Re: [ULOGD PATCH 2/5] New MARK filtering module.

[Patrick McHardy <kaber@trash.net>]

Eric Leblond wrote:
> +static struct config_keyset libulog_kset = {
> +	.num_ces = 2,
> +	.ces = {
> +		{
> +			.key 	 = "mark",
> +			.type 	 = CONFIG_TYPE_INT,
> +			.options = CONFIG_OPT_NONE,
> +			.u.value = 0,
> +		},
> +		{
> +			.key 	 = "mask",
> +			.type 	 = CONFIG_TYPE_INT,
> +			.options = CONFIG_OPT_NONE,
> +			.u.value = 0xffffffff,
> +		},
> +
> +	}
> +};
> +	
> +#define mark_ce(x)	(x->ces[0])
> +#define mask_ce(x)	(x->ces[1])


This magic value stuff is an error-prone mess in my opinion.
Could we for future modules agree to always do something like
this:

enum foo_keys {
	FOO_MARK,
	FOO_MASK,
	...
};

struct config_keyset foo_kset = {
...
	.ces = {
		[FOO_MARK] = {
			...
		},
	},
};

Or maybe resolve them at runtime using the key string in
performance uncritical paths?