Re: [PATCH] Change "refs/" references to symbolic constants

From: Andy Parkins <andyparkins@gmail.com>
Date: Feb 20, 2007, 10:57 UTC
Message-ID: <200702201057.21398.andyparkins@gmail.com>
In-Reply-To: <7vabz9w270.fsf@assigned-by-dhcp.cox.net>

On Tuesday 2007 February 20 10:30, Junio C Hamano wrote:

Show 6 quoted lines

> If we _were_ doing the inline function, I would actually prefer:
>
>         static inline ref_is_head(const char *ref)
>         {
> 		return !prefixcmp(ref, PATH_REFS_HEADS);
>         }

As you brought it up...

I've never really liked "!" on strcmp() lines (but I accept that that is the tradition in git) because it implies the the output of prefixcmp is boolean, but it's actually ternary. strcmp() (I think), should be thought of as outputting

enum {
 STRING1_LESS_THAN_STRING2,
 STRINGS_EQUAL,
 STRING1_GREATER_THAN_STRING2
}

Given that, it makes me uncomfortable to use !strcmp(). Of course in the case of strcmp(), that form is so well known that it makes very little difference to the reader.

I have similar feelings about

 if( !something )

being incorrect when you meant

 if( something == NULL )

While they are identical in what they generate, they send a different message to someone reading the code.

Regardless, I'm not so stubborn as to refuse to go with the flow...

Show 9 quoted lines

> But at least to me,
>
> 	if (!prefixcmp(head, PATH_REFS_HEADS))
> 		head += strlen(PATH_REFS_HEADS);
>
> is easier to follow than:
>
>         if (ref_is_head(head))
>                 head += STRLEN_PATH_REFS_HEADS;

Fine.  I don't really mind - and it's less work on my patch :-)

My argument in favour of the ref_is_head() method is that the prefixcmp() method requires knowledge from the caller about how you tell whether a given ref is a head - the second pushes that information further down the call tree, abstracting it out just a little more.

As I say though - it's not a problem for me.

Andy

-- 
Dr Andy Parkins, M Eng (hons), MIEE
andyparkins@gmail.com