Re: [PATCH] Adds 'stash.index' configuration option

From: Michael J Gruber <git@drmicha.warpmail.net>
Date: May 12, 2011, 08:14 UTC
Message-ID: <4DCB96F9.2020700@drmicha.warpmail.net>
In-Reply-To: <20110512080425.GA11870@sigill.intra.peff.net>

Jeff King venit, vidit, dixit 12.05.2011 10:04:

Show 37 quoted lines

> On Thu, May 12, 2011 at 09:14:09AM +0200, Michael J Gruber wrote:
> 
>> This is yet another incarnation of
>>
>> foo.bar = true
>>
>> meaning that command "git foo" defaults to "git foo --bar". (Admittedly,
>> this is about subcommands of foo.)
>>
>> It has the same problems (possibly breaking scripts). But more
>> importantly, it inflates the code with every such incarnation we add.
>> Have we really agreed that we introduce these one-by-one rather than
>> doing something generic like
>>
>> uiopts.<cmd> = <optionlist>
>>
>> with which you would do
>>
>> uiopts.stash = "--index"
>>
>> and hopefully be script-safe (again, ignoring the subcommand issue)?
> 
> I would love to see something like this, but have we yet figured out all
> of the issues, like:
> 
>   1. How do scripts wanting to call git programs suppress expansion of
>      uiopts when they want predictable behavior?
> 
>   2. Depending on the solution to (1), how do scripts specify that they
>      _do_ want to allow uiopts (e.g., because they know they are
>      presenting the output to the user) for certain commands?
> 
>   3. Depending on (1) and (2), how do scripts differentiate when some
>      options are OK in uiopts, but others are not? For example, it may
>      be desirable for an invocation of diff-tree to have renames turned
>      on by the user, but not for them to change the output format.
>

We haven't figured that out, but was the consensus: "Whatever, let's just keep adding single options." ?

> As much as it sucks to have a config option for each individual option,
> there is at least some oversight of which options will not cause too
> much of a problem when triggered automatically.

I just think we have too many commands which are ui and are used in scripts (e.g. log, commit, stash, just to name a few) for being able to decide that ourselves. Are we saying that people using "git stash" in a script have to deal themselves with a breakage caused by "--index" being a default for some users now?

With a generic approach, we could protect all git-sh-setup using scripts right from the start, for example, while still allowing to override some options or to protect only a few (based on the explicit wishes of a uiopts-aware script).

Michael