Re: A project-files implementation for Git projects

[Dmitry Gutov]

On 30.09.2019 20:09, Stefan Monnier wrote:
> > 1. Not having to worry about changing the API to accommodate future needs.
>
> I think we should feel free to change this "API" at will.

So no worry about external VC backends? Like these ones:

  vc-darcs
  vc-fossil
  vc-hgcmd
  vc-msg
  vc-osc

> > 2. Very uneven backend capabilities, and the necessity to set up
> > fallbacks.  As you can see, even now Git requires a modern-enough version to
> > support EXTRA-IGNORES. And, well, project--files-in-directory is private, at
> > least for now.
>
> These seem to affect the code regardless of whether it goes through
> vc-call-backend or not.

Not exactly. vc-call-backend has its own fallback mechanism (using 
project--files-in-directory there would be kinda awkward), but if we do 
all the checks in one place, we don't have to use it.

Further, normally when somebody adds an implementation of a VC action 
for some backend, it's a good thing, even if the implementation is 
imperfect. Not so in this case.

Using a defcustom for "which backends are capable" is also out of line 
with general VC usage.

In any case, we should push the new feature first. The decision whether 
the functionality should move to a VC action can happen later.