08:16 < abentley> Better phrasing: 'what circumstances should cause us to produce a working tree in a repository branch'?

[Denys Duchier]

Aaron Bentley <aaron.bentley at utoronto.ca> writes:

> My position is that "only a few
> combinations are useful, and we can break those out as
> separate commands".  You are arguing against that, so I believe you are
> arguing against commands like branch and init.

No, I merely offered an alternative (Robert suggested 3 other general design
directions).  It grouped creation of repository, branch and checkout under a
common "create" command.  That seemed fairly natural to me and worth some
consideration (if only to rule it out because it turns out to be less convenient
than a competing proposal).

>> The question was about the ability to
>> create the 3 man kinds of things that bzr offers: repository, branch, and
>> checkout.  "bzr branch" does not support that.
>
> Branch creates all of those.

You cannot create just a repository with "bzr branch"; that's the point; nor
just a checkout.

>> I think you still need to detect an existing .bzr dir and whether the command
>> invoked by the user is valid in this context.
>
> It is true that bzr create and bzr checkout would both need to avoid
> performing bad operations on existing directories.  But bzr create would
> also need to ensure that the options passed to it made sense.

Yes.  My point was that in both designs you cannot escape having to check that
what the user requested makes sense.  A very small point possibly in favour of
my suggestion is that you'd only have to check it in one command.

> I am saying "branch", "create-repository", and "checkout" are better
> commands than "create",

Ok, so now we have the beginning of a counter proposal.  That's a lot better
than just dissing my humble suggestion.  Perhaps we could look at the use cases
and see how these commands would work in practice (for example standalone vs
repo branch, etc...)

> because they mean that users who are only trying
> to create a repository do not have to read through information on
> creating checkouts.  All of the help on "create-repository" will be
> information that helps them do what they're trying to do right now.  I
> think that's a win.

I agree with that, but I don't believe that scrolling to the right use case
section when you request extensive help is going to be such a problem.  Finding
the right command is also a problem and that one is addressed by my suggestion.

I don't have the benefit of conventional experience in customer support: my only
"customers" have been colleagues in computer science and computational
linguistics.  They are far, far from being stupid people, but they have no
interest in VCS, and they can only barely remember a very small subset of CVS
commands.  They just want to write joint papers in a common archive.  Anything
beyond add, commit and update, and they'd send me email asking for help.  In
this respect, fewer commands seemed preferable.

BTW, I suspect that your choice of "branch", "create-repository" and "checkout"
is so that they are all verbs; but we also use two of these words as nouns: we
speak of a "branch", a "checkout", but not a "create-repository".  That last one
doesn't play nice with noun/verb puning.

Cheers,

--Denys