Re: [PATCH] check-ref-format: require a repository for --branch

Jeff King [off-list ref] writes:

On Mon, Jul 17, 2017 at 10:27:09AM -0700, Jonathan Nieder wrote:

quoted

 ...
I don't think it's right.  Today I can do

	$ cd /tmp
	$ git check-ref-format --branch master
	master

You might wonder why I'd ever do such a thing.  But that's what "git
check-ref-format --branch" is for --- it is for taking a <branch>
argument and turning it into a branch name.  For example, if you have
a script with an $opt_branch variable that defaults to "master", it
may do

	resolved_branch=$(git check-ref-format --branch "$opt_branch")

even though it is in a mode that not going to have to use
$resolved_branch and it is not running from a repository.

I'm not sure I buy that. What does "turning it into a branch name" even
mean when you are not in a repository? Clearly @{-1} must fail. And
everything else is just going to output the exact input you provided.
So any script calling "check-ref-format --branch" outside of a repo
would be better off not calling it at all.

I threw this topic in stalled category, hoping that one or the other
opinion eventually turns out to be more prevalent, but it didn't
seem to have happened X-<.

Things like @{-1} would not make any sense when the command is run
outside a repository, and the documentation is quite clear that it
is the primary reason why we added "--branch" option to the command,
i.e.

    With the `--branch` option, it expands the ``previous branch syntax''
    `@{-n}`.  For example, `@{-1}` is a way to refer the last branch you
    were on.  This option should be used by porcelains to accept this
    syntax anywhere a branch name is expected, so they can act as if you
    typed the branch name.

So I am tempted to take this patch to make sure that we won't gain
more people who abuse the command outside a repository.

Having said that, there may still be a use case where a Porcelain
script wants a way to see if a $name it has is appropriate as a
branch name before it has a repository (e.g. a wrapper to "git
clone" that wants to verify the name it is going to give to the "-b"
option), and a check desired in such a context is different from
(and is stricter than) feeding refs/heads/$name to the same command
without the "--branch" option.

So I think the right endgame in the longer term is:

 - Find (or add if it doesn't exist) a way to recommend to Porcelain
   scripts to use to expand an end-user generated string, and to map
   it to a branch name (it may be "rev-parse --symbolic-full-name
   $name"; I dunno).

 - Keep check-ref-format as (or revert it to be) a tool to "check".
   This would involve split strbuf_check_branch_ref() into two:

   - one that does not do the @{-1} thing and is used ONLY for
     format validity check (including rejecting a name that begins
     with a dash, which is OK for a random ref but not acceptable as
     a branch name);

   - the other that does @{-1} thing before doing the above.
   
   and then making the code call the former, not the latter.

The end result would be that check-ref-format becomes textual check
only, and can be usable (agian) outside repository, with or without
"--branch".  As the current code does not allow us do that yet, I
think it is safer to forbid use of --branch outside the repository
for now, purely as a bugfix.


[Footnote]

*1* In a sense, @{-1} is not something the scripts need to check its
    validity of---it is the branch you came from, so by definition
    it must be with a good name.  What the scripts want is instead
    see what the branch actually is, which is not what
    "check-ref-format" is about.

    a31dca03 ("check-ref-format --branch: give Porcelain a way to
    grok branch shorthand", 2009-03-21) says:

    The command may not be the best place to add this new feature, but
    
        $ git check-ref-format --branch "@{-1}"
    
    allows Porcelains to figure out what branch you were on before the last
    branch switching.