On Fri, Apr 05, 2019 at 06:54:06AM +0900, Junio C Hamano wrote:
> Jeff King <peff@xxxxxxxx> writes:
>
> > but you have to take pretty specific steps to get a tag of a blob or
> > tree, like:
> >
> > git tag mytag HEAD:path
> >
> > or
> >
> > git tag mytag HEAD^{tree}
>
> And the way to ask for commit and tag are
>
> git tag mytag master
> git tag mytag v1.0.0
>
> I do not see why only the last one should either be forbidden or
> automaticallly peel. Each of these four names an object of a
> specific type, and singling out a tag as "curious" makes the
> interface uneven.
My point is that it's easy to accidentally tag a tag. It's not easy to
accidentally tag a tree. The counter example would be something that
looks as easy as your "git tag mytag v1.0.0" but actually tags a tree.
I couldn't think of one.
I do buy the argument that arguing for safety valves from a point of
"what's the easiest mistake to make in our current interface" might not
be the best one. It yields a set of operations that aren't necessarily
sensible or orthogonal with respect to the underlying reality of what's
possible and what's reasonable. Which I think is the point you're
making.
-Peff
