#python - Sat 21 Apr 2007 between 00:05 and 00:13

NY Lost Funds



koshPeaker: the more stuff you stick all together the more dangerous those bit flips are, the more things are seperated the more resilient the system is
peakerTFKyle: if its never seen
koshaddress space seperation is a useful feature and I don't see how threads really buy you that much to be worth its loss
actionTFKyle doesn't like that too much
TFKyledoesn't like that too much
peakerkosh: What if the bit flip occurs in the kernel, or any shared code of apache?
kosh: its silly -- bit flips are not an argument
koshI just came up with something
peakerkosh: you know that all the processes share the text segment
kosh: and a bit flip there would crash them all
koshheck it could be other things outside what you can figure out
TFKylehaving a bunch of broken possibilities in the threading library and not fixing them just because "the most popular way to make apps doesn't do that"
koshprocesses are safer and scale better
peakerwell - what you really want is separation. Whether its done via complete address space separation - or by semantic separation in a safe language - doesn't really matter. As long as a stupid human can't overwrite some other guy's data
koshand in the event of pretty much any kind of problem if you have a child process you can kill it
peakerkosh: Threads are faster for certain types of calculations
explicit processes are safer than explicit threads. but implicit threads are just safe as either
koshumm threads are faster for certain kinds of parallel problems, not for calculations
however they are faster for a very very tiny subset of problems
sproingiedifference between thread and process is moot if you're not sharing data
would be nice to have a standard way to start new processes based on whether you need shared state, pre-emption, etc
peakerkosh: not so tiny. if a thread computes large amounts of data, as a fork/exec'd process it would have to send it over pipes - which requires far more copying/etc
sproingieand let the runtime figure out what you need from there
koshunless you propose that threads can be run under a different uid then the parent then there are still differences
Peaker: what about shared memory, mmap etc?
or a unix domain socket
peakerkosh: that makes it threads, erally
a unix domain socket does 0-copy transfers?
koshI think so
sproingieit tries to
brianhi
http://rafb.net/p/xgAyr714.html
koshwith processes you can start children under another uid, apache actually does that if you configure it to, so you can have it start certain requests for certain users under certain uids
peakerkosh: anyway, back to point from before: Python allows you to communicate your intent. But a language that FORCES you to, can make many more assumptions
brianI'm having a problem with that code snippet
peakerkosh: I don't believe in uid's. I think that's a stupid way to implement security
koshbrian: we need more information then that
brian: like what the problem is
brianSyntax error
After and
On the first line of the if statement

Page: 4 11 18 25 32 39 46 53 60 

IrcArchive

NY Lost Funds