[ovs-dev] [ext-244 3/4] Remove assumption that there are 64 or fewer fields.

[Jarno Rajahalme]

On Jul 26, 2014, at 10:41 AM, Jarno Rajahalme <jrajahalme at nicira.com> wrote:

> 
> 
> Sent from my iPhone
> 
>> On Jul 26, 2014, at 10:37 AM, Ben Pfaff <blp at nicira.com> wrote:
>> 
>> On Sat, Jul 26, 2014 at 08:47:56AM -0700, Jarno Rajahalme wrote:
>>>> +/* A set of mf_field_ids. */
>>>> +struct mf_bitmap {
>>>> +    unsigned long bm[BITMAP_N_LONGS(MFF_N_IDS)];
>>>> +};
>>>> +#define MF_BITMAP_INITIALIZER { { [0] = 0 } }
>>> 
>>> Is this different from just { { 0 } } ?
>> 
>> It has the same semantics, but it suppresses GCC warnings about
>> only partially initializing an array.
>> 
>>>> @@ -270,9 +276,9 @@ struct mf_field {
>>>>    * Also, some field types are tranparently mapped to each other via the
>>>>    * struct flow (like vlan and dscp/tos fields), so each variant supports
>>>>    * all protocols. */
>>>> -    enum ofputil_protocol usable_protocols; /* If fully/cidr masked. */
>>>> +    uint32_t usable_protocols; /* If fully/cidr masked. */
>>>>   /* If partially/non-cidr masked. */
>>>> -    enum ofputil_protocol usable_protocols_bitwise;
>>>> +    uint32_t usable_protocols_bitwise;
>>> 
>>> These seem unrelated changes?
>> 
>> This is actually a sticky point.  Before this commit, meta-flow.h
>> includes ofp-util.h, to get "enum ofputil_protocol" for these two
>> members.  After this commit, ofp-util.h includes meta-flow.h, to get
>> struct mf_bitmap.  The circular dependency causes a problem.  The best
>> solution I came up with was to change the "enum ofputil_protocol"
>> members to uint32_t.  It's not a great solution.  Do you have a good
>> idea?
> 
> Can the enum be forward declared here instead?

It seems C does not allow that, even if GCC does :-(

  Jarno