On Thu, Nov 08, 2018 at 04:45:16PM +0100, Johannes Schindelin wrote:
> > One thing I had in mind when proposing $VARIABLE is that it opens up a
> > namespace for us to expand more things (*) for example $GIT_DIR (from
> > ~/.gitconfig).
> >
> > (*) but in a controlled way, it may look like an environment variable,
> > but we only accept a selected few. And it's only expanded at the
> > beginning of a path.
>
> I understand that desire, and I even agree. But the fact that it looks
> like an environment variable, but is not, is actually a point in favor of
> *not* doing this. It would violate the principle of least astonishment.
I agree that it is potentially surprising. OTOH, it is at least pretty
obvious when you see in the wild something like:
[core]
foo = $RUNTIME_PREFIX/bar
what it is _trying_ to do. My big concern with "~"-based things is that
they're kind of subtle. If I saw:
[core]
foo = ~~/bar
I'm not sure what I would think it does. ;)
> The form `<RUNTIME_PREFIX>/abc/def` would not be confused with anything
> that it is not, I would think. The only thing against this form (at least
> that I can think of) is that some people use this way to talk about paths
> that vary between different setups, with the implicit assumption that the
> reader will "interpolate" this *before* applying. So for example, when I
> tell a user to please edit their <GIT_DIR>/config, I implicitly assume
> that they know to not type out, literally, <GIT_DIR> but instead fill in
> the correct path.
So yeah, some alternative syntax that is verbose but distinct makes
sense to me. We use %-substitution elsewhere. Could we do something with
that? "%RUNTIME_PREFIX%" gives me too many flashbacks, but something
like "%(RUNTIME_PREFIX)" matches our formatting language.
I dunno. I actually do not think "$FOO" is so bad, as long as we clearly
document that:
- we do not allow arbitrary $FOO from the environment (though I am
actually not all that opposed to doing so)
- we add some special $FOOs that are not actually environment
variables
-Peff
