Re: [lwip-devel] byte order, packing, optimizations

[address@hidden]

Stéphane Lesage wrote:
> Yeah that's more elegant when having 1 byte packing...
> The problem is that people who optimize will use 2 bytes or no packing,
> and it won't work for them.
>    
Oh, you're right, these structs aren't aligned, I didn't think of 
that... However, that's an example of what can go wrong when 
"optimizing" and using pack(2): you'd need some kind check (e.g. 
LWIP_ASSERT for alignment) to make sure that future changes don't break 
the alignment assumptions.
> I was preparing another patch for
> - combine load/stores and ntohX/htonX (lwip packing)
>    
Could you explain that? I'd like to make sure it's OK before you 
actually do the work to prepare a patch.
> - other macros for reading/writing unaligned words for such protocols
>    
Hmm, I'm not sure I understand you correctly, but we already discussed 
such macros (for processors/compilers that can't do struct packing) and 
decided we don't want to support this...
>   >  >  - change ip_addr_is_broadcast to a more efficient macro
>   >  >
>   >  That change actually changes the behaviour of ip_addr_is_broadcast().
>   >  Was that intended? I don't think the change is correct, or is it?
>
> I removed:
> - the broadcast with 0.0.0.0 "old style comment", is it really used ?
>    
It is! DHCP/Bootp is allowed to use this.
> - comparison to netif address, useful only when mask=255.255.255.255 ->
> the netif should not support broadcast
>    
Don't know about that...

Simon