panic recovery

panic recovery

[Levend Sayar]

Hi, all.

Yesterday one of my colleague tried to commit her two weeks of work.
Nearly 10 source code files.
After an unsuccessful am and apply commands, she tried rebase. And git
bothered something like you have to finish your am first.
Then I suggested to do

git am --abort

After that, we saw that all the sources she developed for two weeks
are gone. We really got panic.But I know that git stores everthing as
objects, namely commit,tree,blob and tag objects.

We sorted every file according to modification time under .git/objects
directory and use "git show" for every single object.
After a cumbersome work, we found blob objects for our source codes
that we lost and commit objects. Namely we rescued her work.

Is there a better way to use in this kind of situations ?

For example, to list every single objects under objects directory
according to their types ?
Or search a phrase in every blob obejcts to find the source code we
are looking for ?

_lvnd_
(^_^)

Re: panic recovery

[Thomas Rast]

Levend Sayar wrote:
> git am --abort
> 
> After that, we saw that all the sources she developed for two weeks
> are gone.

You can dig up the lost commits from the HEAD reflog after that, which
records all movements of HEAD.  See

  http://www.kernel.org/pub/software/scm/git/docs/git-reflog.html

The method you described is usually only applicable if you have truly
unreachable objects, which happens e.g. if you lost some content that
was once staged in the index but never committed.

-- 
Thomas Rast
trast@{inf,student}.ethz.ch

Re: panic recovery

[Thomas Rast]

By the way:

Levend Sayar wrote:
> Yesterday one of my colleague tried to commit her two weeks of work.

If you really mean she wanted to make the first commit in two weeks,
that is just wrong.  Usually you should commit at least every few
hours, and changes should be small and self-contained.  Even in the
case of starting a new project or self-contained module, you should
commit a lot so you can go back to an older state if required.  You
can use rebase -i later to clean up the work-in-progress commits.

-- 
Thomas Rast
trast@{inf,student}.ethz.ch

Re: panic recovery

[Junio C Hamano]

Levend Sayar <levendsayar@gmail.com> writes:

> Yesterday one of my colleague tried to commit her two weeks of work.
> Nearly 10 source code files.

Ok.

> After an unsuccessful am and apply commands, she tried rebase. And git
> bothered something like you have to finish your am first.

I do not get this for two reasons.

What does "am" and "apply" have anything to do with committing her own
work in her own working tree to begin with? And even if "am" and/or
"apply" were the right tools for committing hear work, that would mean she
had the changes to be committed neatly made into patch form suitable to be
fed to these commands, so I would imagine that the recovery is just the
matter of "reset --hard" followed by attempts to apply those patches again
and fixing rejects more carefully than the first failed attempt?

A practical suggestion, without knowing what really went wrong, that would
be valuable, especially for git beginners but applicable also to git
experts, is this. First commit your own work proper before doing anything
that may cause conflicts such as merging other's possibly unrelated work
(e.g. merge, am, ...). Then do such mergy operation to make separate
commits. You can do fancy things like combining changes into a single
commit _after_ doing so, and it will be much safer because at the worst
case you will be unable to achieve the fanciness (e.g. combining the
changes) but will have the working results (i.e. a commit with your own
changes, and separate commits recording others changes).