|
01:56
ilbot3 joined
02:17
ggoebel joined
03:03
bloatable6 joined,
coverable6 joined,
nativecallable6 joined,
quotable6 joined,
evalable6 joined,
bisectable6 joined,
greppable6 joined,
benchable6 joined,
releasable6 joined,
unicodable6 joined,
squashable6 joined,
statisfiable6 joined
05:55
evalable6 joined
06:05
committable6 joined,
committable6_ joined
06:07
committable6 joined
06:09
committable6 joined
07:12
AlexDaniel- joined
07:33
domidumont joined
07:39
domidumont joined
09:05
rba_ joined
09:16
rba joined
09:31
rba_ joined
09:47
rba joined
10:01
rba_ joined
10:16
rba joined
10:31
rba_ joined
10:46
rba joined
11:02
rba_ joined
11:16
rba joined
11:31
rba_ joined
11:46
rba joined
12:01
rba_ joined
|
|||
| Geth | MoarVM: a9267cb67f | (Timo Paulssen)++ | src/core/interp.c don't allow low-level null from getdynlex |
12:10 | |
| timotimo | ^- RT #132300 | 12:11 | |
| synopsebot | RT#132300 [new]: rt.perl.org/Ticket/Display.html?id=132300 [SEGV] say nqp::getlexdyn('&DEPRECATED') | ||
| jnthn | timotimo++ | 12:12 | |
| timotimo | i'm going to check the other functions of that type, too | ||
| there's one that explicitly stores NULL if the result from the function in question was null | |||
| jnthn | Stores NULL in a register? | 12:13 | |
| We should never do that | |||
| timotimo | m: use nqp; say nqp::getlexrel(nqp::ctx(), "nope") | 12:14 | |
| camelia | (signal SEGV) | ||
| timotimo | m: use nqp; say nqp::getlexreldyn(nqp::ctx(), "nope") | ||
| camelia | (signal SEGV) | ||
|
12:16
rba joined
|
|||
| Geth | MoarVM: 650e7976f0 | (Timo Paulssen)++ | src/core/interp.c no bare null from getlexrel or getlexreldyn either |
12:16 | |
| timotimo | they could need changes in the jit, too, though | ||
| interestingly, only getlexdyn has that problem | 12:18 | ||
| nice. the only two usages of getdynlex check the result for null anyway, so we can just as well move the check to the func that does the work | 12:21 | ||
| Geth | MoarVM: 676723d736 | (Timo Paulssen)++ | 2 files move null check into getdynlex func for benefit of jit the return value was already checked for null by every user except the jit, so just have the function do it. |
12:23 | |
| MasterDuke | btw, this is the link i was looking for earlier www.moarvm.org/measurements/perl6-bench/ | 12:29 | |
| timotimo | hasn't run in ages | ||
| MasterDuke | there was a period where new files weren't being produced at all. they are now, but are empty | 12:30 | |
| dogbert17 stumbles on a SEGV /o\ | |||
| or more specifically asan barfage, i.e. no SEGV | 12:31 | ||
| gist.github.com/dogbert17/89a3383f...10317e1a5a | |||
|
12:31
rba_ joined
|
|||
| timotimo | wat, accessing a destroyed threadcontext? | 12:32 | |
| MasterDuke | where do the process that creates those benchmarks live? | ||
| timotimo | that's japhb's perl6-bench | 12:33 | |
| jnthn | It was just a cron job running perl6-bench and sticking the results somewhere | 12:34 | |
| Not sure what broke it | 12:35 | ||
| MasterDuke | where's the cron job? | ||
| timotimo | jnthn's system | ||
| dogbert17 | timotimo: if it's any clue the barfage disappears if I refrain from setting MVM_GC_DEBUG=2 | 12:36 | |
| timotimo | i wonder why it'd die checking a register assignment, but not actually use whatever's in that? | ||
| dogbert17 tries with MVM_GC_DEBUG=1 | 12:37 | ||
| with MVM_GC_DEBUG set to 1 it fails even earlier ! | 12:39 | ||
| ok 12 - Threads got different IDs | |||
| #0 0xb55d379a in bind_pos /home/dogbert/repos/rakudo/nqp/MoarVM/src/6model/reprs/VMArray.c:387 | |||
| #1 0xb55bb140 in MVM_repr_bind_pos_o /home/dogbert/repos/rakudo/nqp/MoarVM/src/6model/reprconv.c:232 | |||
| #2 0xb57646a0 in MVM_spesh_stats_cleanup /home/dogbert/repos/rakudo/nqp/MoarVM/src/spesh/stats.c:645 | |||
| #3 0xb575c317 in worker /home/dogbert/repos/rakudo/nqp/MoarVM/src/spesh/worker.c:13 | |||
| heap-use-after-free | |||
| timotimo | what :) | 12:40 | |
| dogbert17 | bizarre, with MVM_GC_DEBUG=0 the problems disappear | 12:41 | |
| MasterDuke | didn't jnthn fix some bug in debug code recently? | 12:42 | |
| dogbert17 | I think so yes | ||
| MasterDuke | wonder if the problem is actually in MVM_GC_DEBUG!=0 code | 12:43 | |
| dogbert17 | that would be my noobish guesss | ||
| guess this should be reported | 12:45 | ||
|
12:46
rba joined
|
|||
| dogbert17 | I am running with a very small nursery, i.e. 32768 * 2 | 12:46 | |
| ok, reported as github.com/MoarVM/MoarVM/issues/728 | 13:00 | ||
|
13:01
rba_ joined
|
|||
| dogbert17 tries to repair a broken Roland TR-707 instead & | 13:03 | ||
|
13:16
rba joined
13:31
rba_ joined
13:46
rba joined
14:02
rba_ joined
14:17
rba joined
14:31
rba_ joined
14:47
rba joined
14:58
zakharyas joined
15:01
rba_ joined
15:47
quotable6 joined,
coverable6 joined,
nativecallable6 joined,
evalable6 joined,
bloatable6 joined,
greppable6 joined,
committable6 joined,
unicodable6 joined,
benchable6 joined,
releasable6 joined,
bisectable6 joined,
squashable6 joined
16:01
buggable joined
16:22
domidumont joined,
rba joined
16:32
rba_ joined
17:45
evalable6 joined,
statisfiable6 joined
|
|||
| japhb notices he's been flagged ... | 18:16 | ||
| What's up with perl6-bench? | |||
| timotimo | just that we used to have daily results up on moarvm.org, but that has stopped at some point | 18:17 | |
| japhb | Huh. Bitrot? I haven't patched it in quite a while, I suppose it could have fallen prey to a deprecation or so. | 18:18 | |
| timotimo | i'd expect it's something not quite right with jnthn's setup | 18:20 | |
|
18:43
nebuchadnezzar joined
|
|||
| samcv | timotimo, do you know how long testing takes to run? | 18:45 | |
| timotimo | you mean spectests? | 18:50 | |
| samcv | the things you are citing. perl6-bench | 18:54 | |
| timotimo | oh | ||
| ironically, the faster your code is, the longer it'll take in total :P | 18:55 | ||
| i haven't run it in a long time | |||
| it starts with a tiny piece of work and increases the workload until a threshold is reached | |||
| perl6-bench has this nice tool that builds versions of rakudo and perl5 and it's all batch-friendly | 18:57 | ||
|
20:34
MasterDuke joined
21:57
MasterDuke_ joined
22:31
MasterDuke joined
|
|||
| MasterDuke | jnthn: is there an easy way to know if a method gets inlined? | 22:42 | |
| jnthn | MasterDuke: --profile | 22:52 | |
| MasterDuke | `In total, 5316427 call frames were entered and exited by the profiled code. Inlining eliminated the need to create 34691 call frames (that's 0.65%).` | 22:58 | |
| so i doubt it for the method i'm interested in, but i don't see specific information about inlining in the Routines tab | 23:00 | ||
| jnthn | It's in the call stack tab iirc | 23:01 | |
| So really it tells you about whether it was inlined at a particular callsite | 23:02 | ||
| MasterDuke | ah, so it is. `Calls (Inlined) NaN + (NaN%) ` | ||
| jnthn | o.O | 23:07 | |
| Sleep time o/ | |||
|
23:53
evalable6 joined
|
|||