Re: Alphabetize git config --list

From: Jeff King <peff@peff.net>
Date: May 29, 2012, 08:49 UTC
Message-ID: <20120529084940.GA15145@sigill.intra.peff.net>
In-Reply-To: <4FC45CEB.4070808@alum.mit.edu>

On Tue, May 29, 2012 at 07:21:47AM +0200, Michael Haggerty wrote:

Show 11 quoted lines

> >>Essentially this
> >>$ git config --list | sort
> >
> >No, it's not a good idea; the order of keys is important. For example,
> >[...]
> 
> Is the order of *keys* important, or only the order of *values* for a
> given key?  I believe that one could sort the output by key as long
> as the values for each key are kept in their original order (i.e.,
> sort stably by key).  This would indeed make it easier to read the
> output.

Good question. I was thinking of the order of values, but you're right that a stable key-only sort would handle that.

In general, keys are independent, but there are two exceptions I can think of:

  1. We sometimes have aliased keys that get parsed to the same memory
     location (usually this is to maintain backwards compatibility when
     a key is renamed, or wants to grow incompatible features). For
     example, "diff.color" and "color.diff" override each other, and
     their order is important. Also, "diff.funcname" and "diff.xfuncname".
     There may also be others.

  2. Keys which involve pattern matching. The url.*.insteadOf keys seem
     to be order-independent, as they the longest match is used, not the
     first. The credential context-matching code takes all matches
     (though technically, things like the order in which helpers are
     called would be changed by a sort). I can't think of any others
     offhand.

So yes, I think the key order does matter, though only in a few corner cases.

-Peff