Re: [RFC] [PATCH]: use explicit enums for CCID 3 states

[Eddie Kohler <kohler@cs.ucla.edu>]

Sure, straight enums use more space, but it's possible to use an enum 
bitfield.  That would get the best of both worlds.

Eddie


Arnaldo Carvalho de Melo wrote:
> On 11/15/06, Gerrit Renker <gerrit@erg.abdn.ac.uk> wrote:
>> This patch tackles the following problem:
>>         * the ccid3_hc_{t,r}x_sock define ccid3hc{t,r}x_state as `u8', 
>> but
>>           in reality there can only be a few, pre-defined enum names
>>         * this necessitates addiditional checking for unexpected values
>>           which would otherwise be caught by the compiler
> 
> My problem is with the changes to the structs, from u8 to enums, that
> take 4 bytes, <shameless plug>see using my new tool, codiff (code
> diff):
> 
> [acme@newtoy net-2.6.20]$ codiff --structs --verbose
> /tmp/ccid3.o.before /tmp/ccid3.o.enum
> /pub/scm/linux/kernel/git/acme/net-2.6.20/net/dccp/ccids/ccid3.c:
>  struct ccid3_hc_tx_sock       |   +4
>    ccid3hctx_state;
>     from: u8                      /*    30(0)     1(0) */
>     to:   enum ccid3_hc_tx_states /*    32(0)     4(0) */
>  struct ccid3_hc_rx_sock       |   +4
>    ccid3hcrx_state:8;
>     from: u64                     /*    16(36)    8(8) */
>     to:   enum ccid3_hc_rx_states /*    20(0)     4(0) */
> 2 structs changed
> [acme@newtoy net-2.6.20]$
> 
> </shameless plug>
> 
> I removed the other parts of the patch (replacing printks+dump_stack
> with BUG) to see what would be the code impact of such change and was
> surprised:
> 
> [acme@newtoy net-2.6.20]$ codiff --functions /tmp/ccid3.o.before
> /tmp/ccid3.o.enum
> /pub/scm/linux/kernel/git/acme/net-2.6.20/net/dccp/ccids/ccid3.c:
>  ccid3_hc_rx_set_state         |  -33
>  ccid3_hc_tx_set_state         |   -5
>  ccid3_hc_rx_send_feedback     |  -16
>  ccid3_hc_rx_init              |   -8
>  ccid3_hc_tx_init              |   +3
>  ccid3_hc_tx_no_feedback_timer |  -14
>  ccid3_hc_rx_packet_recv       |  -48
>  ccid3_hc_rx_get_info          |   -5
> 8 functions changed, 3 bytes added, 129 bytes removed
> [acme@newtoy net-2.6.20]$
> 
> Anyway, there are always tradeoffs, compiler gets more info, but we
> end up using more memory per socket instance, I guess we can stay like
> we are now, no? :-)
> 
> - Arnaldo
> 
> Probably because we stop using a u64 bitfield in the rx case...
> -
> To unsubscribe from this list: send the line "unsubscribe dccp" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html