Hello!
Here's a snip of an IRC conversation between Darren Reed (Avalon) and myself
(danmcd) that's related to this thread on detangle.
Dan
===================== (Cut up to and including here.) =====================
<Avalon> dan - why are you so caught up on using obscure programming tricks
for microoptimisations of the code?
<danmcd> Avalon - Too much assembly as a child, but that's not the point of
detangle.
<danmcd> Avalon - And I took Jeremy's advice, the != thing is far more
readable.
<danmcd> I just didn't update the opensolaris.org webrev with it.
<Avalon> dan - the bigger question I have re. the detangle is should there be
a better architected interface through which this is achieved?
<danmcd> There might have been --> rewriting connp->conn_recv() on a
UDP_NAT_T_ENDPOINT socket
<danmcd> to point to a zero-SPI-stripper, but IP does some of the other
options processing prior to calling
<danmcd> connp->conn_recv() so I have to place that functionality in IP.
<danmcd> Either that, or I attempt to refactor UDP/IP interfacing to move the
appropriate bits out of IP and
<danmcd> into UDP.
<Avalon> That's probably out of scope for what you're doing...
<danmcd> Agreed.
<danmcd> It's a damned shame, because I was all psyched about Yosemite's use
of the conn_t structure, but
<danmcd> then I couldn't get it to bend to my will.
<Avalon> Have you blogged about this?
<danmcd> About why UDP/IP interactions is broken?
<Avalon> it's worth recording these design/implementation comments somewhere..
<danmcd> Or about detangle?
<Avalon> all of the above.
<danmcd> I was going to blog about detangle closer to putback.
<Avalon> ok.
<danmcd> As for the UDP thing, maybe we need to have this exchange on
networking-discuss?
<danmcd> THat's a public record!
<danmcd> We could even be lazy and cut/paste this IRC chat there! :)
<Avalon> hahahaha
<danmcd> It has precedent...
<Avalon> just so long as someone doesn't come out and say "when are we making
sun's irc network open?"
<danmcd> HA!
_______________________________________________
networking-discuss mailing list
[email protected]
