00:12 sena_kun joined 00:14 Altai-man_ left 02:12 Altai-man_ joined 02:14 sena_kun left 02:15 MasterDuke left 04:13 sena_kun joined 04:14 Altai-man_ left 05:07 vrurg left 05:08 vrurg joined 05:15 Kaiepi left 05:32 Kaiepi joined
nwc10 good *, #moarvm 06:04
06:12 Altai-man_ joined 06:14 sena_kun left 06:44 patrickb joined 06:54 domidumont joined 07:27 leont joined
nwc10 MasterDuke: is that warning new (and not on master)? 07:29
tellable6 nwc10, I'll pass your message to MasterDuke
07:58 zakharyas joined 08:13 sena_kun joined 08:14 Altai-man_ left 08:29 MasterDuke joined
MasterDuke . 08:30
tellable6 2020-07-31T07:29:55Z #moarvm <nwc10> MasterDuke: is that warning new (and not on master)?
MasterDuke nwc10: yeah, not on master 08:31
nwc10 OK, interesting/odd as it's inside a macro that I didn't change. 08:33
I'm not sure what flags I need to enable to see those. Do we get the same warnign about aliasing violations in other parts of the code? Or is that a first? 08:34
$ork thing is being silly/slow today. But it's SEP from today because I'm on holiday.
MasterDuke you don't have -Wstrict-aliasing set?
nwc10 (actually going away for 2 nights in the middle of the 2 weeks in the garden) 08:35
I guess that I don't
MasterDuke i don't remember seeing one of those before
nwc10 (I do on other $ork code)
MasterDuke well, i know i sometimes see different things when building on my desktop vs laptop
nwc10 I shall try to figure this out after $ork 08:36
MasterDuke different gcc versions, etc
nwc10 well, more, after I am not actively having to figure out $ork
MasterDuke even though i'm using the exact same configuration/commands to build moarvm
nwc10 at some point today (preferably yesterday) I should migrate to the my regular "about to be AFK for a while" plan - fixing documentation 08:37
MasterDuke always a thrill ride 08:38
nwc10: but i can repro what dogbert17 said earlier, rakudo build times were a little faster on your branch 08:41
nwc10 good.
I estimated it at about 4%. (No-one else seems brave enough to put a number on it. I was hoping I could go second on numbers) 08:42
which you can take as "our hashes were serously L.T.A>"
lizmat: but don't quote me on that ^^^ instead, I'd suggest something closer to "so much awesomeness has been showered down upon MoarVM that every other bottleneck has been fixed - hence hashes only became "slow" because all their competition had been eliminated" 08:45
MasterDuke stage parse master is 39.342, XXX-better-hash is 36.251
lizmat nwc10: I guess a lot of work was put into making sure they were safe from algorithmic complexity attacks ? 08:46
nwc10 I'd like someone to review that, but 08:47
a) I believe it's as safe as our previous implementation
MasterDuke no obvious difference in spectest time
nwc10 b) these things are overblown. (and I believe that I *am* quailified to say that)
lizmat I know it scared the heck out of people at former $work
as they got routinely attacked by a *lot* of people 08:48
nwc10 that was possibly the second round of fun
MasterDuke nwc10: i'm not qualified to really review your branch, but i did notice a bunch of malloc/free instead of MVM_malloc/MVM_free. is that intentional?
nwc10 MasterDuke: no. that's not intentional. That's a good spot. Thanks
there will be a rebase at some poitn. Maybe as YYY- 08:49
and then ZZZ-
and then I usualyl run out of ideas (at least, based on $ork test commit names)
lizmat he... some of our earliest servers were called that :-)
nwc10 lizmat: IIRC Python 2.7 did *not* enable protection from randomisation *by default*, and I don't remember ever reading about anyone running django getting burned (etc) because of this 08:50
the other one was, having (I think) chatted with someone at said $ork
the two "obvious" general ways to attack an online system are
1) figure out expensive search queries for the read-only system
Geth MoarVM: MasterDuke17++ created pull request #1332:
Fix some new compiler warnings
nwc10 2) once you are in the transacting system (ie the "checkout" of a web store, or similar - anything that needs to write to the DB), do your DOS there 08:51
as most sites are largely read only, and that part can be easily scaled
but the read/write or write parts can't
I'm not *denying* that the complexity attacks could worjk
and I think I figured out how to attack Perl 5 (pre 5.16) in quite an interesting way that can't be mitigated 08:52
oh yes,
also what was found (the second time) was that the mitigations put in place for 5.8.2 could *themselves* be attacked
and actually it was scarily more effective to attack the mitigations than the original problem.
lizmat otoh, one time at $ork, we ddossed ourselves
nwc10 "just the once?" :-) 08:53
lizmat by having javascript errors sent back
and making an error with that, so that in some cases the client would just send back the same error continuously
take that times the number of clients and whee! 08:54
that was one of the few distributed ones :-)
nwc10 aspects of current $ork that I like - 08:55
1) we are not globally visible, so not a massive target
2) we don;t deal with credit cards (or anything transacting money in real time)
and (at least I thought)
3) it's on site, but easily commutable 08:56
09:13 MasterDuke left 10:12 Altai-man_ joined 10:14 sena_kun left 10:17 MasterDuke joined
Geth MoarVM: f6f12f5a9d | (Daniel Green)++ | 2 files
Fix some new compiler warnings

Remove an unused variable. Use right size in format string. Make variable const.
MoarVM: 0a06f80c70 | niner++ (committed using GitHub Web editor) | 2 files
Merge pull request #1332 from MasterDuke17/fix_some_new_compiler_warnings

Fix some new compiler warnings
11:25 zakharyas left 12:18 brrt joined
brrt \o 12:19
lizmat brrt o/ 12:22
brrt ohai lizmat 12:25
lizmat wonders what brrt is up to these days 12:30
brrt not enough sleep, for one thing :-D 12:33
lizmat well, wake up later, go to sleep earlier ? :-)
brrt :-) 12:37
$work, is mostly what I'm doing
nwc10 good *, brrt 12:44
brrt ohai nwc10 12:46
nwc10 brrt: I made a branch. (technically, I had already made a different branch). But this branch is better. 12:47
brrt where does the branch go to 12:52
nwc10 it got to MoarVM/XXX-A-Better-Hash 12:53
brrt hah
nwc10 and it has all the UT hash implemetation replaced with Robin Hood hashing
well, it's XXX because it has some XXX "FIXME" type comments 12:54
and I didn't want anyone to merge it yet
it is Safe For Work
unless, well,
brrt RH hashing was when you had a flat hash table and you'd swap entries to 'average' the search length, correct?
MasterDuke you can do a draft PR and then it won't be merged, but you get to look at it in with the github ui a little easier
nwc10 IIRC in some lightning advert at a not-called-YAPC::Europe I said something like
"if you like the idea of Perl 6 going faster, dontate to the P6CMF"
"and if you don't" [pause] 12:55
"donate to the P5CMF"
brrt lol
I find the GH UI to be not all that optimal
slow, for one
nwc10 I have 3 or 4 more commits to push
then I'll make a PR
er, "then" might be tomorrow morning
I'm at $ork
and we have visitors soon after
and there will be barbeque and beer
brrt hmm, no barbeque here today, nor beer 12:57
lizmat: hows' the heat wave treating you?
nwc10 *how* far are you from Belgium? And is that distance your excuse?
MasterDuke no bbq, yes beer
nwc10 and, hurrah, I read in the local free newspaper that the beer offer is on again at the supermarket nearby 12:58
so I must visit
lizmat Belgium: about 4km as the crow flies, 8km to the nearest bridge
nwc10 but tomorrow. or Monday. In the morning. When sober
(sadly the branch within walking distance no longer plays along with the beer offer)
brrt I am in fact very close to Belgium 12:59
nwc10 and Luxemburg? "Too close for comfort"?
brrt But the province I'm closest too (Antwerpen) is CLOSED FOR CORONA
nwc10 although I buy their explanation that they are full of (fithly) foreiners
visiting daily. to do work.
brrt so I could go but I'd have to spend 14 days in quarantine
nwc10 that's a bugger. doesn't sound like a good idea
brrt it sucks. 13:00
nwc10 hence why no beer? :-(
brrt no, the becasue of the no-beer is the fact that I'm the designated driver :-)
MasterDuke doesn't belgium have the highest deaths/million of population ratio?
brrt idk 13:01
nwc10 brrt: I'm glad that your group is sensible, but saddened to learn that no-one else takes their turn at it. 13:03
brrt Well, the fact that I'm the only one in the group that has a drivers license kind of forces the issue :-) 13:07
nwc10 they should learn!
brrt absoluteluy
nwc10 or they should set up a pub locally. 13:08
(not recommended. I understand it to be very hard work)
brrt indeed 13:11
I'm trying to find... where are my notes for the linear 'expression' JIT 13:12
nwc10 "behind the sofa?" is a usual faceitous answer. But are these disk files, or real bits of paper?
13:13 sena_kun joined 13:14 Altai-man_ left
brrt disk files 13:16
if they were bits of paper then I knew where they were 13:17
I still know the plan vaguely but I had it worked out to a considerable level of detail
nwc10 I thought that ripgrep was the newer shiny, replacing ag, replacing ack
so just leave a brute-force search of the disk running while you do your solem duty as designated driver? 13:18
lizmat brrt: to answer your question: 33.2 outside, 25.2 inside
outside == in the shade
nwc10 lizmat: that sounds very close to the numbers here 13:19
lizmat expectation was 36 here, but I don't think we're going to reach that
nwc10 #11955 When in doubt, use brute force.
13:20 domidumont left
brrt ripgrep is fairly shiny 13:20
I sometime doesn't quite do what I want it to, but usually it finds stuff 13:21
I had codesearch installed before, which has an index, and is faster overall
Geth MoarVM/XXX-A-Better-Hash: 53 commits pushed by (Nicholas Clark)++
review: github.com/MoarVM/MoarVM/compare/8...d2fca65b21
nwc10 that seemed to end up being a rebase, and then with new commits
I'm not sure what I rebased
aha: 13:45
-+ * (Doesn't allocate, so can't deadlock. */
++ * (Doesn't allocate, so can't deadlock.) */
might be fractionally faster, because it now pre-sizes the hashes for bytecode deserialisation and hash copying 13:46
but I don't know if that will make any measurable difference
dogbert17 nwc10: I did a git pull on my XXX-A-Better-Hash and I can't build. There's a merge conflict in src/core/str_hash_table.c 14:36
[Coke] dogbert17: I think that's just your copy 14:39
I just did a git pull --rebase, don't see any merge conflict markers.
dogbert17 and I did git pull only :( 14:40
[Coke] did you have local changes?
I'm at add2fca65 here in MoarVM, seems ok 14:41
dogbert17 I didn't have any local changes but I believe it's fixed now after some googling 14:43
[Coke] huh 14:47
14:50 brrt left 14:51 brrt joined
dogbert17 I did a 'git reset --hard <previously checked out commit> 14:53
15:02 zakharyas joined
nwc10 that was likely the "oops, I seem to have rebased it" 15:09
oh my, wireless is not as good as the CAT-5 15:10
(have unplugged the CAT-5 because we have small visitors)
only visiting to get another beer
[Coke] moshi moshi++ 15:11
nwc10: got some failures today on rakudo test 16:06
nwc10: gist.github.com/coke/cc744eade897a...14b93495ec
(os x)
16:12 Altai-man_ joined 16:14 sena_kun left
dogbert17 what might this be? 17:07
ok 16 - CAS on linked list with lexical head works (4)
MoarVM oops: insert conflict, 0x7fa1d72bfd18 is 114, 94727014987304 != 94727014987256
at gen/moar/Metamodel.nqp:1285 (/home/dogbert/repos/rakudo/blib/Perl6/Metamodel.moarvm:compute_mro)
happens intermittently when running t/spec/S17-lowlevel/cas.t 17:08
17:12 domidumont joined 17:27 brrt left 17:38 patrickb left 17:43 domidumont left 17:55 lucasb joined 18:05 zakharyas left 18:13 sena_kun joined 18:14 Altai-man_ left 19:39 brrt joined 20:12 Altai-man_ joined 20:14 sena_kun left 20:39 brrt left 22:12 sena_kun joined 22:14 Altai-man_ left 22:33 leont left 22:38 Kaiepi left 22:41 Kaiepi joined