Re: [PATCH] optionally disable overwriting of ignored files

On Fri, Aug 20, 2010 at 01:35:32PM -0700, Junio C Hamano wrote:

If (and this may be a big IF) it is reasonable to add paths to .gitignore
that you do not want to lose, then you would want to have three classes of
untracked paths: "precious but ignored", "trashable" (and by definition
ignored), and "unignored" (and by definition is not ignored and is
precious).

I am not sure anybody would bother categorizing their files into
several classes. On the other hand, .gitignore and
.git/info/exclude may already serve such a purpose.

Files that go into a tracked .gitignore are most likely generated
files, and therefore trashable. Files that go into
.git/info/exclude or into an untracked .gitignore (e.g. echo '*' >
precious-simulation-results/.gitignore), are not always generated
and may not be trashable. At least they would not likely get in the
way of checkout or merge.

(As you noted below, except for the untracked .gitignore part, this
is how git behaves already.)

As I already pointed out, we don't support the "precious but ignored"
class.  So an obvious alternate solution to your particular case is not to
add such a path to the gitignore mechanism.

That would defeat the point of ignoring it.

I have a suspicion that the approach this patch takes would not help very
much in the real life.  You just traded the lack of "precious but ignored"
with "no file is trashable, even if it is listed in .gitignore".

Granted, as long as it is not default, it won't negatively affect people
who do not enable it, other than that it may add maintenance burden on the
resulting bloated code, but I find it hard to swallow new code that does
not convincingly solve the real problem.

Yes, I am not a big fan of this solution either. But if we do not
find a better solution, I think having it configurable cannot hurt.
The code required is minimal. As far as I am concerned, we can even
remove the -i part.

By the way, we seem to have a few longstanding bugs that trashes an
unignored and untracked (hence by definition precious) path during branch
switching, and does not give a correct escape hatch.

Doing this:

    $ git checkout maint
    $ echo foo >t/t2018-checkout-branch.sh
    $ git checkout master

does correctly error out because the unignored file needs to be
overwritten.  But doing this after the above still errors out, which is
probably wrong:

    $ echo t/t2018-checkout-branch.sh >>.git/info/exclude
    $ git checkout master

I understand your point. But this is actually a great example. I
have a bunch of such tests, which are not in shape for upstream,
but I want to keep them around anyways (and run them). Do you
really think that an untracked test which was added to
.git/info/exclude should be considered trashable? If it were a
generated file, it would have been added to .gitignore.

After doing the above:

    $ ed .git/info/exclude ;# remove the extra entry in info/excludes
    $d
    w
    q
    $ rm t/t2018-checkout-branch.sh
    $ echo foo >po
    $ git checkout pu

should error out, as "po" is a directory that has tracked contents, and we
never said the untracked regular file "po" is trashable, but the above
sequence happily checks the branch out.

Ok, that's bad.

Clemens