bug#21559: 25.0.50; auto-revert-mode breaks git rebase

[Michael Albinus]

Dmitry Gutov <dgutov@yandex.ru> writes:

Hi Dmitry,

>> I'm not so happy seeing git specific code in autorevert.el (yes, I know,
>> it's just an environment variable). All git specifics shall be kept in
>> vc-git.el. Otherwise, we would go into a dependency nightmare, soon.
>
> But this way the patch is much smaller, isn't it? That has
> maintainability benefits too.

A dependency nightmare counts much more, IMHO.

> And using a variable would make more sense if we determine that other
> facilities, not just autorevert, make VC calls that we want to
> consider "background".

autorevert wouldn't care who uses this variable. In a broader sense, it
could let-bind it for the whole auto-revert-handler body.

>> If this variable hurts for standard operation of vc-git, we could set it
>> depending whether (auto)revert is running, or not. There is the variable
>> `revert-buffer-in-progress-p', which is non-nil if reverting is in
>> progress. I could set it also before calling `vc-refresh-state' in
>> `auto-revert-handler'. And in vc-git.el, `process-environment' could be
>> extended or not, depending on the value of `revert-buffer-in-progress-p'.
>
> In which part of vc-git.el? Changing the implementation of two
> commands, like the original patch proposed, makes for a bigger change.
>
> We could do that in vc-git-command, I suppose...

Perhaps. autorevert shouldn't know anything about vc-git, it should just
let-bind the variable, and let other packages decide whether they use
it.

FWIW, I'm also not so enthusiastic, that aut-revert-handler calls
vc-refresh-state directly. This would be better organized by a hook.

Best regards, Michael.