Re: [PATCHv2 2/4] Implement loss counting on TFRC-SP receiver

[Leandro Sales <leandroal@gmail.com>]

Hi Gerrit,

On Mon, Oct 19, 2009 at 2:26 AM, Gerrit Renker <gerrit@erg.abdn.ac.uk> wrote:
>
> | --- dccp_tree_work03.orig/net/dccp/ccids/lib/packet_history_sp.c      2009-10-08 22:58:21.418908270 -0300
> | +++ dccp_tree_work03/net/dccp/ccids/lib/packet_history_sp.c   2009-10-08 22:59:07.442411383 -0300
> | @@ -243,6 +243,7 @@
> |  {
> |       u64 s0 = tfrc_rx_hist_loss_prev(h)->tfrchrx_seqno,
> |           s1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_seqno,
> | +         n1 = tfrc_rx_hist_entry(h, 1)->tfrchrx_ndp,
> |           s2 = tfrc_rx_hist_entry(h, 2)->tfrchrx_seqno,
> |           s3 = DCCP_SKB_CB(skb)->dccpd_seq;
> I have removed the old definition of n1, which was further below and which caused this warning.
>
> net/dccp/ccids/lib/packet_history_sp.c:276:7: warning: symbol 'n1' shadows an earlier
> net/dccp/ccids/lib/packet_history_sp.c:247:6: originally declared here
>
>

Well done!

> I thought again about the earlier suggestion to make 'num_losses' u64. Since li_losses sums the values
> stored in num_losses, it needs to have the same size (currently it is u32). But then another thought is
> that if there are so many losses that u32 overflows, then the performance is so bad anyway that it is
> better to turn off the receiver. Hence I have reverted it to u32, as per your original patch.
>

OK

> Please find attached a patch of the changes I made. As per posting, I have separated out the dccp.h part,
> since it is also useful in general.

OK, agreed!

Thank you,

BR,
Leandro.