Re: [PATCH v3] build: add default aliases

From: Felipe Contreras <felipe.contreras@gmail.com>
Date: Sep 29, 2013, 03:37 UTC
Message-ID: <CAMP44s12c2_nEkwzgWXYzaJTeymSyLbDMWoRygMwZv8-m8wG4g@mail.gmail.com>
In-Reply-To: <52479C04.8060000@alum.mit.edu>

On Sat, Sep 28, 2013 at 10:18 PM, Michael Haggerty <mhagger@alum.mit.edu> wrote:

Show 42 quoted lines

> On 09/29/2013 12:41 AM, Felipe Contreras wrote:
>> On Tue, Sep 24, 2013 at 1:39 PM, Jonathan Nieder <jrnieder@gmail.com> wrote:
>>>> On Sat, Sep 21, 2013 at 02:20:21PM -0500, Felipe Contreras wrote:
>>>>> For now simply add a few common aliases.
>>>>>
>>>>>   co = checkout
>>>>>   ci = commit
>>>>>   rb = rebase
>>>>>   st = status
>>>>
>>>> [...]
>>
>>> But making 'ci' a synonym of another command by default while still
>>> keeping its definition configurable would be doing people a
>>> disservice, I fear.
>>
>> And I and many (most) users disagree.
>>
>>> As long as 'ci' works out of the box, it will
>>> start showing up in examples and used in suggestions over IRC, etc,
>>> which is great.
>
> ...and in scripts.
>
>> It might, or...
>>
>>> Unfortunately that means that anyone who has 'ci'
>>> defined to mean something different can no longer use those examples,
>>> that advice from IRC, and so on.  So in the world where 'ci' is a
>>> synonym for 'commit' by default, while people still *can* redefine
>>> 'ci' to include whatever options they like (e.g., "-a"), actually
>>> carrying out such a personal customization is asking for trouble.
>>
>> Precisely for this reason it might not. If people know aliases can be
>> different in different machines, they would avoid them in
>> documentation which is meant for all machines.
>
> My experience contradicts your prediction.  I have 'ci'/'co' aliases in
> my own configuration.  But even though I am aware of the fact that other
> people might not have the same aliases, I have on multiple occasions
> used them in documentation and/or scripts meant for other people.  The
> muscle memory is just too strong.

If you are already making that error, then this patch wouldn't make it any worst. In fact, it would make the situation better.

If previously you had 10 persons complaining about the "ci" command not working, now 9 of them wouldn't complain because "git ci" does actually work, even if you have aliased it to something slightly different, like "commit -v". Instead, you would have 1 person complaining, because he has a different alias, which makes the command fail somehow. In reality, that 1 person might not even exist.

The solution before and after my patch is the same; avoid the 'ci'/'co' aliases.

> My error was discovered by other people who didn't have those aliases.

And after this patch it still will.

> If *most* people had the same aliases as I did, and others had defined
> their own slightly different ones, then the scripts would have subtly
> malfunctioned for the latter set of users and I would have had trouble
> reproducing the errors.

Doubtful. But a warning that a default alias is being used, or simply showing the actual command in the standard output would fix the problem.

It certainly looks like you are not even looking for solutions for the hypothetical problems you put forward.

> So, even though I think such aliases are a great convenience factor, I
> am -0 on including pre-defined but overrideable aliases and -1 on
> including pre-defined and non-overrideable aliases.

And still, somehow every other VCS out there manages default aliases just fine, and Mercurial even allows overriding commands. How do you explain that the world hasn't ended for them?

-- 
Felipe Contreras