Re: [PATCH 00/19] Add git-list-files

From: Jeff King <peff@peff.net>
Date: Dec 2, 2014, 20:18 UTC
Message-ID: <20141202201837.GC23461@peff.net>
In-Reply-To: <CACsJy8DBkGVhaJnCTs9_E+g6FYYr4V-6S=XB5wrGBFPjHnEu1A@mail.gmail.com>

On Tue, Dec 02, 2014 at 06:45:52PM +0700, Duy Nguyen wrote:

Show 6 quoted lines

> > As a side note, I wonder if it would be sensible to whitelist some
> > commands as porcelain, and allow aliases to override them (either
> > entirely, or just to add-in some options).
> 
> Agreed. Maybe not all porcelain (some like git-branch almost functions
> like plumbing).

Yeah, many things straddle the plumbing/porcelain line (e.g., "commit" is porcelain, but it's basically the only sane way for scripts to make a commit, so many use it). So I'd pick just a few things that should be safe to override.

> We also need away to stop alias (e.g. in scripts).

Do we? I think the point of allowing this only for porcelain is that you do not have to care about scripts. That is, a script running "git ls" would get whatever the user's preferences are for "ls" output. A script parsing the output of "ls" deserves whatever crap it gets.

Show 5 quoted lines

> In scripts I can specify full path to a command to make sure I won't
> hit an alias. I guess we can't do the same here. The closet to "full
> path" is git-cmd form, as opposed to "git cmd" form) but I think we
> don't want to bring back git-cmd. Maybe just a "git --no-alias cmd"
> and GIT_NO_ALIAS..

Yeah, I think "--no-alias"/GIT_NO_ALIAS would work. But the problem is one of compatibility. Scripts are not written to specify no-alias, so you cannot just turn on the override-commands-with-aliases feature immediately (and likewise, scripts have little incentive to bother annotating their calls if it the override feature is not enabled).

-Peff