00:01:47 our code doesn't depend on SSAness yet, so that won't break correctness. 00:06:22 I'm lost, I'm afraid. Might just be the wine, though. 00:11:45 What SSA gets you is an efficient way to base analyses on the (conservative approximation of) set of potential defining assignments at each use. 00:14:19 we can get a less efficient way to base analyses on a more conservative approximation by inserting copy nodes as needed. 00:16:09 Ah. 00:16:43 I think that I might need to understand the analysis that SBCL already does, then. 00:20:40 so far, the best thing I can think of is to do it during when converting into IR1 00:21:21 or, at least at some point between IR1 conversion and type assignment. 00:23:24 Who's this bill hart and what is he trying to do? 00:24:33 I don't know, but he clearly doesn't have any sort of a feel for the performance model involved. 00:24:48 Or, for that matter, the /language/ model. 00:26:21 -!- attila_lendvai [~attila_le@catv-80-98-24-21.catv.broadband.hu] has quit [Quit: Leaving.] 00:39:54 -!- tcr1 [~tcr@80-218-246-66.dclient.hispeed.ch] has quit [Quit: Leaving.] 00:50:24 some prof at warwick into computerised number theory 00:51:19 I guess he might be trying to compile computational kernels at runtime instead of going through C. 01:24:31 -!- homie [~levgue@xdsl-78-35-147-195.netcologne.de] has quit [Quit: ERC Version 5.3 (IRC client for Emacs)] 01:26:42 homie [~levgue@xdsl-78-35-147-195.netcologne.de] has joined #sbcl 02:29:12 -!- hargettp [~hargettp@96.237.127.144] has quit [Quit: Leaving...] 02:55:06 -!- nyef [~nyef@pool-70-20-57-204.man.east.myfairpoint.net] has quit [Quit: G'night all.] 03:33:55 lambda-avenger [~roman@adsl-99-185-244-104.dsl.pltn13.sbcglobal.net] has joined #sbcl 06:29:30 homie` [~levgue@xdsl-78-35-182-71.netcologne.de] has joined #sbcl 06:32:37 tcr1 [~tcr@80-218-246-66.dclient.hispeed.ch] has joined #sbcl 06:32:39 -!- homie [~levgue@xdsl-78-35-147-195.netcologne.de] has quit [Ping timeout: 276 seconds] 07:09:35 -!- tcr1 [~tcr@80-218-246-66.dclient.hispeed.ch] has quit [Quit: Leaving.] 07:30:38 flip214 [~marek@unaffiliated/flip214] has joined #sbcl 07:35:19 antgreen` [~user@CPE00222d6c4710-CM00222d6c470d.cpe.net.cable.rogers.com] has joined #sbcl 07:36:47 -!- antgreen [~user@CPE00222d6c4710-CM00222d6c470d.cpe.net.cable.rogers.com] has quit [Ping timeout: 246 seconds] 07:53:56 -!- antgreen` [~user@CPE00222d6c4710-CM00222d6c470d.cpe.net.cable.rogers.com] has quit [Ping timeout: 246 seconds] 08:15:41 maybe I should go and talk to said prof at warwick 08:39:30 tcr1 [~tcr@80-218-246-66.dclient.hispeed.ch] has joined #sbcl 09:34:39 -!- Krystof [~csr21@csrhodes.plus.com] has quit [Ping timeout: 240 seconds] 10:06:26 attila_lendvai [~attila_le@catv-80-98-24-21.catv.broadband.hu] has joined #sbcl 10:33:09 -!- tcr1 [~tcr@80-218-246-66.dclient.hispeed.ch] has quit [Ping timeout: 276 seconds] 10:35:54 tcr1 [~tcr@217-162-131-235.dynamic.hispeed.ch] has joined #sbcl 10:41:05 -!- homie` [~levgue@xdsl-78-35-182-71.netcologne.de] has quit [Quit: ERC Version 5.3 (IRC client for Emacs)] 10:43:32 homie [~levgue@xdsl-78-35-182-71.netcologne.de] has joined #sbcl 10:48:05 ilmari [ilmari@knuth.ping.uio.no] has joined #sbcl 11:02:52 -!- lambda-avenger [~roman@adsl-99-185-244-104.dsl.pltn13.sbcglobal.net] has left #sbcl 11:04:44 hargettp [~hargettp@96.237.127.144] has joined #sbcl 15:16:32 ccl-logbot [~ccl-logbo@setf.clozure.com] has joined #sbcl 15:16:32 -!- names: ccl-logbot lnostdal green` hargettp ASau flip214 tcr1 @Krystof ilmari homie slyrus cmm tsuru angavrilov mnemonicsloth |3b|``` jiacobucci luis redline6561 christoph_debian derrotebaron jsnell gnooth pkhuong minion specbot lisppaste2 fe[nl]ix @Xof deepfire froydnj joshe 15:54:34 -!- flip214 [~marek@unaffiliated/flip214] has quit [Remote host closed the connection] 16:06:44 mega1 [~quassel@2001:470:1f05:548:215:58ff:fe7d:773b] has joined #sbcl 16:28:34 Kaek [~b@c-9ccde253.97-16-64736c12.cust.bredbandsbolaget.se] has joined #sbcl 16:43:20 -!- Kaek [~b@c-9ccde253.97-16-64736c12.cust.bredbandsbolaget.se] has quit [Read error: Connection reset by peer] 16:45:46 Kaek [~b@c-9ccde253.97-16-64736c12.cust.bredbandsbolaget.se] has joined #sbcl 17:43:08 nyef [~nyef@pool-70-20-57-204.man.east.myfairpoint.net] has joined #sbcl 17:43:18 Hello all. 17:47:03 rmarynch [~roman@88.135.194.233] has joined #sbcl 17:50:19 hi nyef 17:50:50 Anything interesting happening? 17:51:12 Other than the regression that I'm bisecting now, that is? 17:52:26 -!- hargettp [~hargettp@dhcp-162.mirrorimage.net] has quit [Remote host closed the connection] 18:01:22 in the TODO file I see 'Cross-platform GUI'. What does it mean for SBCL (and is it actual at all)? 18:03:26 probably doesn't mean anything at this point 18:03:35 any such development should take place externally to sbcl 18:04:17 Heh. Didn't the "function names" one get done? 18:04:49 I don't think it did, but it's possible 18:04:51 And I /know/ that "undefined function / variable restarts" have been... mostly ignored for quite a while, even with proof-of-concept patches available. 18:05:00 I thought nikodemus fixed it within the past year? 18:05:57 I only vaguely understand IR1 patches :) 18:06:13 hargettp [~hargettp@dhcp-162.mirrorimage.net] has joined #sbcl 18:06:38 I don't understand how we could get a fault at *0* exactly. 18:06:43 Ahh. The cross-platform gui entry is clearly a joke. 18:07:05 I wondered about that too 18:07:08 pkhuong: I didn't the one time I tried it and looked at the error message, it was somewhere in the #x1xxx range. 18:07:13 the global variables entry can get crossed off! 18:07:47 The socket, RPC, and GUI items really should be external projects anyway. 18:09:19 yes 18:09:48 there were some peephole opt. patches in sbcl-devel this year, but it seems that they are not committed. 18:10:35 Weren't they more proof-of-concept than anything else? 18:11:20 IIRC, the author showed some working cases. Maybe Nikodemus knows better, he reviewed the code 18:12:05 Mmm. Between most/all of the committers having lives, and not everyone being able to reasonably review the various patches... 18:17:52 nyef: at, that makes more sense. 18:17:59 *ah 18:21:34 ... "There's an EMF cache? Unbelievable!" 18:23:04 where ? 18:23:16 "Memory fault at b4efe (pc=0x4fb82d4c sp=(nil))" 18:25:07 Ooh. 1.0.43.67 is "bad", which means the d-x closure stuff can't be at fault! 18:25:27 Finally, something unrelated to d-x closures! :-D 18:28:51 I have one point to add into TODO: line-based code profiling. disassembler annotations are good, but not always enough 18:30:04 rmarynch: form-based source locations, at best, and I'm not sure all the transformations we use make that possible. 18:30:33 pkhuong: even with (debug 3) ? 18:33:01 rmarynch: unless you want to radically redo a lot of the compiler, yes 18:34:31 No, I would like to stay away from massive changes, they are hard to merge back to the upstream :) 18:41:06 foom [~jknight@ita4fw1.itasoftware.com] has joined #sbcl 18:41:20 -!- foom [~jknight@ita4fw1.itasoftware.com] has quit [Client Quit] 18:41:41 foom [~jknight@ita4fw1.itasoftware.com] has joined #sbcl 18:54:46 -!- rmarynch [~roman@88.135.194.233] has quit [Quit: Leaving] 19:04:13 homie` [~levgue@xdsl-84-44-178-68.netcologne.de] has joined #sbcl 19:07:17 -!- homie [~levgue@xdsl-78-35-182-71.netcologne.de] has quit [Ping timeout: 264 seconds] 19:14:49 stassats [~stassats@wikipedia/stassats] has joined #sbcl 19:19:16 rmarynch [~roman@88.135.194.233] has joined #sbcl 19:30:06 ok, so AFAICT, the bug is because we call find-or-create-constraint on a conset-less lambda-var 19:31:55 Which means...? 19:32:07 dun dun dunnnn 19:32:08 *nyef* hasn't even looked at consets yet. 19:32:30 nyef: that function assume that it's a conset 19:32:56 so i reads at some offset away from NIL and uses that as a bitvector 19:33:52 -!- christoph_debian [~user@oteiza.siccegge.de] has quit [Remote host closed the connection] 19:34:29 safety 0? 19:34:52 right 19:35:18 dun dun dunnnn 19:36:40 Ugh. Reminds me of that thing with the simple-error or whatever it was where the condition class slot wasn't typechecked, but the compiler believed the type declaration... I don't think we solved that one. 19:37:32 christoph_debian [~user@oteiza.siccegge.de] has joined #sbcl 19:40:12 -!- rmarynch [~roman@88.135.194.233] has quit [Quit: Leaving] 19:40:35 ah. 19:40:56 with types, we can have a binding that's constant even when setfed 19:43:31 Makes a certain amount of sense, yes. 19:43:50 This breaks the set code somehow? 19:44:50 it breaks the 3rd leg of the cond in add-test-constraints 19:46:34 where it assumes that if arg2 is constant-lvar-p, and is bound to a leaf, the leaf is constant. 19:46:40 but it's constant, yet SETFed. 19:48:10 If it's known to be set to the same value it already has, surely you can eliminate the SET node? 19:49:10 that would be one way, true. 19:49:58 eliminate the SET, recompute lambda-var-sets and closure-var-p. 19:53:42 I have a fix for the assumption in add-test-constraints still 19:54:58 pfdietz' random tester is a fantastic tool 19:55:06 it is. 19:55:26 I wonder if it'd be better known if hed invented a term for it as well ;) 19:58:42 so... we're getting surprisingly strong stuff by piggy backing on type and constraints. 20:06:33 How strong, and how surprising? 20:07:04 almost sufficiently smart? 20:07:39 nyef: well, strong enough to break assumptions in our own code ;) 20:08:14 the sbcl singularity! 20:11:04 still, it does stuff like compiling (if (eql x 1d0) (1+ x) x) to an eql and a cmov (and no arithmetic) 20:11:22 random flow sensitive constant propagation ;) 20:17:16 that's handy :) 20:19:49 I wonder why that doesn't work for float comparisons instead of arithmetic 20:19:59 also, a bit sad that it can compile to a cmov of eql values 20:43:37 -!- stassats [~stassats@wikipedia/stassats] has quit [Remote host closed the connection] 20:44:42 stassats [~stassats@wikipedia/stassats] has joined #sbcl 21:08:25 Fare [~Fare@ita4fw1.itasoftware.com] has joined #sbcl 21:16:50 what are your feelings on creating dummy /dev/null TNs to help VOPs recognize unused results? 21:19:03 Do we have the SC-NUMBERs for it? 21:19:56 I was actually thinking of representing as extra offsets? 21:19:59 It'd certainly be nice for some VOPs to be able to stop emission after their side-effect. 21:20:52 it's either that, or defining VOPful setters as returning (values), and wrapping them appropriately. 21:22:39 That's almost worse, you know? 21:22:57 really? 21:23:34 (how?) 21:23:49 Yeah, it basically pushes the load up a level. 21:24:13 mm. right, potential regalloc issues. 21:25:01 Actually, on x86 and x86-64, we apparently have a NOISE SB used by an IGNORE-ME SC. 21:25:33 It's unique to the x86oids, though. 21:25:34 It would be pretty bad to have representation selection badness on (let ((tmp [double-float value])) (%void-aset vector index tmp) tmp) 21:26:49 -!- stassats [~stassats@wikipedia/stassats] has quit [Ping timeout: 240 seconds] 22:32:04 -!- tcr1 [~tcr@217-162-131-235.dynamic.hispeed.ch] has quit [Quit: Leaving.] 23:09:22 phil [~hargettp@96.237.127.144] has joined #sbcl 23:11:28 -!- hargettp [~hargettp@dhcp-162.mirrorimage.net] has quit [Remote host closed the connection] 23:15:38 -!- phil [~hargettp@96.237.127.144] has quit [Quit: Linkinus - http://linkinus.com] 23:15:45 hargettp [~hargettp@96.237.127.144] has joined #sbcl 23:19:17 Is it realistic, in SBCL as it stands today, to expect that an escaped closure function will only be called synchronously, given the existence of SIGALARM and related asynchronous events? 23:20:51 -!- ilmari [ilmari@knuth.ping.uio.no] has left #sbcl