git worktree add /foo <TAB>

git worktree add /foo <TAB>

[Daniel Shahaf]

The second argument to 'git worktree add' can be any committish.  That
committish becomes HEAD of the new worktree, and if it's a branch name,
then that branch becomes the current branch of the new worktree, else
the new worktree starts as a detached HEAD.

How should that be completed?

Currently it's completed as follows:

+++ b/Completion/Unix/Command/_git
@@ -2026,11 +2026,7 @@ _git-worktree() {
       curcontext=${curcontext%:*}-$line[1]:
       case $line[1] in
         (add)
 	  if (( $words[(I)--detach] )); then
 	    args=( ':commit:__git_commits' )
 	  else
 	    args=( ':branch:__git_branch_names' )
 	  fi

… and while simply using __git_commits unconditionally wouldn't be
wrong, I think this case might prefer a variant of __git_commits that
offers branch names in preference to the N other things __git_commits
offers.  Do we have such a helper?

(Or should I just add another tag-order style to my config...?)

I see __git_recent_branches but it's not precisely what I have in mind.

Cheers,

Daniel

Re: git worktree add /foo <TAB>

[Oliver Kiddle]

On 9 Aug, Daniel Shahaf wrote:
> The second argument to 'git worktree add' can be any committish.  That
> committish becomes HEAD of the new worktree, and if it's a branch name,
> then that branch becomes the current branch of the new worktree, else
> the new worktree starts as a detached HEAD.
>
> How should that be completed?
>
> Currently it's completed as follows:
>
> +++ b/Completion/Unix/Command/_git
> @@ -2026,11 +2026,7 @@ _git-worktree() {
>        curcontext=${curcontext%:*}-$line[1]:
>        case $line[1] in
>          (add)
>  	  if (( $words[(I)--detach] )); then
>  	    args=( ':commit:__git_commits' )
>  	  else
>  	    args=( ':branch:__git_branch_names' )
>  	  fi

I can't remember why it was done this way but I'm wondering if the
two halves of the if…else…fi are the opposite way around from the
intention. Specifying --detach would be superfluous for anything but
a branch because it is the default for any other committish so after
--detach there is sense in only completing branches. I often apply logic
such as this but it might not always be helpful, such as if --detach
comes from an alias.

Otherwise, as you say, any committish is valid and our usual approach
is to just complete them all and expect user's to set tag-order to give
preference to the matches they find most useful. I suggested years back
that we perhaps add a mechanism for giving a default tag-order and
perhaps should consider it again.

> I see __git_recent_branches but it's not precisely what I have in mind.

What would be useful in this case?

Oliver

Re: git worktree add /foo <TAB>

[Daniel Shahaf]

Oliver Kiddle wrote on Mon, Aug 14, 2017 at 13:57:33 +0200:
> On 9 Aug, Daniel Shahaf wrote:
> > The second argument to 'git worktree add' can be any committish.  That
> > committish becomes HEAD of the new worktree, and if it's a branch name,
> > then that branch becomes the current branch of the new worktree, else
> > the new worktree starts as a detached HEAD.
> >
> > How should that be completed?
> >
> > Currently it's completed as follows:
> >
> > +++ b/Completion/Unix/Command/_git
> > @@ -2026,11 +2026,7 @@ _git-worktree() {
> >        curcontext=${curcontext%:*}-$line[1]:
> >        case $line[1] in
> >          (add)
> >  	  if (( $words[(I)--detach] )); then
> >  	    args=( ':commit:__git_commits' )
> >  	  else
> >  	    args=( ':branch:__git_branch_names' )
> >  	  fi
> 
> I can't remember why it was done this way but I'm wondering if the
> two halves of the if…else…fi are the opposite way around from the
> intention. Specifying --detach would be superfluous for anything but
> a branch because it is the default for any other committish so after
> --detach there is sense in only completing branches.

+1

> I often apply logic such as this but it might not always be helpful,
> such as if --detach comes from an alias.

I don't understand.  Firstly, both zsh aliases and git aliases are
'seen through' in completion; secondly, the perfect shouldn't be the
enemy of the good.

> Otherwise, as you say, any committish is valid and our usual approach
> is to just complete them all and expect user's to set tag-order to give
> preference to the matches they find most useful. I suggested years back
> that we perhaps add a mechanism for giving a default tag-order and
> perhaps should consider it again.

Let's collect a few such settings before we invent a mechanism; it's
easier to invent when there are concrete details to work with.

To start with, here's one style I use:

    # Prefer recent commit object name completion
    zstyle ':completion::complete:git-(format-patch|send-email):argument-1:' group-order commits

> > I see __git_recent_branches but it's not precisely what I have in mind.
> 
> What would be useful in this case?

Just __git_branch_names — namely, names of local branches — is exactly
what I was after.  (Don't know how I managed to miss it)

So, yes, reversing the if and adding a tag-order/group-order style
should suffice.

Additionally, it would be nice if the group description said
"(alternatively specify name of new branch)" like _mkdir does.

Thanks,

Daniel

Re: git worktree add /foo <TAB>

[Oliver Kiddle]

Daniel Shahaf wrote:
> > intention. Specifying --detach would be superfluous for anything but
> > a branch because it is the default for any other committish so after
> > --detach there is sense in only completing branches.
> +1

> > I often apply logic such as this but it might not always be helpful,
> > such as if --detach comes from an alias.

> I don't understand.  Firstly, both zsh aliases and git aliases are
> 'seen through' in completion; secondly, the perfect shouldn't be the
> enemy of the good.

If --detach comes from an alias, the user didn't go to the effort of
typing it. So there's a higher chance of the user wanting the
pedantically correct (or perfect) behaviour of completing all commits
rather than our too clever by half (or good) behaviour. However, I tend
to find that cutting down the number of completion matches where
possible is nearly always helpful.

> > Otherwise, as you say, any committish is valid and our usual approach
> > is to just complete them all and expect user's to set tag-order to give
> > preference to the matches they find most useful. I suggested years back
> > that we perhaps add a mechanism for giving a default tag-order and
> > perhaps should consider it again.
>
> Let's collect a few such settings before we invent a mechanism; it's
> easier to invent when there are concrete details to work with.

Fair enough. I don't have any tag or group ordering styles for git
myself though I may have been put off trying on a couple of occasions
due to nested tag loops making it not entirely trivial.

> So, yes, reversing the if and adding a tag-order/group-order style
> should suffice.

Ok, patch below reverses the if.

> Additionally, it would be nice if the group description said
> "(alternatively specify name of new branch)" like _mkdir does.

Doesn't it need -b to create a new branch?

In such cases, compadd -x should be used for explanations instead of -X.

Oliver

diff --git a/Completion/Unix/Command/_git b/Completion/Unix/Command/_git
index 518e6d198..26554de05 100644
--- a/Completion/Unix/Command/_git
+++ b/Completion/Unix/Command/_git
@@ -2027,9 +2027,9 @@ _git-worktree() {
       case $line[1] in
         (add)
 	  if (( $words[(I)--detach] )); then
-	    args=( ':commit:__git_commits' )
-	  else
 	    args=( ':branch:__git_branch_names' )
+	  else
+	    args=( ':commit:__git_commits' )
 	  fi
           _arguments \
 	    '(-f --force)'{-f,--force}'[checkout branch even if already checked out in another worktree]' \