Re: [PATCH v3 0/5] stash: support pathspec argument

[Jeff King <peff@peff.net>]

On Mon, Feb 13, 2017 at 10:35:31PM +0100, Matthieu Moy wrote:

> > Is it really that dangerous, though? The likely outcome is Git saying
> > "nope, you don't have any changes to the file named drop". Of course the
> > user may have meant something different, but I feel like "-p" is a good
> > indicator that they are interested in making an actual stash.
> 
> Indeed -p is not the best example. In the old thread, I used -q which is
> much more problematic:
> 
>   git stash -q drop => interpreted as: git stash push -q drop
>   git stash drop -q => drop with option -q

Yeah, I'd agree with that. I wouldn't propose to loosen it entirely, but
rather to treat "-p" specially.

> It's not really "dangerous" at least in this case, since we misinterpret
> a destructive command for a less destructive one, but it is rather
> confusing that changing the order between command and options change the
> behavior.
> 
> I actually find it a reasonable expectation to allow swapping commands
> and options, some programs other than git allow it.

I think we may have already crossed that bridge with "git -p stash".

Not to mention that the ordering already _is_ relevant (we disallow one
order but not the other). If we really wanted to allow swapping, it
would mean making:

  git stash -p drop

the same as:

  git stash drop -p

I actually find _that_ more confusing. It would perhaps make more sense
with something like "-q", which is more of a "global" option than a
command-specific one. But I think we'd want to whitelist such global
options (and "-p" would not be on that list).

> > The complexity is that right now, the first-level decision of "which
> > stash sub-command am I running?" doesn't know about any options. So "git
> > stash -m foo" would be rejected in the name of typo prevention, unless
> > that outer decision learns about "-m" as an option.
> 
> Ah, OK. But that's not really hard to implement: when going through the
> option list looking for non-option, shift one more time when finding -m.

No, it's not hard conceptually. It just means implementing the
option-parsing policy in two places. That's not too bad now, but if we
started using rev-parse's options helper, then I think you have corner
cases like "git stash -km foo".

My "-p" suggestion suffers from a similar problem if you treat it as
"you can omit the 'push' if you say "-p", rather than "if -p is the
first option, it is a synonym for 'push -p'".

-Peff