Re: [RFC PATCH] Introduce "precious" file concept

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Nov 12, 2018, 09:49 UTC
Message-ID: <87o9au39s7.fsf@evledraar.gmail.com>
In-Reply-To: <1205132135.1189562.1542013731020.JavaMail.zimbra@matthieu-moy.fr>

On Mon, Nov 12 2018, Matthieu Moy wrote:

Show 22 quoted lines

> "Per Lundberg" <per.lundberg@hibox.tv> wrote:
>
>> On 11/11/18 5:41 PM, Duy Nguyen wrote:
>> > On Sun, Nov 11, 2018 at 1:33 PM Ævar Arnfjörð Bjarmason
>> > <avarab@gmail.com> wrote:
>>
>> >> That will lose no data, and in the very rare cases where a checkout of
>> >> tracked files would overwrite an ignored pattern, we can just error out
>> >> (as we do with the "Ok to overwrite" branch removed) and tell the user
>> >> to delete the files to proceed.
>> > There's also the other side of the coin. If this refuse to overwrite
>> > triggers too often, it can become an annoyance.
>
> I may have missed some cases, but to me the cases when checkout may try
> to overwrite an ignored file are essentially:
>
> * Someone "git add"ed a file meant to be ignored by mistake (e.g.
>   "git add -f *.o").
>
> * A file that was meant to be kept private (e.g. config.mak.dev) ends
>   up being tracked. This may happen when we find a way to make per-developer
>   settings the same for everyone.

Yes, the cases under discussion here are all cases where a tracked file clobbers a file matching a pattern in in .gitignore.

What I'd add to your list is:

* Some projects (I've seen this in the wild) add e.g. *.mp3 or whatever
  else usually doesn't belong in the repo as a "soft ignore". This is
  something we've never recommended, but have implicitly supported since
  the only caveats are a) you need a one-off "git add -f" and then
  they're tracked b) them being missing from "status" before being
  tracked c) the issue under discussion here.

* Cases where e.g. filename changes to a directory or globs because of
  that make this more complex.

Show 27 quoted lines

> I both cases I'd want at least to be notified that something is going on,
> and in the second I'd probably want to keep my local file around.
>> If we feel thrashable is stretching it too far (which I don't think it
>> is), we could add a "core.ignore_files_are_trashable" setting that
>> brings back the old semantics, for those who have a strong feeling about it.
>
> May I remind an idea I sugested in an old thread: add an intermediate level
> where ignored files to be overwritten are renamed (eg. foo -> foo~ like Emacs'
> backup files):
>
> https://public-inbox.org/git/vpqd3t9656k.fsf@bauges.imag.fr/
>
> One advantage of the "rename" behavior is that it's safer that the current,
> but still not very disturbing for people who like the current behavior. This
> makes it a good candidate for a default behavior.
>
> This could come in complement with this thread's "precious" concept:
>
> * If you know what you're doing and know that such or such file is precious,
>   mark it as such and Git will never overwrite it.
>
> * If you don't know about precious files, just keep the default setting and
>   the worse that can happen is to get your file overwritten with a bakup
>   of the old version kept around.
>
> This would probably play better with a notion of "precious" files than with
> a notion of "trashable" files.

I used to think this foo -> foo~ approach made the most sense (and said as much in https://public-inbox.org/git/871s8qdzph.fsf@evledraar.gmail.com/) but I think it's probably best not to do it and just error out, because:

 * We'd still need to handle the cases where "tests" the file collides
   with "tests" the directory. Then where do we move the colliding file?
   ~/.git/lost+found/* ? We could handle the subdir case with another
   special-case though...

 * I think such silent action will just leave users more confused, and
   in many cases (e.g. a merge) whatever message we print out will be
   missed in a deluge of other messaging, and they'll probably miss it.

   I'd like to avoid a case where a bunch of *~ files get committed
   because the user's workflow is (and some beginner git users do this):

       git pull && git add . && git commit && git push

   As the "pull" would now invoke a merge that would do this rename.

 * If I have the "foo" file open in my editor (a plausible way to run
   into this) I switch to another terminal, do the merge, miss the
   message, then re-save "foo". Now I have both "foo" and "foo~"
   on-disk. Another case where we should just refuse until the user
   resolves the situation to avoid the confusion.