Re: [PATCH] rev-parse: fix --parse-opt --keep-dashdash --stop-at-non-option

["Uwe Kleine-König" <u.kleine-koenig@pengutronix.de>]

Hi Pierre,

On Thu, Jul 08, 2010 at 12:18:48PM +0200, Pierre Habouzit wrote:
> On Thu, Jul 08, 2010 at 09:26:23AM +0200, Uwe Kleine-König wrote:
> > On Wed, Jul 07, 2010 at 02:41:33PM -0700, Junio C Hamano wrote:
> > > Uwe Kleine-König  <u.kleine-koenig@pengutronix.de> writes:
> > > 
> > > > The ?: operator has a lower priority than |, so the implicit associativity
> > > > made the 6th argument of parse_options be PARSE_OPT_KEEP_DASHDASH if
> > > > keep_dashdash was true discarding PARSE_OPT_STOP_AT_NON_OPTION and
> > > > PARSE_OPT_SHELL_EVAL.
> > > 
> > > Wow, this is an age-old breakage dating back to 6e0800e (parse-opt: make
> > > PARSE_OPT_STOP_AT_NON_OPTION available to git rev-parse, 2009-06-14) that
> > > dates back to the very original --stop-at-non-option patch, isn't it?
> > I made a quick C-quiz at my company asking what's wrong with 6e0800e.
> > 
> > Apart from the bug fixed in my patch a colleague wondered about
> > stop_at_non_option being static.  I think it doesn't do any harm, still
> > I think being an automatic variable would be more common.  Is the static
> > intended here?  This was introduced in
> > 21d4783538662143ef52ed6967c948ab27586232, so I cc:d Pierre.
> 
> Well, the sole difference is that it makes &stop_at_non_option been
> computed at compile time instead of runtime, which is pretty much the
> same. cmd_parseopt isn't meant to be reentrant so it's not important.
I don't know about x86, but I think on arm computing at compile time
isn't cheaper than at runtime, it's just pc-relative instead of
sp-relative.  But having the variable automatic saves a bit of heap.
Probably not worth to discuss about these two ints though.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |