Re: [PATCH 07/10] color: delay auto-color decision until point of use

> Jeff King <peff@peff.net> writes:
> 
> > diff --git a/color.h b/color.h
> > index a190a25..d715fd5 100644
> > --- a/color.h
> > +++ b/color.h
> > @@ -49,6 +49,16 @@ struct strbuf;
> >  #define GIT_COLOR_NIL "NIL"
> >  
> >  /*
> > + * The first three are chosen to match common usage in the code, and what is
> > + * returned from git_config_colorbool. The "auto" value can be returned from
> > + * config_colorbool, and will be converted by want_color() into either 0 or 1.
> > + */
> > +#define GIT_COLOR_UNKNOWN -1
> > +#define GIT_COLOR_ALWAYS 0
> > +#define GIT_COLOR_NEVER  1
> > +#define GIT_COLOR_AUTO   2
> 
> The ALWAYS/NEVER somehow go against my intuition. Let me trace one
> codepath starting from git_branch_config().
> 
>     branch_use_color is set from git_config_colorbool("color.branch");
>     -> given "never", git_config_colorbool() returns 0;
>     branch_get_color() asks want_color(branch_use_color);
>     -> want_color() returns if the given value is positive.
> 
> Because git_config_colorbool() does not use the above symbolic constants,
> everything goes well, but aren't these two swapped?