| 01:40.18 | *** join/#riftuidev Natch (~Natch@c-eccde155.25-4-64736c10.cust.bredbandsbolaget.se) |
| 01:42.25 | seebs | So, ZorbaBeta: Idle question, is the addon API in general in Rift's main thread? Although I'd imagine it would be a ton of work, if it *were* running as a separate thread, it would give us a ridiculous amount of free CPU time, since basically everyone running Rift has at least one CPU core at <5% usage. :) |
| 01:44.31 | Imhothar | it's runnign in the main thread together with the game renderer |
| 01:45.10 | Imhothar | At least that's what he said long time ago in a thread about multicore stuff |
| 01:51.12 | seebs | I am going to keen softly until he fixes it. Or I lose interest in the task, whichever comes first. |
| 01:51.48 | seebs | I have spent most of the last two days cleaning my desks. Well. Cleaning desk and doing things that are part of that category of tasks, like getting the loose music hardware plugged in and wired up again. |
| 01:51.56 | Imhothar | I would be quite a challenging task |
| 01:52.16 | Imhothar | you'd need to synchronize every Inspect.* call with the game logic database |
| 01:52.41 | seebs | Hmm, I see the issues. |
| 01:52.45 | Imhothar | probably the same datastructures the rest of the game is using |
| 01:53.28 | Imhothar | unless you turned every Inspect.* into an callback-driven asynchronous model |
| 01:54.03 | seebs | Hmm. |
| 01:54.14 | seebs | If only memory were cheap, you could do it pretty well by using snapshots. |
| 01:54.37 | Imhothar | doubt you could update the snapshots fast enough |
| 01:54.48 | Imhothar | imagine a raud situation |
| 01:54.50 | Imhothar | raid |
| 01:54.56 | seebs | Addon API gets called on a data set which is at most one frame old, anything it wants to do which has to get back to the main environment is queued up for processing, etcetera. |
| 01:55.20 | seebs | Well, keep in mind, for many people right now, raids and the like already imply dropping to 10fps or less. |
| 01:55.36 | seebs | 20fps with data one frame old would actually be better. :) |
| 01:55.50 | Imhothar | you're assuming it's the addon envuronment dragging down fps |
| 01:55.56 | Imhothar | the game is doing so many other things |
| 01:56.57 | seebs | True. Well, I'm mostly assuming that the general "one main thread doing everything" is bad for us. *sigh* |
| 01:57.02 | seebs | I would really, really, love to know how that happened. |
| 01:57.30 | seebs | Pretty sure that, when the first line of code was written for Rift, you simply couldn't *buy* a single-core PC except maybe a netbook. |
| 01:57.34 | Imhothar | you know, it may be misleading |
| 01:57.54 | Imhothar | the OS does it's own magic to task scheduling, including thread to cpu assignment |
| 01:58.14 | Imhothar | you could have two threads running on the same core because the OS thinks it's a good thing to do |
| 01:58.32 | seebs | We've been told a few times that there really is one main thread that does (almost) everything. |
| 01:58.38 | Imhothar | and there are good reasons for doing that |
| 01:58.44 | Imhothar | oh |
| 01:58.49 | Imhothar | in that case |
| 01:59.07 | seebs | And this is why changing mainthreadcpu to 0 instead of 3 doesn't really help; it just lets the OS move the thread around, it doesn't make things more multithreaded. |
| 01:59.23 | Imhothar | in that case, find that guy, tar and feather him and tell him to fix it |
| 02:01.25 | seebs | I just want to know how it happened. Did the version of Gamebryo they started with impose this limit somehow? Did the guy who was going to do the multithreading suddenly get hit by a bus right after converting to a thematically appropriate religion? |
| 02:01.53 | Imhothar | oh well |
| 02:01.59 | seebs | I mean, the game's not perfect or anything, but just about all the other crazy stuff I've seen or inferred at least made some kind of sense at some level. |
| 02:02.00 | Imhothar | multirhreading is Pandora's Box |
| 02:02.08 | Imhothar | if you don't know what exactly to expect don't open it |
| 02:02.39 | seebs | heh |
| 02:02.54 | Imhothar | hah interesting. I opened rift.exe by accident and got the "datafiles corrupt" error. is that the default message when something's not right? |
| 02:02.58 | seebs | ask me someday about how pseudo's thread-safety code interacts with signal handlers. |
| 02:03.03 | seebs | I am not sure. |
| 02:04.06 | Imhothar | Ì have a collegue with no IT education and he did some multhreading in his (selling) product. He had no idea about deadlocks, race doncitions, starvation or what a mutex is. |
| 02:04.13 | Imhothar | he was veeeeery lucky it worked |
| 02:04.38 | Imhothar | he was very far on the edge of the bell curve |
| 02:05.00 | seebs | wow |
| 02:05.11 | seebs | You familiar with the "dining philosophers" problem? |
| 02:05.12 | Imhothar | and even after he did some reading about it he still doesn't get it right |
| 02:05.28 | Imhothar | maybe in german |
| 02:05.49 | Imhothar | ah yes |
| 02:06.15 | seebs | (Trivia point: I studied German for a year or two in school, then ended up learning Chinese and losing most of the German.) |
| 02:06.17 | Imhothar | just wasn't aware what the english term is for it |
| 02:06.45 | seebs | So, I never did a computer science program, and a friend posed the problem for me as an illustration of the kind of thing people would want you to be able to do if you had a CS background. |
| 02:06.53 | seebs | I spent... oh, three or four *years* thinking about it from time to time. |
| 02:07.03 | seebs | And I eventually admitted that I was completely unable to see how it could be solved. |
| 02:07.39 | seebs | I could find no way of resolving this that didn't require adding some extra item to use to keep track of things, like a lock, or a random retry that could conceivably deadlock for a very long time but in practice will always resolve quickly. |
| 02:08.03 | seebs | ... It turns out that those ARE the solutions. I just thought they didn't count because the problem as described had not stated that I could add new items. |
| 02:08.58 | Imhothar | deadlocks are interesting, though race conditions create the hardest-to-find bugs |
| 02:10.32 | Imhothar | 1. race conditions are extremely nondeterministic and 2. it's very difficult to find out a bug is caused by a race condition |
| 02:12.10 | Imhothar | plus: it's very difficult for most untrained people to imagine multiple concurrently running processes |
| 02:12.30 | seebs | Yeah. |
| 02:12.40 | seebs | One of the things I'm relatively good at. |
| 02:13.15 | seebs | The trick is getting the processes sort of simultaneously mapped out, and then shuffling their order of execution a bit until something breaks. |
| 02:13.41 | seebs | hah! I found a place on my desk where I could set up my gaming laptop. |
| 02:13.51 | Imhothar | UML is also a big help as in most other situations |
| 02:14.11 | seebs | I have one monitor left that I haven't found a place to put it, but the rest of my setup is mostly done. |
| 02:14.34 | seebs | I still have very very little formal background in computer science. Never took classes, though I've read books. |
| 02:14.43 | seebs | Mostly, I get by on autism and a fanatical desire to understand things. |
| 06:14.13 | *** join/#riftuidev Imhothar (~miranda@ppp-188-174-82-119.dynamic.mnet-online.de) |
| 06:16.00 | *** join/#riftuidev ChanServ (ChanServ@services.) |
| 06:16.00 | *** mode/#riftuidev [+o ChanServ] by leguin.freenode.net |
| 06:19.37 | *** join/#riftuidev zz_Cidan (~cidan@conspiracy/developer/Cidan) |
| 06:21.37 | *** join/#riftuidev Cairenn (~Owner@MMOI/Administratrix/Cairenn) |
| 06:21.37 | *** join/#riftuidev bctrainers (~bc@unaffiliated/bctrainers) |
| 06:21.37 | *** mode/#riftuidev [+o Cairenn] by leguin.freenode.net |
| 06:22.01 | *** join/#riftuidev Kaelten (~Kaelten@WoWUIDev/WoWAce/WoWIFA/CurseStaff/kaelten) |
| 06:22.01 | *** mode/#riftuidev [+o Kaelten] by leguin.freenode.net |
| 08:13.23 | *** join/#riftuidev Noshei (~quassel@c-67-161-146-244.hsd1.co.comcast.net) |
| 09:49.11 | *** join/#riftuidev Noshei (~quassel@c-67-161-146-244.hsd1.co.comcast.net) |
| 10:04.37 | *** join/#riftuidev Noshei (~quassel@c-67-161-146-244.hsd1.co.comcast.net) |
| 10:07.43 | *** join/#riftuidev Noshei (~quassel@c-67-161-146-244.hsd1.co.comcast.net) |
| 13:13.17 | *** join/#riftuidev g0urra (~g0urra@unaffiliated/gourra) |
| 14:18.11 | *** join/#riftuidev Noshei (~quassel@c-67-161-146-244.hsd1.co.comcast.net) |
| 20:26.33 | *** join/#riftuidev Natch (~Natch@c-eccde155.25-4-64736c10.cust.bredbandsbolaget.se) |