Newest at the top
2024-05-21 21:17:23 +0200 | waleee | (~waleee@h-176-10-144-38.NA.cust.bahnhof.se) (Ping timeout: 264 seconds) |
2024-05-21 21:08:59 +0200 | Guest86 | (~Guest86@186.82.99.37) (Quit: Client closed) |
2024-05-21 20:58:18 +0200 | <EvanR> | worker threads that can wait for work, furiously accomplish something and go back to sleep in ~0 time, vs the thread running the black hole merger simulation |
2024-05-21 20:55:05 +0200 | mesaoptimizer | (~mesaoptim@user/PapuaHardyNet) |
2024-05-21 20:54:51 +0200 | mesaoptimizer | (~mesaoptim@user/PapuaHardyNet) (Quit: mesaoptimizer) |
2024-05-21 20:52:29 +0200 | yin | (~yin@user/zero) |
2024-05-21 20:50:40 +0200 | waleee | (~waleee@h-176-10-144-38.NA.cust.bahnhof.se) |
2024-05-21 20:50:01 +0200 | kuribas | (~user@ptr-17d51encis8jg2ccf48.18120a2.ip6.access.telenet.be) (Remote host closed the connection) |
2024-05-21 20:43:28 +0200 | ezzieyguywuf | (~Unknown@user/ezzieyguywuf) |
2024-05-21 20:42:39 +0200 | <[exa]> | like, most threads in existence just wait for some resource to arrive anyway, right |
2024-05-21 20:42:38 +0200 | raehik | (~raehik@rdng-25-b2-v4wan-169990-cust1344.vm39.cable.virginm.net) |
2024-05-21 20:42:12 +0200 | <dolio> | Although I assume they weren't doing much. |
2024-05-21 20:41:59 +0200 | <dolio> | There's an old ghc tracker ticker where Simon Marlow says he ran 1 million threads without a problem. |
2024-05-21 20:41:24 +0200 | <dolio> | Yeah. |
2024-05-21 20:41:02 +0200 | <[exa]> | dolio: btw the last time I read the RTS the total haskell IO-thread (_not_ OS thread) count was much more like memory-bounded than actually switch-starvation-bounded |
2024-05-21 20:40:41 +0200 | Guest86 | (~Guest86@186.82.99.37) |
2024-05-21 20:39:42 +0200 | <mauke> | (mind, this was a 32-bit system with limited RAM) |
2024-05-21 20:39:09 +0200 | <[exa]> | let's take a moment now to commemorate the magnificent glorious 1 globally interpreter-locked thread of python |
2024-05-21 20:39:04 +0200 | <mauke> | haskell got very slow, but still made visible progress at 300,000 threads |
2024-05-21 20:38:25 +0200 | <mauke> | perl was under 100 |
2024-05-21 20:38:02 +0200 | <mauke> | last time I tried to benchmark thread systems (by creating an ever-growing bucket chain of threads in various systems/languages), C/pthreads maxed out at a couple hundred threads IIRC |
2024-05-21 20:35:38 +0200 | <mauke> | fibers, maybe |
2024-05-21 20:35:03 +0200 | ft | (~ft@p508db8fc.dip0.t-ipconnect.de) |
2024-05-21 20:24:07 +0200 | <geekosaur> | Gtk provides its own event loop |
2024-05-21 20:23:44 +0200 | <geekosaur> | pthread_create. and libuv is an event loop, not a thread multiplexer |
2024-05-21 20:22:06 +0200 | <dolio> | Yeah, that sounds right. |
2024-05-21 20:21:57 +0200 | <Rembane> | The uv thing? |
2024-05-21 20:21:56 +0200 | <monochrom> | But lazy evaluation is the major difference. It also has the domino effect of causing many other differences e.g. how and why GHC does heap, closures, and GC in a way a C compiler doesn't. |
2024-05-21 20:21:53 +0200 | gentauro | (~gentauro@user/gentauro) |
2024-05-21 20:21:43 +0200 | <dolio> | Are they actually using OS threads? I thought there was some other C thing that people used when they wanted that level of concurrency. |
2024-05-21 20:21:11 +0200 | <lxsameer> | mauke: cheers |
2024-05-21 20:20:32 +0200 | <monochrom> | We certainly recommend "don't bother writing your own select event loop, just fork more threads". So the RTS has to actually optimize for that. >:) |
2024-05-21 20:20:01 +0200 | <geekosaur> | C/Gtk programmers have told me otherwise (thousands of threads) |
2024-05-21 20:19:21 +0200 | <dolio> | At least, last I heard. |
2024-05-21 20:18:44 +0200 | <dolio> | Yeah, I guess. The problem is that you can't get away with having as many OS threads as people want Haskell threads. |
2024-05-21 20:18:20 +0200 | raehik | (~raehik@rdng-25-b2-v4wan-169990-cust1344.vm39.cable.virginm.net) (Ping timeout: 260 seconds) |
2024-05-21 20:17:59 +0200 | peterbecich | (~Thunderbi@syn-047-229-123-186.res.spectrum.com) (Ping timeout: 264 seconds) |
2024-05-21 20:17:19 +0200 | <monochrom> | It also turns out we want to liberally move Haskell threads to any OS thread at a whim. |
2024-05-21 20:16:25 +0200 | <monochrom> | OS stacks would be OK if there were a 1-1 mapping from Haskell threads to OS threads. OS threads already enjoy individual stacks. But of course we are more ambitious, we want our own cramming M Haskell threads into 1 OS thread. |
2024-05-21 20:16:02 +0200 | gentauro | (~gentauro@user/gentauro) (Read error: Connection reset by peer) |
2024-05-21 20:16:02 +0200 | awnmp | (~awnmp@user/awnmp) |
2024-05-21 20:14:25 +0200 | <dolio> | Yeah. Switching out OS stacks is bad news, I think. |
2024-05-21 20:14:21 +0200 | <monochrom> | As a bonus, stack is also growable and movable. |
2024-05-21 20:14:02 +0200 | <monochrom> | Ah yeah, then also N Haskell threads can have N distinct Haskell stacks too. |
2024-05-21 20:13:26 +0200 | Guest86 | (~Guest86@186.82.99.37) (Client Quit) |
2024-05-21 20:12:33 +0200 | <dolio> | That's probably not the only reason. |
2024-05-21 20:11:29 +0200 | <monochrom> | The fact that Haskell FFI works best if Haskell code leave OS-sanctioned stack alone (so C code can use it unconfused) so ghc-generated code uses another register and another memory area for Haskell stack. |
2024-05-21 20:11:16 +0200 | <mauke> | lxsameer: I like direct-sqlite. no built-in support for migrations, but pretty trivial to add IMHO. depends on what you expect from a migration feature |
2024-05-21 20:11:01 +0200 | Guest86 | (~Guest86@186.82.99.37) |
2024-05-21 20:07:52 +0200 | <monochrom> | ghc-generated code looks different from gcc-generated code because of better reasons than this. Evaluation order. Heap system. Closures. How they decide to use registers. |