology.
>

I now understand why using the actual names with a hypothetical
topology would only confuse the user. I'll update the diagrams
with hypothetical names.
Also, other than these commands, what others should I include in
the list? I think we may also have warnings for the following:
* git checkout -- <path>
* git rm --cached
* git stash drop [<stash>]
* git stash clear
* All plumbing commands?
Comments?

Thanks and regards,
Sidhant Sharma

[1]: https://git-scm.com/book/en/v2/Git-Branching-Rebasing#The-Perils-of-Rebasing

[RFC/GSOC] Git Beginner | Warnings for potentially destructive commands (v2)

[Sidhant Sharma]

Updated the warning messages, and added warnings for two more commands
(`checkout --` and `stash clear`). The list is now:
$ git rebase
$ git reset --hard
$ git clean -f
$ git gc --prune=now --aggressive
$ git push -f <branch>
$ git push remote [+/:]<branch>
$ git branch -D
$ git checkout [-p] [<tree-ish>] -- <path>
$ git stash clear

Looking forward to your comments.

Thanks and regards,
Sidhant Sharma

---

$ ggit rebase

[WARNING] You are about to rebase your commits in $TOPIC_BRANCH onto the
$BASE_BRANCH, which will essentially replay the work done in $TOPIC_BRANCH
since last merge onto $BASE_BRANCH.
For instance, assume the following history exists and the current branch is
master:

    o---o---A---B  master
         \
          X---Y  topic

After rebasing, the history would be:

    o---o---A---B  master
                 \
                    X'---Y'  topic

where X' and Y' are the commits making changes identical to those made by X and
Y respectively.
Rebasing is not usually problematic except in cases when you are rebasing
commits that exist outside your repository (such as on a remote or on someone
else's computer).

$ ggit reset --hard

Resetting to <destination-commit-hash>
[WARNING] You are about to hard reset the current HEAD ($CURRENT_BRANCH) to a
previous commit in history, possibly in another branch. This will discard all
changes make thereafter.
For instance, assume the following history exists and the current branch is
master:

    o---o---A---B---C---D---E  master

After resetting 3 commits:

    o---o---A---B  master

The commits C, D and E and the changes made by them will be lost.
Note that if you make commits on top of B, you would have rewritten the history
and would have trouble restoring it easily.
You can undo this action by resetting the HEAD to the last commit you were on,
which can be seen by running `git reflog`. The first entry (HEAD{1}) points to
the current HEAD location, second entry (HEAD{1}) points to the last position of
your HEAD and so on.
If you want to reset while retaining the changes made since, use --soft instead
of --hard. This will reset without discarding the changes made in the previous
commits.

$ ggit push --force

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are about to purge commits from the $REMOTE/$BRANCH branch and
overwrite it's history to match yours.
For instance, assume the following history exists where 'origin' is a configured
remote and the current branch is master:

        o---o---o---A---B  origin/master
             \
              X---Y---Z  your master

After force push:

        o---o---o---X---Y---Z  origin/master and your master

Commit A and B will be gone. If other people have worked on top of A or B then
they won't be able to merge their changes easily.
To revert this, you would have to force push from a computer that has not yet
pulled the changes you pushed and still has commits A and B as they were in
origin/master previously.

$ ggit push <remote> :<branch> ( ggit push <remote> --delete <branch> )

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are about delete a remote branch, which will result in the loss
of commits made on that branch. This may cause a problem if other people are
working on the branch you are deleting, as they would not be able to push or
merge their changes easily.
You can undo this by pushing the same branch to the remote with upstream flag,
that is, by running:
`$ ggit push -u $REMOTE $BRANCH`
This will work unless you have deleted the branch locally as well. In that case,
you need to restore it first.

$ ggit push <remote> +<branch>
( or ggit push <remote> +<basebranch>:<targetbranch> )

Pushing changes to $REMOTE/$BRANCH
[WARNING] You are attempting to push changes from $BASE_BRANCH to $TARGET_BRANCH
while allowing non-fast-forward updates. This can leave unreferenced commits
dangling in the origin repository.
For instance, assume the following history exists where 'origin' is a configured
remote and the current branch is master:

        o---o---A---B  origin/master
             \
              X---Y---Z  your master

State after forced push:

        o---o---A---B    (unnamed branch)
             \
              X---Y---Z  origin/master and your master

Commits A and B would no longer belong to a branch with a symbolic name, and so
would be unreachable. Also, people who have based their work on A or B would
have to either merge or rebase after you have pushed.
If you wish to keep both the changesets (A,B and X,Y,Z), you may want to use
`merge` or `rebase`.

$ ggit branch -D
( or ggit branch -d -f )

Deleting branch $TARGET_BRANCH
[WARNING] You are about to force delete the $TARGET_BRANCH branch. This will
cause git to delete the branch even if it has not been merged, which may result
in loss of commits. Unless you are confident the branch has been merged, use
`-d` or `--delete` without `--force` which will warn you if the branch has not
been merged.
You can restore a forcefully deleted branch by running:
`$ git branch <branch-name> <sha1>`
where <branch-name> is the name of the branch you wish to restore and <sha1> is
the last commit made to the branch you want to restore. You can find this
information by running `git reflog`.
To undo deletion of this branch, you can run:
`$ ggit branch $TARGET_BRANCH $LAST_COMMIT_SHA`

$ ggit clean -f [<path>]

Cleaning <path>
[WARNING] You are about to clean <path>, which will delete all files and
directories at the given path which are not tracked by git. Note that it would
not be possible to revert this action, and all files not in present in the index
will be lost permanently.
If you only wish to see what will be deleted on running `ggit clean`, use the
`-n` or `--dry-run` option.

$ ggit gc --prune=now --aggressive
( ggit  gc --prune=all --aggressive)

Running garbage collection
[WARNING] You are about to prune all objects, regardless of their age or
reachability. The `--aggressive` flag tells git to aggressively optimize the
repository at the cost of taking more time to complete the operation. By setting
`--prune` option to `now` or `all`, you ask git to optimize all objects, and not
just the unreachable ones. Unless the repository is quiescent, you will lose
newly created objects that haven’t been anchored with the refs and end up
corrupting your repository.
It is also not possible to revert the actions taken by `git gc`.

$ ggit checkout [-p] [<tree-ish>] -- <path>

Discarding changes from working tree
[WARNING] You are about to permanently discard the unstaged changes present in
<path>. This will bring <path> to its state as stored in the index, or to the
state at <tree-ish>.
If you wish to see what changes will be discarded and/or choose which changes
are will be discarded, use the `-p` or `--patch` flag.
The actions made by this command cannot be undone.

$ ggit rm [--cached] <file>

Removing file(s): <file>
[WARNING] You are about to remove files/directories from the git index. This
will cause git to stop tracking them, as well as delete them from the local
filesystem. Using the `--cached` flag will only remove them from the index and
not the filesystem.
If you only wish to see which files will be removed, use the `-n` or `--dry-run`
flag.
You can revert the effects of this command by either restoring the files
manually and running `ggit add <file>` or by running `ggit reset --hard`.

$ ggit stash clear

Dropping all stashed states
[WARNING] You are about to remove all the stashed states. Note that those states
will then be subject to pruning, and may be impossible to recover. To view all
stashes, use `ggit stash list`. To selectively drop stashes, use
`ggit stash drop <stash>`.

Accepting input
Are you sure you want to continue? [Y/n]

Re: [RFC/GSOC] Git Beginner | Warnings for potentially destructive commands (v2)

[Remi Galan Alfonso]

Sidhant Sharma <tigerkid001@gmail.com> wrote:
> $ ggit push --force
> 
> Pushing changes to $REMOTE/$BRANCH
> [WARNING] You are about to purge commits from the $REMOTE/$BRANCH branch and
> overwrite it's history to match yours.
> For instance, assume the following history exists where 'origin' is
> a configured remote and the current branch is master:
> 
>         o---o---o---A---B  origin/master
>              \
>               X---Y---Z  your master
> 
> After force push:
> 
>         o---o---o---X---Y---Z  origin/master and your master

It should be:

        o---o---X---Y---Z  origin/master and your master

(The last 'o' commit should have been overwritten by the force push)

Thanks,
Rémi