#haskell - Wed 11 Apr 2007 between 00:00 and 00:06

NY Lost Funds



dibblegodcoutts, I agree entirely
sjanssenI don't understand what these potential problems are
locking inside the handle?
procyon112Especially if correctness proofs can be done at compile time, and abstracted away at runtime.
dibblegoI think Backus said it most succintly
donsoh, i have to subscribe to my own list.
monochromThe problem with most programmers is they code harder not think harder.
sjanssens/problems/performance problems
dibblegomonochrom, I've been through three keyboards this year (perhaps I should think more instead?)
dcouttssjanssen: as far as I know there are no performance problems with Handles if you use them right.
dibblegogranted, they are MS keyboards with very fimsy keys
*flimsy
monochromI guess you don't want to be through three brains next year :)
dibblegoI hope my brain's enter key is not manufactured by Microsoft
dcouttssjanssen: who says there are potential problems? I must have missed that part of the conversation
sjanssendcoutts: whatever Bulat complains about
sorearI never realized Bulat was complaining about performance.
dcouttssjanssen: well, I'm not sure I understand what he thinks is bad, all I remember is that he's wrong :-)
sorearI thought he was complaining about IO being too big for base.
donsok, welcome msg sent.
sorearand I agree on that point :)
donsdid everyone get that?
actionsorear subs
sorearsubs
jcreighdons: yes
dcouttssjanssen: sorear: right he certainly complains about the Handle stuff being monolithic
donssjanssen: isn't subscribed yet.
sjanssensorear: search archives for "high performance io lib", I'm almost certain you'll get dozens of Bulat hits
dcouttssjanssen: sorear: but as I understand it he also thinks there are higher speed OS-native 'zero-copy' things that we're not using and the Handle is too inflexible to take advantage of.
sorearhahaha.
donssecret syscalls!
sorearno, not secret
about as well documented as +RTS -xc
jcreigh"zero copy"?
dcouttssjanssen: sorear: he's half right. Zero copy is nearly worthless but it might be fun to try async IO rather than the current non-blocking IO, but I expect the gains to be very marginal.
sorearIE they are well documented; nobody but me seems to have ever bothered to read the relevant docs though
dcouttsjcreigh: mmap and similar techniques
donswell, we do have a full epoll binding in the 'unify' lib.
hmm, that might be fun to cabalise ..
should win award for best code not yet in a cabal repo
dcouttsepoll is easy to bind, it's just 4 functions

Page: 2 9 16 23 30 37 44 51 58 65 72 79 

IrcArchive

NY Lost Funds