Re: [PATCH v2 1/4] i18n: factorize more 'incompatible options' messages

[Johannes Sixt <j6t@kdbg.org>]

Am 29.01.22 um 00:58 schrieb Junio C Hamano:
> Johannes Sixt <j6t@kdbg.org> writes:
>> A bit of code duplication could be avoided if die_if_incompatible_opt3()
>> forwarded with an additional pair 0, "" to die_if_incompatible_opt4().
> 
> I wondered if a single varargs function
> 
>     void die_if_incompatible_optN(const char *name1, int opt1, ...);
> 
> that takes a name=NULL terminated sequence of <name, opt> would
> work, but
> 
>  (1) it would require the caller to always spell out the terminating
>      NULL, which may be ugly.
> 
>  (2) it would tempt people to programatically generate message for N
>      options, which leads to sentence lego, which is the exact
>      opposite of what this series wants to achieve.

The reason I did not suggest a varargs version is because

  (3) it is not typesafe.

A varargs implementation could be used as an implementation helper, but
should not be a public interface.

> In any case, I do agree with you that the callers of _opt3()
> variants can just pass (0, "unused") in the 4-th slot.  If this were
> made varargs, then it only needs to pass NULL at the end, so that
> might be an improvement, though.
> 
> Also, isn't "if" in the name of the function misleading?  as far as
> I can tell, this function is not "check if these options are
> compatible and die if some are incompatible with each other", but
> the caller is convinced that these options are incompatible before
> it decides to call this function and there is no "if" in what this
> function can do.

Good point.

> void die_for_incompatible_opts(const char *name1, int opt1, ...)
> {
> 	const char *name[4];
> 	int count = 0;
>         va_list params;
> 
>         va_start(params, name1);
> 
>         if (opt1)
> 	        name[count++] = name1;
>         while (count < ARRAY_SIZE(name)) {
>                 const char *n = va_arg(params, const char *);
> 
>                 if (!n)
> 			break;
>                 if (va_arg(params, int))
> 	                name[count++] = n;
>         }
>         va_end(params);
> 
>         switch (count) {
> 	default:
> 		BUG("die-for-incompatible-opts can only take up to %d args",
> 		    ARRAY_SIZE(name));

The problems here are: (1) this case triggers only if there is an actual
invocation with all 5+ incompatible options, and (2) at that time, the
out-of-bounds write to the name array has already happened.

I know this implementation is just a rough show-case. But since it's
been written by a very proficient mind, I'm now convinced: it's not
sufficiently easy to write the varargs version, so let's not go there.

> 	case 4:
> 		die(_("options '%s', '%s', '%s', and '%s'"
> 		      " cannot be used together"),
> 		    name[0], name[1], name[2], name[3]);
> 		break;
> 	case 3:
> 		die(_("options '%s', '%s', and '%s'"
> 		      " cannot be used together"),
> 		    name[0], name[1], name[2]);
> 		break;
> 	...
> 	}
> }
> 
> might be easier to extend when somebody discovers there needs the
> "opt5" variant. 

-- Hannes