00:05:07 Bike, don't suggest that heading should be deleted! 00:05:15 I wouldn't dream. 00:05:23 :))) 00:05:34 It just seemed more language-extension-y than compiler-work-y. 00:06:59 Bike, imagine a world where we could have generic structs or containers, and we wouldn't lose type information! 00:09:08 Bike: package-local nicknames would have made an awesome GSoC project, imo. 00:09:18 it doesn't have to be compiler hacking. 00:09:50 i see. 00:10:04 pkhuong, so you know, having this list is more motivating than reading 4 year old bugs/wishes on the bug tracker 00:12:02 I'm this close to suggesting mathematical optimisation projects, but I don't want to be the person who committed convex optimisation code in SBCL ;) 00:12:36 suggest them! 00:14:21 If compile times weren't an issue, representation selection could likely be improved with a branch-and-bound backed by a specialised relaxation method. 00:15:56 what do you mean representation selection? 00:16:46 I'm writing this one up right now ;) 00:17:16 pkhuong, you've listed so many interesting things, it's as if we are the same person, you're just smarter 00:18:16 quintessence, as it were 00:18:31 I have a long mental list of SBCL todos and wishes. 00:19:56 Bike, :)) 00:42:14 Quadrescence: you can reload from the repo for more on representation selection. 00:51:26 :)) 01:04:23 psilord [~psilord@c-69-180-173-249.hsd1.mn.comcast.net] has joined #sbcl 01:29:18 slyrus [~chatzilla@adsl-99-183-240-66.dsl.pltn13.sbcglobal.net] has joined #sbcl 01:52:04 -!- slyrus [~chatzilla@adsl-99-183-240-66.dsl.pltn13.sbcglobal.net] has quit [Ping timeout: 240 seconds] 02:12:45 how bout: Enhance FFI to support passing and returning structures by value. 03:01:09 -!- Quadrescence [~quad@unaffiliated/quadrescence] has quit [Quit: This computer has gone to sleep] 03:02:39 Quadrescence [~quad@unaffiliated/quadrescence] has joined #sbcl 03:06:42 -!- cmm [~cmm@bzq-79-180-135-215.red.bezeqint.net] has quit [Remote host closed the connection] 03:13:57 erikc: Done already in latest CFFI release. 03:15:00 ohCFFI is different from SBCL FFI? 03:15:04 not just a wrapper? 03:15:37 CFFI manual still lists struct by value as a limitation 03:16:17 ah, via libffi 03:17:02 yep, so it's available portably. 03:41:57 Fare [~fare@173-9-65-97-NewEngland.hfc.comcastbusiness.net] has joined #sbcl 03:59:47 yacks [~yacks@180.151.36.168] has joined #sbcl 04:22:30 attila_lendvai [~attila_le@92.46.0.215] has joined #sbcl 04:22:31 -!- attila_lendvai [~attila_le@92.46.0.215] has quit [Changing host] 04:22:31 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 04:23:46 -!- jsnell [~jsnell@ash.snellman.net] has quit [Ping timeout: 256 seconds] 04:23:52 jsnell [~jsnell@ash.snellman.net] has joined #sbcl 04:32:05 -!- Fare [~fare@173-9-65-97-NewEngland.hfc.comcastbusiness.net] has quit [Ping timeout: 248 seconds] 04:35:53 -!- ebobby [~fms@199.21.86.106] has quit [Quit: Lost terminal] 04:50:43 -!- erikc [~erikc@CPE00222d53fe78-CM00222d53fe75.cpe.net.cable.rogers.com] has quit [Quit: erikc] 05:18:05 dtw [~dtw@pdpc/supporter/active/dtw] has joined #sbcl 05:42:18 -!- drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has quit [Remote host closed the connection] 06:10:14 -!- wbooze [~wbooze@xdsl-78-35-143-201.netcologne.de] has quit [Ping timeout: 255 seconds] 06:15:11 slyrus [~chatzilla@adsl-99-183-240-66.dsl.pltn13.sbcglobal.net] has joined #sbcl 06:19:26 drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has joined #sbcl 06:23:54 sdemarre [~serge@194.81-64-87.adsl-dyn.isp.belgacom.be] has joined #sbcl 06:48:44 -!- sdemarre [~serge@194.81-64-87.adsl-dyn.isp.belgacom.be] has quit [Ping timeout: 246 seconds] 06:53:03 -!- yacks [~yacks@180.151.36.168] has quit [Ping timeout: 260 seconds] 07:11:39 -!- Bike [~Glossina@71-34-68-165.ptld.qwest.net] has quit [Quit: leaving] 07:15:13 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Quit: Leaving.] 07:37:27 stassats [~stassats@wikipedia/stassats] has joined #sbcl 07:59:20 jarod_ch_ [~jarod_che@115.193.163.175] has joined #sbcl 08:10:10 yacks [~yacks@180.151.36.168] has joined #sbcl 08:12:15 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 08:28:07 jarod_c__ [~jarod_che@125.119.103.177] has joined #sbcl 08:28:24 -!- jarod_ch_ [~jarod_che@115.193.163.175] has quit [Ping timeout: 264 seconds] 08:30:17 -!- jarod_c__ [~jarod_che@125.119.103.177] has quit [Read error: Connection reset by peer] 08:31:38 jarod_ch_ [~jarod_che@125.119.103.177] has joined #sbcl 08:45:12 -!- psilord [~psilord@c-69-180-173-249.hsd1.mn.comcast.net] has quit [Ping timeout: 264 seconds] 08:48:09 akovalen` [~user@95.72.171.19] has joined #sbcl 08:50:19 -!- akovalenko [~user@77.51.63.57] has quit [Ping timeout: 264 seconds] 08:50:32 -!- jarod_ch_ [~jarod_che@125.119.103.177] has quit [Ping timeout: 246 seconds] 08:51:02 jarod_ch_ [~jarod_che@183.128.143.223] has joined #sbcl 08:58:24 prxq [~mommer@mnhm-5f75c082.pool.mediaWays.net] has joined #sbcl 08:59:23 psilord [~psilord@c-69-180-173-249.hsd1.mn.comcast.net] has joined #sbcl 09:00:02 stassats` [~stassats@wikipedia/stassats] has joined #sbcl 09:03:43 -!- stassats [~stassats@wikipedia/stassats] has quit [Ping timeout: 260 seconds] 09:13:50 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Ping timeout: 255 seconds] 09:28:13 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 10:10:08 Try cos of 0.41117862 and 0.41117865. Sigh. 10:11:06 could that be because we coerce to double, call libm cos, then coerce back? 10:13:08 *stassats`* had somewhere a branch which used cosf 10:13:22 https://github.com/stassats/sbcl/tree/libm-single-float 10:14:09 stassats`: no. I generate these test cases from C. 10:16:03 i don't understand the significance of this 10:16:35 We can't compute derive numeric ranges correctly. 10:17:34 is there any numeric ranges derivation going on when simply doing (cos 0.41117862)? 10:17:44 uh? 10:17:57 maybe i'm missing some context 10:19:01 foreignFunction [~niksaak@94.27.88.240] has joined #sbcl 10:19:13 fwiw, on that branch, (cos 0.41117865) => 0.9166504 10:19:13 we derive numeric ranges for monotone functions by evaluating at the bounds.. 10:20:41 (mapcar #'cos '(0.41117862 0.41117865)) => (0.91665035 0.91665035) ; mainline => (0.91665035 0.9166504); the branch 10:21:33 indeed. I'm asking libm directly. 10:22:14 so, what was the sighing about? 10:22:24 glibm seems to be off by one ULP at most, but I don't think we have any guarantee elsewhere. In fact, I think x87 is known for being off by up to a dozen ULPs. 10:22:53 *stassats`* sees there was some discussion he missed 10:22:55 we'll most likely have the same phenomenon on double floats. Monotone FP functions aren't. 10:25:49 ah, i see the problem now 10:26:33 there were some changes to single float functions in newer glibc 10:26:52 stassats [~stassats@wikipedia/stassats] has joined #sbcl 10:29:04 though, the double versions appear to be suffering from the same problem 10:30:04 or not 10:30:51 fbsd/x86-64 seems OK. 10:30:52 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Ping timeout: 256 seconds] 10:33:43 on that branch, with glibc-2.17 => (mapcar #'cos '(0.41117862 0.41117865)) => (0.91665035 0.91665035) 10:33:54 the previous result was with glibc-2.13 10:35:50 glibc-2.17 is using sse for single floats 10:36:44 (2.13 does only for doubles) 10:42:04 attila_lendvai [~attila_le@87.247.56.52] has joined #sbcl 10:42:07 -!- attila_lendvai [~attila_le@87.247.56.52] has quit [Changing host] 10:42:07 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 10:54:31 -!- stassats [~stassats@wikipedia/stassats] has quit [Ping timeout: 264 seconds] 10:57:13 angavrilov [~angavrilo@217.71.227.190] has joined #sbcl 11:11:38 stassats`: on your branch, I guess compiling (LAMBDA (X) (DECLARE (TYPE (SINGLE-FLOAT 0.30006453 0.30006456) X)) (COS X)) fails strangely. 11:12:12 The value # is not of type SB-KERNEL:VALUES-TYPE. 11:13:04 works fine with glibc-2.17 11:13:34 that's why i didn't commit that branch, because the current glibc in debian (even testing) is not good enough 11:13:42 err, even unstable 11:14:16 I don't think we can really blame libm. 11:15:03 well, they fixed it, at lest for sse enabled cpus 11:16:06 should sbcl use some safeguards? 11:17:07 I think we should just always derive trivial bounds for everything but basic functions (arithmetic, maybe sqrt). 11:17:38 perhaps maybe log and exp. 11:20:18 maybe use a slow but correct algorithm? 11:20:38 but then it might derive something which differs from what libm returns 11:21:18 the contrary, or you derive things, then upgrade libc, and the fals with derived type are broken 11:22:52 exactly. 11:26:32 what do you mean trivial bounds? rounded off? 11:26:41 [-1, 1] 11:27:05 Stuff like [0, 1] seems reasonable as well 11:27:10 i see 11:27:40 do many optimization depend on exact bound derivations? 11:28:02 I don't think so. We do like the derive tight bouds to provably avoid complex arithmetic, I guess. 11:28:15 well, don't ever change the rounding mode before calling into libm (: 11:32:21 -!- jarod_ch_ [~jarod_che@183.128.143.223] has quit [Quit: Textual IRC Client: http://www.textualapp.com/] 11:53:04 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Quit: Leaving.] 12:02:50 leuler [~user@p548FC74B.dip.t-dialin.net] has joined #sbcl 12:24:38 loke_ [~elias@bb119-74-80-38.singnet.com.sg] has joined #sbcl 12:34:51 -!- stassats` [~stassats@wikipedia/stassats] has quit [Remote host closed the connection] 12:35:18 stassats [~stassats@wikipedia/stassats] has joined #sbcl 12:43:37 LiamH [~none@pdp8.nrl.navy.mil] has joined #sbcl 12:48:04 -!- drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has quit [Remote host closed the connection] 12:48:17 drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has joined #sbcl 12:49:30 -!- leuler [~user@p548FC74B.dip.t-dialin.net] has quit [Quit: ERC Version 5.1.2 $Revision: 1.796.2.6 $ (IRC client for Emacs)] 12:53:13 leuler [~user@p548F9849.dip.t-dialin.net] has joined #sbcl 13:47:33 -!- scymtym_ [~user@2001:638:504:2093:226:b9ff:fe7d:3e1f] has quit [Ping timeout: 240 seconds] 13:53:05 hlavaty`` [~user@friedrichstrasse.knowledgetools.de] has joined #sbcl 13:54:56 -!- hlavaty` [~user@friedrichstrasse.knowledgetools.de] has quit [Ping timeout: 258 seconds] 13:55:51 scymtym_ [~user@2001:638:504:2093:226:b9ff:fe7d:3e1f] has joined #sbcl 13:57:04 -!- drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has quit [Remote host closed the connection] 14:26:17 drmeister [~drmeister@166.137.85.80] has joined #sbcl 14:31:15 -!- drmeister [~drmeister@166.137.85.80] has quit [Remote host closed the connection] 14:31:39 -!- asedeno_ [asedeno@nat/google/x-mcujdcjjwpoqrohp] has quit [Ping timeout: 245 seconds] 14:33:09 asedeno [asedeno@nat/google/x-aciigosfjenipyry] has joined #sbcl 14:41:40 drmeister [~drmeister@farnsworth.chem.temple.edu] has joined #sbcl 14:44:11 zophy [ceb02321@gateway/web/freenode/ip.206.176.35.33] has joined #sbcl 14:59:37 -!- zophy [ceb02321@gateway/web/freenode/ip.206.176.35.33] has quit [Ping timeout: 245 seconds] 15:29:57 Fare [~fare@173-9-65-97-NewEngland.hfc.comcastbusiness.net] has joined #sbcl 15:46:08 Bike [~Glossina@67-5-232-29.ptld.qwest.net] has joined #sbcl 15:53:57 Krystof: page updated. 16:23:22 fab. 16:25:51 -!- loke_ [~elias@bb119-74-80-38.singnet.com.sg] has quit [Remote host closed the connection] 16:38:41 wbooze [~wbooze@xdsl-78-35-131-209.netcologne.de] has joined #sbcl 16:40:10 attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has joined #sbcl 17:26:46 sdemarre [~serge@194.81-64-87.adsl-dyn.isp.belgacom.be] has joined #sbcl 17:29:26 -!- drmeister [~drmeister@farnsworth.chem.temple.edu] has quit [Remote host closed the connection] 17:30:06 drmeister [~drmeister@farnsworth.chem.temple.edu] has joined #sbcl 17:34:42 -!- drmeister [~drmeister@farnsworth.chem.temple.edu] has quit [Ping timeout: 252 seconds] 18:13:20 drmeister [~drmeister@pool-173-59-25-70.phlapa.fios.verizon.net] has joined #sbcl 18:43:05 -!- attila_lendvai [~attila_le@unaffiliated/attila-lendvai/x-3126965] has quit [Ping timeout: 255 seconds] 19:02:36 -!- wbooze [~wbooze@xdsl-78-35-131-209.netcologne.de] has quit [Ping timeout: 264 seconds] 19:37:24 -!- angavrilov [~angavrilo@217.71.227.190] has quit [Ping timeout: 264 seconds] 19:57:20 -!- yacks [~yacks@180.151.36.168] has quit [Quit: Leaving] 20:12:37 ebobby [~fms@199.21.86.106] has joined #sbcl 20:16:34 Damnation. I knew there was something I forgot to fix before this release. (Do Not Do Releases Late At Night Without A Checklist) 20:18:22 still, at least the release script works 20:20:26 pkhuong: i was interested in the serialization one. 20:20:36 -!- stassats [~stassats@wikipedia/stassats] has quit [Ping timeout: 264 seconds] 20:27:30 zophy [ceb02321@gateway/web/freenode/ip.206.176.35.33] has joined #sbcl 20:35:17 -!- foreignFunction [~niksaak@94.27.88.240] has quit [Quit: Leaving.] 20:49:54 Bike: I found it hard to justify a learning angle. I still think it's an interesting project though. 20:50:51 pkhuong: I've been thinking about how serializing and deserializing closures would work, so I was especially interested in that part...maybe you could sell it as learning about the sorts of things object file formats do, or relocation or something. 20:50:52 wbooze [~wbooze@xdsl-78-35-152-86.netcologne.de] has joined #sbcl 20:59:00 -!- Bike [~Glossina@67-5-232-29.ptld.qwest.net] has quit [Ping timeout: 264 seconds] 21:02:43 Bike [~Glossina@67-5-232-29.ptld.qwest.net] has joined #sbcl 21:02:50 pkhuong: awesome (and awesomely-long) list :-) 21:05:06 do you want to post to the mailing list? 21:06:11 Krystof: oh yeah, that'd be a good idea. 21:09:59 -!- dtw [~dtw@pdpc/supporter/active/dtw] has quit [Quit: ERC Version 5.3 (IRC client for Emacs)] 21:17:03 foreignFunction [~niksaak@94.27.88.158] has joined #sbcl 21:18:57 Krystof: re guidance for proto hackers, a friend suggested that we borrow some other projects' policy and first ask the mentee to build SBCL and submit a patch that is eventually merged in. 21:30:20 Hydan [~hydan@ip-89-103-110-116.net.upcbroadband.cz] has joined #sbcl 21:32:19 That list of projects is a) wow, long, and b) unfortunately nearly unreadable with github's stylesheet. 21:33:13 foom: is http://www.sbcl.org/gsoc2013/ideas better? 21:33:26 yes. 21:34:36 pkhuong: sadly it is too late for me to edit the organization application 21:34:40 but that is a good idea 21:35:15 Well, you can contact the people applying and ask them directly to do that, can't you? 21:35:33 mentees, yes, if we are accepted as an organization 21:35:47 honestly I would expect our chances are rather low 21:35:59 Aw, why? 21:36:23 It seems as if Gsoc has been selecting fewer orgs in recent years than they did in the past. 21:36:55 I'm not doing this for GSoC, frankly. Hopefully, people will be attracted regardless. 21:37:01 indeed. 21:37:42 i'd like to do one of these projects, but it would be great to get it as part of gsoc. I mean, money. 21:43:36 It'd be great if the list of projects had links to existing applicable work/email threads, where they exist. 21:44:17 foom: good idea. I'll see when I can do that. 21:47:08 -!- ebobby [~fms@199.21.86.106] has quit [Ping timeout: 255 seconds] 21:47:24 pkhuong: at the very least I can point undergrads / masters students to it 21:47:40 of course, most of ours would say "whut?" faced with that list, but... 21:47:41 Are the projects sorted according to anyone's idea of suitability for a new hacker or importance? 21:48:31 it's more or less sorted by difficulty, as best as I could estimate 21:49:20 then again, the estimate is probably less accurate for students who don't follow umontreal-style curricula. 21:50:02 It seems a bit eye-glazingly long right now; might be better if the huge list was split up somehow, maybe into chunks by estimated difficulty. 21:50:24 foom: do you think themes would be better? 21:50:35 Yes, that works too. 21:50:54 ebobby [~fms@199.21.86.106] has joined #sbcl 21:51:31 Just seeing a list of 24 items in a single list is kinda like wow, too many choices. 21:53:44 -!- LiamH [~none@pdp8.nrl.navy.mil] has quit [Quit: Leaving.] 21:59:05 It might also be useful to put the gain in context. "Knowing about GC" seems cool, but what kind of career, etc, can I get with that? (just brainstorming here) 21:59:31 prxq: I'll be damned if I know ;) 21:59:40 heh 21:59:49 :-) 22:00:20 I tried to phrase these things the way I'd present them in my CV, but it's not always obvious what to write except "it'd be incredibly awesome" 22:00:58 Now that you mention it, that actually is missing 22:00:58 kind of career? 22:01:32 foom: what kind of job can that knowledge get you 22:01:46 a software engineering job. :) 22:02:00 -!- sdemarre [~serge@194.81-64-87.adsl-dyn.isp.belgacom.be] has quit [Ping timeout: 252 seconds] 22:02:26 Yeah, which of these would be most incredibly awesome? Which of the projects will most excite the sbcl community to have someone working on them? 22:05:01 I guess to me, the GC investigations sound the most exciting to have someone work on, but they're of the type of project that might end up in a total lack of delivery of anything useful after a few months of work, despite best efforts. 22:05:17 btw, some of the GC ones seem to be contradictory. 22:05:43 foom: because it may be too hard? 22:05:50 prxq: yup. 22:06:20 foom: they're at the bottom of my list for a reason (: I consider all the runtime work extra hard, because you need to fit more stuff in your head to make progress. 22:06:23 prxq: mainly because they're about performance, and it's hard to figure out if the thing you're gonna do will improve performance or not. 22:06:29 until after you've done it 22:07:41 The current GC has problems when you run threads that cons on different cores, as it synchronizes them, so everything runs on one core. Fixing that would be great. 22:07:58 pkhuong: Right, I'm not sure that a newbie would actually understand that it's probably impossible. :) 22:08:00 that would mean lots of performance. 22:08:52 "Make SBCL GC ready for the multicore era" sounds halfway buzzword compliant :-) 22:09:16 prxq: I'm pretty sure it'd be easier to fit MPS in than to write a parallel GC (never mind a mostly concurrent one) 22:09:42 otherwise, mark and sweep for tenured data would likely be better suited to parallelisation. 22:10:48 After considering a bit more... 22:11:30 I think the uncomfortableness I've been having with this list is that none of the things seem like they're actually concrete proposals that have already been discussed/accepted by a/some maintainers in principle. 22:11:33 The most successful internships I've been involved with or watched have been things which everyone vaguely knows/agrees on how it should happen, but nobody's had the time to do the implementation work and fully flesh it out. 22:11:35 prxq: I'd probably try to work out some sort of shared multi process heap hack instead (: 22:12:00 foom: actually, a lot of them, I have branches. 22:12:51 pkhuong: sounds interesting! what is that? :-) 22:13:13 pkhuong: so basically, what you need is students 22:13:43 Unicode has Krystof; GMP, division, switch/case, coarse types, peephole, structures, software write barriers, allocator, and mark and sweep, I have code/branches. 22:15:29 For the record, my experience with sbcl GC is that it's quite good. I've observed 10% overhead with code that consed without much regard for anything. 22:15:50 Id est, looks hard to beat. 22:15:56 -!- edgar-rft [~GOD@HSI-KBW-149-172-63-75.hsi13.kabel-badenwuerttemberg.de] has quit [Quit: eternal darkness] 22:16:47 prxq: run multiple processes, implement stuff like fork and join with actual forks, and find a way to pun unix shared memory in the lisp heap. 22:17:13 pkhuong: uh oh. 22:17:16 :-) 22:18:11 -!- Bike [~Glossina@67-5-232-29.ptld.qwest.net] has quit [Ping timeout: 255 seconds] 22:20:08 Bike [~Glossina@174-25-51-73.ptld.qwest.net] has joined #sbcl 22:26:57 -!- wbooze [~wbooze@xdsl-78-35-152-86.netcologne.de] has quit [Remote host closed the connection] 22:27:48 pkhuong: Wow, you have a lot of branches. I didn't know there was existing work for most of those. :) 22:28:31 I think I only knew about peephole, barriers, and struct layout. 22:28:47 wbooze [~wbooze@xdsl-78-35-152-86.netcologne.de] has joined #sbcl 22:29:14 foom: I've discussed some of the other stuff with _8david` and nikodemus here. I believe they have code as well (: 22:31:32 I wonder if there's some way to make the work people are doing but that hasn't been finished more publicly visible. 22:32:09 I guess a list like this. :) 22:34:33 -!- Hydan [~hydan@ip-89-103-110-116.net.upcbroadband.cz] has quit [Ping timeout: 240 seconds] 22:39:32 Krystof maintained a text file of private branches for a while 22:43:05 did I? I've expunged what must have been a miserable experience in administration from my mind 22:43:08 I blame the children 22:43:38 I believe you had a script do it. 22:44:16 that does ring a bell 22:49:15 -!- zophy [ceb02321@gateway/web/freenode/ip.206.176.35.33] has quit [Quit: Page closed] 22:51:36 -!- Fare [~fare@173-9-65-97-NewEngland.hfc.comcastbusiness.net] has quit [Ping timeout: 260 seconds] 23:03:37 foom: does it seem mroe approachable now? 23:04:40 Fare [fare@nat/google/x-gvnyzkhcjrbtziwk] has joined #sbcl 23:15:02 -!- prxq [~mommer@mnhm-5f75c082.pool.mediaWays.net] has quit [Quit: Leaving]