threads / discuss / 41894

GIT_CONFIG - what's the point?

Subject: GIT_CONFIG - what's the point?

## tl;dr

7 messages between Apr 1, 2016 and Apr 3, 2016.

replies: 6people: 4as markdown or json

Matthew Persico· Apr 1, 2016, 00:54 UTC · lore
Greetings.

Given the GIT_CONFIG environment variable can change 'git config' behaves, it stands to reason that if GIT_CONFIG is defined, then ALL git commands obey the value of GIT_CONFIG and use that file for config info.

As a test, exported GIT_CONFIG=/tmp/ohm, copied ~/.gitconfig to /tmp/ohm, moved ~/.gitconfig to ~/.gitconfig.hold and then tried git st, where 'st' is an alias for status in my config file.

No dice. git st was unrecognized.

So, what's the point of GIT_CONFIG if only git-config uses it? Or did I miss a step?

Thanks
-- 
Matthew O. Persico
Christian Couder· Apr 1, 2016, 11:19 UTC · re: Matthew Persico · lore

Re: GIT_CONFIG - what's the point?

On Fri, Apr 1, 2016 at 2:54 AM, Matthew Persico <matthew.persico@gmail.com> wrote:

Show 9 quoted lines
> Greetings.
>
> Given the GIT_CONFIG environment variable can change 'git config'
> behaves, it stands to reason that if GIT_CONFIG is defined, then ALL
> git commands obey the value of GIT_CONFIG and use that file for config
> info.
>
> As a test, exported GIT_CONFIG=/tmp/ohm, copied ~/.gitconfig to
> /tmp/ohm,

Is /tmp/ohm a directory? If that is the case, then you should probably have exported "GIT_CONFIG=/tmp/ohm/.gitconfig", as the git config doc says it specifies a filename.

Jeff King· Apr 1, 2016, 12:38 UTC · re: Matthew Persico · lore

Re: GIT_CONFIG - what's the point?

On Thu, Mar 31, 2016 at 08:54:26PM -0400, Matthew Persico wrote:
> So, what's the point of GIT_CONFIG if only git-config uses it? Or did
> I miss a step?

There isn't a point to it. It's historical cruft that has been left in to avoid breaking older scripts. The same thing is generally better accomplished by using git-config's "--file" parameter. We should probably do a better job of making that clear in the documentation.

Or possibly deprecate it and eventually remove it entirely, as discussed in:

  http://thread.gmane.org/gmane.linux.debian.devel.bugs.general/1195694/focus=257770
-Peff
Matthew Persico· Apr 1, 2016, 14:31 UTC · re: Jeff King · lore

Re: GIT_CONFIG - what's the point?

Let me explain my scenario. I have an nfs mounted home directory. It is used across multiple machines. I use different colored xterms for each machine. But that means that the one set of colors in my one .gitconfig file don't work against all my screen backgrounds. I'm trying to find a way to tune the git colors per login. The ability to set colors in an environment variable (like most UNIX utils support) would be the easiest way to do this. Failing that, I was hoping that by setting GIT_CONFIG per login, I could tune the color schemes with different config files.

Since that is not how GIT_CONFIG is used, I have simply decided to squint where necessary, or open up a neutral colored xterm for the diff, regardless of machine.

Yes, I could probably do diffs in many other ways, but git diff at the command line is usually the most expedient.

Unless I wanted to define a GIT_CONFIG_OVER environment variable upon login, place inside it the appropriate -c<name>=<value> overrides for colors, and then define a bash function git as

git () {
   $(which git) $GIT_CONFIG_OVER "$@"
   return $?
}
which seems silly.
Thanks anyway.
On Fri, Apr 1, 2016 at 8:38 AM, Jeff King <peff@peff.net> wrote:
Show 16 quoted lines
> On Thu, Mar 31, 2016 at 08:54:26PM -0400, Matthew Persico wrote:
>
>> So, what's the point of GIT_CONFIG if only git-config uses it? Or did
>> I miss a step?
>
> There isn't a point to it. It's historical cruft that has been left in
> to avoid breaking older scripts. The same thing is generally better
> accomplished by using git-config's "--file" parameter. We should
> probably do a better job of making that clear in the documentation.
>
> Or possibly deprecate it and eventually remove it entirely, as discussed
> in:
>
>   http://thread.gmane.org/gmane.linux.debian.devel.bugs.general/1195694/focus=257770
>
> -Peff
-- 
Matthew O. Persico
Jeff King· Apr 1, 2016, 14:53 UTC · re: Matthew Persico · lore

Re: GIT_CONFIG - what's the point?

On Fri, Apr 01, 2016 at 10:31:12AM -0400, Matthew Persico wrote:
Show 16 quoted lines
> Let me explain my scenario. I have an nfs mounted home directory. It
> is used across multiple machines. I use different colored xterms for
> each machine. But that means that the one set of colors in my one
> .gitconfig file don't work against all my screen backgrounds. I'm
> trying to find a way to tune the git colors per login. The ability to
> set colors in an environment variable (like most UNIX utils support)
> would be the easiest way to do this. Failing that, I was hoping that
> by setting GIT_CONFIG per login, I could tune the color schemes with
> different config files.
> 
> Since that is not how GIT_CONFIG is used, I have simply decided to
> squint where necessary, or open up a neutral colored xterm for the
> diff, regardless of machine.
> 
> Yes, I could probably do diffs in many other ways, but git diff at the
> command line is usually the most expedient.

I think per-environment config is a reasonable thing to want. I can think of three approaches with varying drawbacks:

  1. The cleanest thing would be for git's config-include directive to
     respect environment variables somehow. We do expand $HOME in
       [include]
       path = ~/.whatever
     but only $HOME (and if you're willing to set $HOME, you can just
     point to a different ~/.gitconfig in the first place).
     So the drawback is that it doesn't currently work, and would need a
     patch to git. ;)
  2. If you don't want to change $HOME, you might be OK with changing
     $XDG_CONFIG_HOME, which could then point to your machine-specific
     user-level config (and could use an include to pull in the common
     bits). See "git help config" for more details (particularly, I
     think the XDG source only kicks in if you have no ~/.gitconfig, but
     I might be misremembering the details).
     The drawback is that other things besides git use $XDG_CONFIG_HOME,
     so you might be affecting them.
     You could probably come up with some elaborate scheme where each
     host has its own $XDG_CONFIG_HOME, and you symlink in the common
     bits. Or something. But that gets complicated pretty fast.
  3. The "git -c" mechanism communicates config to sub-processes via the
     $GIT_CONFIG_PARAMETERS variable. Its format and even existence are
     undocumented, but something like this would probably do what you
     want:
       export GIT_CONFIG_PARAMETERS="'config1=value1' 'config2=value2'"
     The drawback is that it is definitely not officially supported.
     However, I don't think there any plans to get rid of it (and if
     anything, I'd suspect in the future the parser may get less picky,
     and it might become an officially supported interface; but don't
     quote me on that).
-Peff
SZEDER Gábor· Apr 1, 2016, 17:28 UTC · re: Matthew Persico · lore

Re: GIT_CONFIG - what's the point?

Show 27 quoted lines
> Let me explain my scenario. I have an nfs mounted home directory. It
> is used across multiple machines. I use different colored xterms for
> each machine. But that means that the one set of colors in my one
> .gitconfig file don't work against all my screen backgrounds. I'm
> trying to find a way to tune the git colors per login. The ability to
> set colors in an environment variable (like most UNIX utils support)
> would be the easiest way to do this. Failing that, I was hoping that
> by setting GIT_CONFIG per login, I could tune the color schemes with
> different config files.
> 
> Since that is not how GIT_CONFIG is used, I have simply decided to
> squint where necessary, or open up a neutral colored xterm for the
> diff, regardless of machine.
> 
> Yes, I could probably do diffs in many other ways, but git diff at the
> command line is usually the most expedient.
> 
> Unless I wanted to define a GIT_CONFIG_OVER environment variable upon
> login, place inside it the appropriate -c<name>=<value> overrides for
> colors, and then define a bash function git as
> 
> git () {
>    $(which git) $GIT_CONFIG_OVER "$@"
>    return $?
> }
> 
> which seems silly.

Yeah, that 'return $?' at the end of the function does indeed seem silly :) (sorry, couldn't resist...)

You could use machine-specific config includes instead of that GIT_CONFIG_OVER environment variable. I.e. store machine-specific color configuration in ~/.gitcolors.<machine> or something and define the shell function as:

git () {
        command git -c include.path=~/.gitcolors.$HOSTNAME "$@"
}

The impact on your .bashrc would be much smaller than with the GIT_CONFIG_OVER approach. You could even turn this into an alias, if you want.

Matthew Persico· Apr 3, 2016, 20:11 UTC · re: SZEDER Gábor · lore

Re: GIT_CONFIG - what's the point?

On Fri, Apr 1, 2016 at 1:28 PM, SZEDER Gábor <szeder@ira.uka.de> wrote:
Show 30 quoted lines
>> Let me explain my scenario. I have an nfs mounted home directory. It
>> is used across multiple machines. I use different colored xterms for
>> each machine. But that means that the one set of colors in my one
>> .gitconfig file don't work against all my screen backgrounds. I'm
>> trying to find a way to tune the git colors per login. The ability to
>> set colors in an environment variable (like most UNIX utils support)
>> would be the easiest way to do this. Failing that, I was hoping that
>> by setting GIT_CONFIG per login, I could tune the color schemes with
>> different config files.
>>
>> Since that is not how GIT_CONFIG is used, I have simply decided to
>> squint where necessary, or open up a neutral colored xterm for the
>> diff, regardless of machine.
>>
>> Yes, I could probably do diffs in many other ways, but git diff at the
>> command line is usually the most expedient.
>>
>> Unless I wanted to define a GIT_CONFIG_OVER environment variable upon
>> login, place inside it the appropriate -c<name>=<value> overrides for
>> colors, and then define a bash function git as
>>
>> git () {
>>    $(which git) $GIT_CONFIG_OVER "$@"
>>    return $?
>> }
>>
>> which seems silly.
>
> Yeah, that 'return $?' at the end of the function does indeed seem
> silly :)  (sorry, couldn't resist...)
Part OCD, part OAC. :-)
Show 9 quoted lines
>
> You could use machine-specific config includes instead of that
> GIT_CONFIG_OVER environment variable.  I.e. store machine-specific
> color configuration in ~/.gitcolors.<machine> or something and define
> the shell function as:
>
> git () {
>         command git -c include.path=~/.gitcolors.$HOSTNAME "$@"
> }

BINGO! THAT was the redirection I needed! One thing I was trying to figure out early one was how to put HOSTNAME-based include.path-s in the .gitconfig. So I put them OUTSIDE the .gitconfig like this.

Much obliged. Next time you are in NYC - I owe you a beer!
Show 6 quoted lines
>
> The impact on your .bashrc would be much smaller than with the
> GIT_CONFIG_OVER approach.
> You could even turn this into an alias, if you want.
>
>
-- 
Matthew O. Persico

← back to recent threads