DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

Hi,

Peter and I just stumbled on the following interpretations of the
checkout command.

 git checkout origin/frotz      -> detaches HEAD
 git checkout -t origin/frotz   -> git checkout -b frotz origin/frotz
 git checkout frotz             -> git checkout -b frotz origin/frotz
 git checkout -t frotz          -> fatal: Missing branch name

I was surprised by the third one, which I was expecting to fail.
Apparently it's a DWIM feature introduced in commit 70c9ac2f. I
doubt that is what the user was trying to do if they ran that
command. If anything, this behavior will serve to confuse them.

I think git checkout -t <remote>/<branch> is short enough for
creating a new tracking branch. For something even shorter, we can
create an alias. But I'd prefer to keep the already complex
checkout command a little simpler.

Regards,
Clemens

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Sverre Rabbelier]

Heya,

On Sat, Jun 5, 2010 at 13:09, Clemens Buchacher <drizzd@aon.at> wrote:
> I think git checkout -t <remote>/<branch> is short enough for
> creating a new tracking branch. For something even shorter, we can
> create an alias. But I'd prefer to keep the already complex
> checkout command a little simpler.

You're a little late to the party ;). Did you read the original thread
discussing this?

-- 
Cheers,

Sverre Rabbelier

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sat, Jun 05, 2010 at 03:29:48PM +0200, Sverre Rabbelier wrote:
>
> On Sat, Jun 5, 2010 at 13:09, Clemens Buchacher <drizzd@aon.at> wrote:
> > I think git checkout -t <remote>/<branch> is short enough for
> > creating a new tracking branch. For something even shorter, we can
> > create an alias. But I'd prefer to keep the already complex
> > checkout command a little simpler.
> 
> You're a little late to the party ;). Did you read the original thread
> discussing this?

Well, I just noticed this by accident, it is not documented
anywhere, and you cannot expected me to know everything that is
going on on the list. Should I not offer my opinion because of
that?

I could not make much sense of the thread that came with the patch.
The discussion first seems to conclude that it is a bad idea, only
to implement it anyway.

Well, I also think it is a bad idea, and I am not surprised to find
users confused by it.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Sverre Rabbelier]

Heya,

On Sat, Jun 5, 2010 at 15:58, Clemens Buchacher <drizzd@aon.at> wrote:
> Well, I just noticed this by accident, it is not documented
> anywhere, and you cannot expected me to know everything that is
> going on on the list. Should I not offer my opinion because of
> that?

No, I just meant that because you are late to the party you should
first read the original thread, and then perhaps summarize it, like
you have done below.

> I could not make much sense of the thread that came with the patch.
> The discussion first seems to conclude that it is a bad idea, only
> to implement it anyway.

I think it would be useful here to have some links to relevant posts,
or at least to the relevant thread.

> Well, I also think it is a bad idea, and I am not surprised to find
> users confused by it.

I think it was an idea from the last mentor summit, Peff?

-- 
Cheers,

Sverre Rabbelier

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sat, Jun 05, 2010 at 04:03:30PM +0200, Sverre Rabbelier wrote:
> On Sat, Jun 5, 2010 at 15:58, Clemens Buchacher <drizzd@aon.at> wrote:
> 
> > I could not make much sense of the thread that came with the patch.
> > The discussion first seems to conclude that it is a bad idea, only
> > to implement it anyway.
> 
> I think it would be useful here to have some links to relevant posts,
> or at least to the relevant thread.

The discussion ends with this posting
http://mid.gmane.org/7viqesz3mk.fsf@alter.siamese.dyndns.org ,
which basically says "we all agree this is bad."

Then Junio changes his mind his mind and posts the patches.
http://mid.gmane.org/7vzl7pyvzl.fsf@alter.siamese.dyndns.org

Whether the feature has merit or not is suddenly not an issue any
more and the discussion veers off in other directions about minor
technical details and about the --nodwim/--noguess option, the
notion of which I find ridiculous by the way. But it further
illustrates how random and anti-DWIM this feature really is.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Nicolas Pitre]

On Sat, 5 Jun 2010, Clemens Buchacher wrote:

> On Sat, Jun 05, 2010 at 04:03:30PM +0200, Sverre Rabbelier wrote:
> > On Sat, Jun 5, 2010 at 15:58, Clemens Buchacher <drizzd@aon.at> wrote:
> > 
> > > I could not make much sense of the thread that came with the patch.
> > > The discussion first seems to conclude that it is a bad idea, only
> > > to implement it anyway.
> > 
> > I think it would be useful here to have some links to relevant posts,
> > or at least to the relevant thread.
> 
> The discussion ends with this posting
> http://mid.gmane.org/7viqesz3mk.fsf@alter.siamese.dyndns.org ,
> which basically says "we all agree this is bad."
> 
> Then Junio changes his mind his mind and posts the patches.
> http://mid.gmane.org/7vzl7pyvzl.fsf@alter.siamese.dyndns.org
> 
> Whether the feature has merit or not is suddenly not an issue any
> more and the discussion veers off in other directions about minor
> technical details and about the --nodwim/--noguess option, the
> notion of which I find ridiculous by the way. But it further
> illustrates how random and anti-DWIM this feature really is.

Amen.


Nicolas

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Jeff King]

On Sat, Jun 05, 2010 at 04:03:30PM +0200, Sverre Rabbelier wrote:

> On Sat, Jun 5, 2010 at 15:58, Clemens Buchacher <drizzd@aon.at> wrote:
> > Well, I just noticed this by accident, it is not documented
> > anywhere, and you cannot expected me to know everything that is
> > going on on the list. Should I not offer my opinion because of
> > that?
> 
> No, I just meant that because you are late to the party you should
> first read the original thread, and then perhaps summarize it, like
> you have done below.

Moreover, the burden for arguing against it is a little higher. It is no
longer "I think this is a bad idea" but "I think this is a bad idea, and
it is so bad that it is worth changing behavior that users may have
become accustomed to".

> > Well, I also think it is a bad idea, and I am not surprised to find
> > users confused by it.
> 
> I think it was an idea from the last mentor summit, Peff?

No, I complained about it at the mentor summit, but the idea predates
it. I think all of the relevant discussion is on the list.

-Peff

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Matthieu Moy]

Clemens Buchacher <drizzd@aon.at> writes:

> Hi,
>
> Peter and I just stumbled on the following interpretations of the
> checkout command.
>
>  git checkout origin/frotz      -> detaches HEAD
>  git checkout -t origin/frotz   -> git checkout -b frotz origin/frotz
>  git checkout frotz             -> git checkout -b frotz origin/frotz
>  git checkout -t frotz          -> fatal: Missing branch name
>
> I was surprised by the third one, which I was expecting to fail.
> Apparently it's a DWIM feature introduced in commit 70c9ac2f. I
> doubt that is what the user was trying to do if they ran that
> command. If anything, this behavior will serve to confuse them.

As already mentionned, it's obviously a bit late to revert the commit
which itroduced this DWIM. I tend to agree that this is confusing, but
still, this is very often what the user actually wanted. The original
patch was to sugest the user the appropriate command, and I find the
motivating example rather convincing:

http://thread.gmane.org/gmane.comp.version-control.git/129587

$ git clone git://git.kernel.org/pub/scm/git/git.git
$ cd git
$ git checkout next
error: pathspec 'next' did not match any file(s) known to git.
To create a local branch from the same named remote branch, use
  git checkout -b next origin/next

The newbie does not immediately understand the difference between
local and remote branches, he's been told about the branch "next", and
wants to check it out.

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sun, Jun 06, 2010 at 06:18:56PM +0200, Matthieu Moy wrote:

> $ git clone git://git.kernel.org/pub/scm/git/git.git
> $ cd git
> $ git checkout next
> error: pathspec 'next' did not match any file(s) known to git.
> To create a local branch from the same named remote branch, use
>   git checkout -b next origin/next
> 
> The newbie does not immediately understand the difference between
> local and remote branches, he's been told about the branch "next", and
> wants to check it out.

That is exactly my point. Trying to read the user's mind is not a
solution to this problem. The user does not even know what they are
doing and will inevitably end up even more confused.

The suggestion above would be perfect. It is an easy and obvious
solution, and the user is even educated about their mistake.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sun, Jun 06, 2010 at 12:18:05PM -0400, Jeff King wrote:

> Moreover, the burden for arguing against it is a little higher. It is no
> longer "I think this is a bad idea" but "I think this is a bad idea, and
> it is so bad that it is worth changing behavior that users may have
> become accustomed to".

I argue that people are more likely to be surprised and confused by
this feature rather than being accustomed to it already. I am also
waiting to hear from someone who actively uses this feature.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Jacob Helwig]

On Sun, Jun 6, 2010 at 09:55, Clemens Buchacher <drizzd@aon.at> wrote:
> On Sun, Jun 06, 2010 at 12:18:05PM -0400, Jeff King wrote:
>
>> Moreover, the burden for arguing against it is a little higher. It is no
>> longer "I think this is a bad idea" but "I think this is a bad idea, and
>> it is so bad that it is worth changing behavior that users may have
>> become accustomed to".
>
> I argue that people are more likely to be surprised and confused by
> this feature rather than being accustomed to it already. I am also
> waiting to hear from someone who actively uses this feature.
>

I am someone that "actively" uses this feature.  I will often have
repositories that I use on multiple machines, and will forget which
remote branches I have local tracking branches for, on which machines.
 I'll end up just doing `git checkout $random-branch`, thinking I'm on
one of the machines where I've already setup the local branch, and
find it quite handy that Git _does_ DWIM.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sun, Jun 06, 2010 at 09:59:08AM -0700, Jacob Helwig wrote:
> On Sun, Jun 6, 2010 at 09:55, Clemens Buchacher <drizzd@aon.at> wrote:
> >
> > I argue that people are more likely to be surprised and confused by
> > this feature rather than being accustomed to it already. I am also
> > waiting to hear from someone who actively uses this feature.
> >
> 
> I am someone that "actively" uses this feature.  I will often have
> repositories that I use on multiple machines, and will forget which
> remote branches I have local tracking branches for, on which machines.
>  I'll end up just doing `git checkout $random-branch`, thinking I'm on
> one of the machines where I've already setup the local branch, and
> find it quite handy that Git _does_ DWIM.

I see. But would it be so terrible to have to type "git checkout -t
origin/branch" instead?

Also, isn't this more like "foretell what I really want" rather
than "do what it means"? Who would guess that "git checkout
$branch" means "create $branch tracking <random-remote>/$branch"?

And this is exactly _why_ it can be marginally useful if the
foretelling is correct, but all the more confusing if it's not.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Sverre Rabbelier]

Heya,

On Sun, Jun 6, 2010 at 19:32, Clemens Buchacher <drizzd@aon.at> wrote:
> Also, isn't this more like "foretell what I really want" rather
> than "do what it means"? Who would guess that "git checkout
> $branch" means "create $branch tracking <random-remote>/$branch"?

Note that DWIM stands for "Do What I Mean", not "Do What It Means".

-- 
Cheers,

Sverre Rabbelier

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Johan Herland]

On Sunday 06 June 2010, Clemens Buchacher wrote:
> On Sun, Jun 06, 2010 at 12:18:05PM -0400, Jeff King wrote:
> > Moreover, the burden for arguing against it is a little higher. It is
> > no longer "I think this is a bad idea" but "I think this is a bad
> > idea, and it is so bad that it is worth changing behavior that users
> > may have become accustomed to".
> 
> I argue that people are more likely to be surprised and confused by
> this feature rather than being accustomed to it already. I am also
> waiting to hear from someone who actively uses this feature.

I use this feature all the time, and find it very nice and useful.

Moreover, at $DAYJOB, we have a lot of Git newbies (many of them still can't 
properly juggle remote branches. remote-tracking branches, and local 
branches with tracking info). They mostly work on local topic branches that 
tracks corresponding shared topic branches in a central repo. For them, 
switching to one of these topic branches is simply a matter of 'git checkout 
<branchname>', without having to care (or know) if you've already created 
the local branch or not. In this case, Git really does the Right Thing (tm) 
for us.

In several cases this DWIMmery has prompted people to update from an old Git 
version, just to get this feature.


...Johan

-- 
Johan Herland, <johan@herland.net>
www.herland.net

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Jacob Helwig]

On Sun, Jun 6, 2010 at 10:32, Clemens Buchacher <drizzd@aon.at> wrote:
> On Sun, Jun 06, 2010 at 09:59:08AM -0700, Jacob Helwig wrote:
>> On Sun, Jun 6, 2010 at 09:55, Clemens Buchacher <drizzd@aon.at> wrote:
>> >
>> > I argue that people are more likely to be surprised and confused by
>> > this feature rather than being accustomed to it already. I am also
>> > waiting to hear from someone who actively uses this feature.
>> >
>>
>> I am someone that "actively" uses this feature.  I will often have
>> repositories that I use on multiple machines, and will forget which
>> remote branches I have local tracking branches for, on which machines.
>>  I'll end up just doing `git checkout $random-branch`, thinking I'm on
>> one of the machines where I've already setup the local branch, and
>> find it quite handy that Git _does_ DWIM.
>
> I see. But would it be so terrible to have to type "git checkout -t
> origin/branch" instead?
>

Given that `git checkout branch` already exists & works: Yes.

You're talking about removing a feature, so (as already pointed out)
there's a fairly high burden of proof to show that it's actually a
mis-feature, that is so dangerous/confusing that it _needs_ to be
removed.

> Also, isn't this more like "foretell what I really want" rather
> than "do what it means"? Who would guess that "git checkout
> $branch" means "create $branch tracking <random-remote>/$branch"?
>

This is very much "fortell what I really want", which is exactly what
DWIM (Do What I Mean) means (as Sverre already pointed out).  I'm not
guessing that `git checkout $branch` means `git branch -t $branch
origin/$branch && git checkout $branch`; I'm assuming that `git branch
-t $branch origin/$branch` was already done at some point in the past,
and Git is adding that bit in for me, which is what _I_ really meant
to do.

> And this is exactly _why_ it can be marginally useful if the
> foretelling is correct, but all the more confusing if it's not.
>

Other than "unexpected success", how is the DWIM behavior confusing,
given that it says exactly what it's doing when the DWIM behavior is
invoked?  I'm still not clear what the confusion you're referring to
is on this one.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Miles Bader]

Clemens Buchacher <drizzd@aon.at> writes:
> The suggestion above would be perfect. It is an easy and obvious
> solution, and the user is even educated about their mistake.

Of course, having been educated as to what's going on, the user would
then be annoyed that they had to type all those boilerplate args when
git clearly knew what they wanted to do... and that would be the case
every time from then on...

I think this DWIM is actually pretty convenient, and very often does
reflect what the user intuitively is trying to do when giving such args.

Given that git _does_ tell you what it's doing, and that it's easy
enough to delete the new branch if it wasn't really wanted, it seems
pretty harmless as well.  A campaign to delete this feature seems kind
of silly...

-Miles

-- 
I'm beginning to think that life is just one long Yoko Ono album; no rhyme
or reason, just a lot of incoherent shrieks and then it's over.  --Ian Wolff

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Paolo Bonzini]

On 06/06/2010 06:46 PM, Clemens Buchacher wrote:
> That is exactly my point. Trying to read the user's mind is not a
> solution to this problem. The user does not even know what they are
> doing and will inevitably end up even more confused.

Maybe you just want branch.autosetupmerge = false?

Paolo

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Sun, Jun 06, 2010 at 02:26:08PM -0700, Jacob Helwig wrote:
>
> > And this is exactly _why_ it can be marginally useful if the
> > foretelling is correct, but all the more confusing if it's not.
> >
> 
> Other than "unexpected success", how is the DWIM behavior confusing,
> given that it says exactly what it's doing when the DWIM behavior is
> invoked?  I'm still not clear what the confusion you're referring to
> is on this one.

I am telling git to checkout a branch. Instead it creates a branch.
That is what is confusing to me. Until I found the commit that
introduced it, I was sure it must be a bug.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Mon, Jun 07, 2010 at 03:41:58PM +0900, Miles Bader wrote:
> Clemens Buchacher <drizzd@aon.at> writes:
> > The suggestion above would be perfect. It is an easy and obvious
> > solution, and the user is even educated about their mistake.
> 
> Of course, having been educated as to what's going on, the user would
> then be annoyed that they had to type all those boilerplate args when
> git clearly knew what they wanted to do... and that would be the case
> every time from then on...

Why should the user make the same mistake over and over again?

> I think this DWIM is actually pretty convenient, and very often does
> reflect what the user intuitively is trying to do when giving such args.
> 
> Given that git _does_ tell you what it's doing, and that it's easy
> enough to delete the new branch if it wasn't really wanted, it seems
> pretty harmless as well.  A campaign to delete this feature seems kind
> of silly...

It may be harmless to users who know what's going on. I can
certainly deal with this feature, whether it's there or not.

But this is supposedly a feature which helps users who type "git
checkout <branch>" by mistake, when they really wanted to do "git
checkout -t <remote>/<branch>". I am certain that most new users
who make this mistake will not understand what's going on, even if
they read the output.

I believe that it's because of things like this that many users
still consider git to be complicated and hard to use. That's what
really bothers me.

And it makes me sad that you think it silly to even talk about it.
Even if the feature does not end up getting removed I still hope
that we will exercise more caution in the future and try to solve
the real problem--which appears to be remote branch
handling--rather than introducing more strange behavior.

Regards,
Clemens

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Matthieu Moy]

Clemens Buchacher <drizzd@aon.at> writes:

> But this is supposedly a feature which helps users who type "git
> checkout <branch>" by mistake, when they really wanted to do "git
> checkout -t <remote>/<branch>".

Not sure what's the argument here, but aren't the two commands
equivalent? Do you prefer the second syntax "git checkout -t
<remote>/<branch>"? It's already a DWIM for "git checkout -b <branch>
-t <remote>/<branch>", and I find this one far more confusing:

git checkout    <remote>/<branch> => detaches HEAD
git checkout -t <remote>/<branch> => creates a local branch automatically

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Mon, Jun 07, 2010 at 09:17:25PM +0200, Matthieu Moy wrote:
> Clemens Buchacher <drizzd@aon.at> writes:
> 
> > But this is supposedly a feature which helps users who type "git
> > checkout <branch>" by mistake, when they really wanted to do "git
> > checkout -t <remote>/<branch>".
> 
> Not sure what's the argument here, but aren't the two commands
> equivalent? Do you prefer the second syntax "git checkout -t
> <remote>/<branch>"? It's already a DWIM for "git checkout -b <branch>
> -t <remote>/<branch>", and I find this one far more confusing:
> 
> git checkout    <remote>/<branch> => detaches HEAD
> git checkout -t <remote>/<branch> => creates a local branch automatically

The intent with -t is clear. It is used only when you create a new
branch. Also, you specify the remote branch you're going to create
a new branch from.

"git checkout <branch>", on the other hand, will create a branch
based on a remote branch, even though you neither asked for a new
branch, nor did you specify any remote at all.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Bruce Stephens]

Clemens Buchacher <drizzd@aon.at> writes:

[...]

> "git checkout <branch>", on the other hand, will create a branch
> based on a remote branch, even though you neither asked for a new
> branch, nor did you specify any remote at all.

Provided such a branch doesn't already exist and that only one remote
matches.  I agree it's magic (and understand why people might not like
it) but personally I find it convenient.

Possibly (because it hides the distinction between local and remote
branches a little) it would be better opt-in than the default, but I
think a reasonable case could be made in both directions.

[...]

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Jan Krüger]

Clemens Buchacher <drizzd@aon.at> wrote:

> > Other than "unexpected success", how is the DWIM behavior confusing,
> > given that it says exactly what it's doing when the DWIM behavior is
> > invoked?  I'm still not clear what the confusion you're referring to
> > is on this one.
> 
> I am telling git to checkout a branch. Instead it creates a branch.
> That is what is confusing to me. Until I found the commit that
> introduced it, I was sure it must be a bug.

I usually see the opposite kind of confusion on #git: someone cloned a
repository and wants to work on one of the branches. After all, we tell
everyone that clone copies all the history.

So they type "git checkout <that branch>"... and they get a weird
error (what's a pathspec, anyway?). OMGWTFBBQ! Not knowing what's going
on, they drop by in #git and hear they need to type something much less
straightforward than "git checkout <that branch>". They don't really
know why, so they probably assume it's because git is just so damn
complicated and overengineered.

I have yet to see any newish users complain about the new syntax, by the
way. You don't qualify, sorry. ;)

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Mon, Jun 07, 2010 at 10:11:34PM +0200, Jan Krüger wrote:
> Clemens Buchacher <drizzd@aon.at> wrote:
> 
> > I am telling git to checkout a branch. Instead it creates a branch.
> > That is what is confusing to me. Until I found the commit that
> > introduced it, I was sure it must be a bug.
> 
> I usually see the opposite kind of confusion on #git: someone cloned a
> repository and wants to work on one of the branches. After all, we tell
> everyone that clone copies all the history.
> 
> So they type "git checkout <that branch>"... and they get a weird
> error (what's a pathspec, anyway?).

I agree, that's a less than perfect error message. So let's improve
it. What about "unknown branch or file name"? Possibly also output
the most common synopsis?

> OMGWTFBBQ! Not knowing what's going
> on, they drop by in #git and hear they need to type something much less
> straightforward than "git checkout <that branch>". They don't really
> know why, so they probably assume it's because git is just so damn
> complicated and overengineered.

Well, but that's how it is. If we want more "global" branches, then
we will have to think about how to achieve that. But hiding what we
really have is not going to make it easier to understand.

> I have yet to see any newish users complain about the new syntax, by the
> way. You don't qualify, sorry. ;)

I started this entire thread only because someone on #git was
confused by it.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Miles Bader]

On Tue, Jun 8, 2010 at 3:54 AM, Clemens Buchacher <drizzd@aon.at> wrote:
>> then be annoyed that they had to type all those boilerplate args when
>> git clearly knew what they wanted to do... and that would be the case
>> every time from then on...
>
> Why should the user make the same mistake over and over again?

What I mean is that every time thereafter that they want that
functionality (making a remote branch local), they'll have to type the
"long form".

> And it makes me sad that you think it silly to even talk about it.

I don't think it's silly to note your concern, I think it's silly to
put a fair amount of effort into a campaign for its removal, which is
what you are doing.  If new points can be made, fair enough, but so
far it seems mainly a lot of back and forth stating the same things
over and over.

> I believe that it's because of things like this that many users
> still consider git to be complicated and hard to use. That's what
> really bothers me.

Why do you think that?  I mean, you could be right, but it's certainly
conceivable that this feature causes users to consider git _simpler_
to use, because it lets them think about local branches only.

-Miles

-- 
Do not taunt Happy Fun Ball.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Tue, Jun 08, 2010 at 09:25:07AM +0900, Miles Bader wrote:
>
> > And it makes me sad that you think it silly to even talk about it.
> 
> I don't think it's silly to note your concern, I think it's silly to
> put a fair amount of effort into a campaign for its removal, which is
> what you are doing.  If new points can be made, fair enough, but so
> far it seems mainly a lot of back and forth stating the same things
> over and over.

Since I started this, I feel responsible for responding to people's
comments. I do not think I am doing anything wrong. So please
forgive me for responding to you as well, even if you think it
silly of me.

> > I believe that it's because of things like this that many users
> > still consider git to be complicated and hard to use. That's what
> > really bothers me.
> 
> Why do you think that?  I mean, you could be right, but it's certainly
> conceivable that this feature causes users to consider git _simpler_
> to use, because it lets them think about local branches only.

Thinking about local branches only is wrong.

Let's assume the user "just wants to check out branch X." Such a
use case has been mentioned previously in this thread. So they say
"git checkout X", and voila, they get what they wanted.

It did what they meant it to do. Everybody is happy. But not for
long.

Because the next time they "just want to check out branch X," they
will remember, it's time for "git checkout X." But what's this?
This is not what they wanted. It does not have the latest changes
on branch X! Didn't they do the same thing they did last time?

No they did not, because last time they _created a branch_.
Eventually, if they are lucky, they will learn in #git that they
were supposed to "git merge <remote>/X." And they will think "WTF?
Why do I have to _merge_? And what's this business about remotes?
Why is git is so complicated?"

So all we did was delay the user's confusion, and make it worse,
because now the mind model of "local branches only" they have
developed, is broken.

Git uses slightly different semantics than other VCSs, especially
when it comes to branches. And that often causes users to become
confused. But I happen to think that git's branch model is
superior, and easy to understand once you unlearn what you know
from other VCSs.

But if we try to hide that fact and imitate other VCSs on the
surface, that will only make it harder to learn.

Regards,
Clemens

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[demerphq]

On 8 June 2010 09:29, Clemens Buchacher <drizzd@aon.at> wrote:
> No they did not, because last time they _created a branch_.
> Eventually, if they are lucky, they will learn in #git that they
> were supposed to "git merge <remote>/X." And they will think "WTF?
> Why do I have to _merge_? And what's this business about remotes?
> Why is git is so complicated?"

I'm confused. Wont they actually be told:

  git checkout $foo
  git pull
  git push

Why would they need to do a git merge $remote/X?

The only reason I can think of is that they are doing cross branch
merges, but then they are by default beyond the novice user and can be
expected to know various things, and to have read the docs.

Yves




-- 
perl -Mre=debug -e "/just|another|perl|hacker/"

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Miles Bader]

On Tue, Jun 8, 2010 at 4:29 PM, Clemens Buchacher <drizzd@aon.at> wrote:
> Eventually, if they are lucky, they will learn in #git that they
> were supposed to "git merge <remote>/X."

They don't have to do that, they just have to do "git pull".  This
concept -- that to get the latest changes, one pulls -- is something
people learn very early.

So, in fact it seems quite natural -- you switch to a local branch
with "git co branch", and then if git tells you it's not up-to-date,
you "git pull".  All nice and smooth and simple and standard.  The
DWIM simply makes the first time less clumsy, and while everybody
benefits, beginners probably benefit a bit more.

-Miles

-- 
Do not taunt Happy Fun Ball.

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Jeff King]

On Tue, Jun 08, 2010 at 09:29:58AM +0200, Clemens Buchacher wrote:

> Let's assume the user "just wants to check out branch X." Such a
> use case has been mentioned previously in this thread. So they say
> "git checkout X", and voila, they get what they wanted.
> 
> It did what they meant it to do. Everybody is happy. But not for
> long.
> 
> Because the next time they "just want to check out branch X," they
> will remember, it's time for "git checkout X." But what's this?
> This is not what they wanted. It does not have the latest changes
> on branch X! Didn't they do the same thing they did last time?
> 
> No they did not, because last time they _created a branch_.
> Eventually, if they are lucky, they will learn in #git that they
> were supposed to "git merge <remote>/X." And they will think "WTF?
> Why do I have to _merge_? And what's this business about remotes?
> Why is git is so complicated?"

FWIW, I made this same complaint at the GitTogether, and the response I
got is that git does help the user along through the second "checkout".
You get something like:

  $ git checkout foo
  Branch private set up to track remote branch foo from origin.
  Switched to a new branch 'foo'

  ... time passes ...

  $ git checkout foo
  Switched to branch 'foo'
  Your branch is behind 'origin/foo' by 5 commits, and can be fast-forwarded.

Now you can certainly argue that a clueless user won't know what to do
with those messages, but it is not entirely true that git leaves them
without any advice. Perhaps we could lessen the "wtf" moment in your
example above by making that advice better.

-Peff

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Michael J Gruber]

Clemens Buchacher venit, vidit, dixit 07.06.2010 21:32:
> On Mon, Jun 07, 2010 at 09:17:25PM +0200, Matthieu Moy wrote:
>> Clemens Buchacher <drizzd@aon.at> writes:
>>
>>> But this is supposedly a feature which helps users who type "git
>>> checkout <branch>" by mistake, when they really wanted to do "git
>>> checkout -t <remote>/<branch>".
>>
>> Not sure what's the argument here, but aren't the two commands
>> equivalent? Do you prefer the second syntax "git checkout -t
>> <remote>/<branch>"? It's already a DWIM for "git checkout -b <branch>
>> -t <remote>/<branch>", and I find this one far more confusing:
>>
>> git checkout    <remote>/<branch> => detaches HEAD
>> git checkout -t <remote>/<branch> => creates a local branch automatically
> 
> The intent with -t is clear. It is used only when you create a new
> branch. Also, you specify the remote branch you're going to create
> a new branch from.
> 
> "git checkout <branch>", on the other hand, will create a branch
> based on a remote branch, even though you neither asked for a new
> branch, nor did you specify any remote at all.

You're making a very important point here:

The existing DWIMery executes *different commands* depending on the
circumstances (<branch> existing or not). "checkout" and "checkout
-t/-b" really are different commands.

For me, DWIMery is OK in these cases:

- completing refs (<name> may be refs/heads/<name> or refs/tags/<name> etc.)

- Adding options without which the other options don't make sense
(independent of the circumstances), such adding "-b" for "-t", or, even
doing the DWIMery above since "--track" is requested explicitly.

Changing the command (mode) is something fundamentally different.

[In this specific case, it also keeps the user from learning what's
going on, but that's a different issue.]

But I'm afraid it's too late.

Michael

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[demerphq]

On 8 June 2010 10:07, Michael J Gruber <git@drmicha.warpmail.net> wrote:
> Clemens Buchacher venit, vidit, dixit 07.06.2010 21:32:
>> On Mon, Jun 07, 2010 at 09:17:25PM +0200, Matthieu Moy wrote:
>>> Clemens Buchacher <drizzd@aon.at> writes:
>>>
>>>> But this is supposedly a feature which helps users who type "git
>>>> checkout <branch>" by mistake, when they really wanted to do "git
>>>> checkout -t <remote>/<branch>".
>>>
>>> Not sure what's the argument here, but aren't the two commands
>>> equivalent? Do you prefer the second syntax "git checkout -t
>>> <remote>/<branch>"? It's already a DWIM for "git checkout -b <branch>
>>> -t <remote>/<branch>", and I find this one far more confusing:
>>>
>>> git checkout    <remote>/<branch> => detaches HEAD
>>> git checkout -t <remote>/<branch> => creates a local branch automatically
>>
>> The intent with -t is clear. It is used only when you create a new
>> branch. Also, you specify the remote branch you're going to create
>> a new branch from.
>>
>> "git checkout <branch>", on the other hand, will create a branch
>> based on a remote branch, even though you neither asked for a new
>> branch, nor did you specify any remote at all.
>
> You're making a very important point here:
>
> The existing DWIMery executes *different commands* depending on the
> circumstances (<branch> existing or not). "checkout" and "checkout
> -t/-b" really are different commands.

"checkout a branch" is the command. The rest is just bookkeeping.

Yves



-- 
perl -Mre=debug -e "/just|another|perl|hacker/"

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Michael J Gruber]

demerphq venit, vidit, dixit 08.06.2010 10:18:
> On 8 June 2010 10:07, Michael J Gruber <git@drmicha.warpmail.net> wrote:
>> Clemens Buchacher venit, vidit, dixit 07.06.2010 21:32:
>>> On Mon, Jun 07, 2010 at 09:17:25PM +0200, Matthieu Moy wrote:
>>>> Clemens Buchacher <drizzd@aon.at> writes:
>>>>
>>>>> But this is supposedly a feature which helps users who type "git
>>>>> checkout <branch>" by mistake, when they really wanted to do "git
>>>>> checkout -t <remote>/<branch>".
>>>>
>>>> Not sure what's the argument here, but aren't the two commands
>>>> equivalent? Do you prefer the second syntax "git checkout -t
>>>> <remote>/<branch>"? It's already a DWIM for "git checkout -b <branch>
>>>> -t <remote>/<branch>", and I find this one far more confusing:
>>>>
>>>> git checkout    <remote>/<branch> => detaches HEAD
>>>> git checkout -t <remote>/<branch> => creates a local branch automatically
>>>
>>> The intent with -t is clear. It is used only when you create a new
>>> branch. Also, you specify the remote branch you're going to create
>>> a new branch from.
>>>
>>> "git checkout <branch>", on the other hand, will create a branch
>>> based on a remote branch, even though you neither asked for a new
>>> branch, nor did you specify any remote at all.
>>
>> You're making a very important point here:
>>
>> The existing DWIMery executes *different commands* depending on the
>> circumstances (<branch> existing or not). "checkout" and "checkout
>> -t/-b" really are different commands.
> 
> "checkout a branch" is the command. The rest is just bookkeeping.

I'm sorry but that is just plain silly.

We have several "commands" which do very different things depending on
options, i.e. comprise different commands/command modes - which is why I
used the phrase command (mode) in the part you chose to leave out.

One mode is checking out an existing branch.

One mode is creating a new branch and checking it out, i.e. doing a "git
branch" followed by a "git checkout" proper.

Current DWIMery switches from the first to the second depending on your
existing branches. The only reason this could sneak in is the fact that
it is non-destructive, just (over-) eagerly constructive.

Michael

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Matthieu Moy]

demerphq <demerphq@gmail.com> writes:

> On 8 June 2010 09:29, Clemens Buchacher <drizzd@aon.at> wrote:
>> No they did not, because last time they _created a branch_.
>> Eventually, if they are lucky, they will learn in #git that they
>> were supposed to "git merge <remote>/X." And they will think "WTF?
>> Why do I have to _merge_? And what's this business about remotes?
>> Why is git is so complicated?"
>
> I'm confused. Wont they actually be told:
>
>   git checkout $foo
>   git pull
>   git push

Right. Just like they learnt clone/push/pull before knowing about
remote and tracking branches, because clone did the magic for them
when creating the repo.

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/

Re: DWIM "git checkout frotz" to "git checkout -b frotz origin/frotz"

[Clemens Buchacher]

On Tue, Jun 08, 2010 at 03:52:42AM -0400, Jeff King wrote:
>
>   $ git checkout foo
>   Branch private set up to track remote branch foo from origin.
>   Switched to a new branch 'foo'
> 
>   ... time passes ...
> 
>   $ git checkout foo
>   Switched to branch 'foo'
>   Your branch is behind 'origin/foo' by 5 commits, and can be fast-forwarded.

I see. That really is somewhat reassuring. I still do not like it,
but considering the fact that I could not come up with a scenario
worse than that, I suppose the feature is not as bad as I made it
out to be.

Thanks all for discussing it with me anyways.

Regards,
Clemens