Geth MoarVM/master: 7 commits pushed by (Timo Paulssen)++ 00:04
timotimo next up: stuff to explore frames with? 00:33
i don't think i want to write bytecode-dumping in python 00:34
good thing we already have that for --dump
let's see if i can hook that up somehow 00:36
MasterDuke timotimo: is this script loaded automatically if you use perl6-gdb? 00:43
timotimo yes, but multiple caveats 00:44
1) you have to copy or link it into the same folder as the "moar" binary
2) you have to whitelist it
i think that's all
MasterDuke how do you whitelist it?
timotimo gdb tells you how to do it 00:45
add-auto-load-safe-path /home/timo/perl6/moarvm/tools/moar-gdb.py
add-auto-load-safe-path /home/timo/perl6/install/bin/moar-gdb.py
this goes into ~/.gdbinit
in theory we can put the python script into the binary with a piece of linker script or something 00:46
i don't want to bother factoring out the stuff in bytecodedump so i can call it for individual frames 00:50
maybe later
what else ...
cat on lap 00:52
cannot continue working
okay, cat released me 01:02
i guess moar-gdb could teach gdb how to find fields in a P6opaque 01:03
and others
like CStruct and CPPStruct
MasterDuke bigints? 01:05
timotimo good point 01:06
Geth MoarVM: 1870340d84 | (Timo Paulssen)++ | tools/moar-gdb.py
todo entries for pretty printing some of our REPRs
 01:11
timotimo i don't see a way to provide a display hint of "no hint" 01:42
though mabe i can invent a value for "no hint" and just shrug it off 01:44
MasterDuke what are display hints? 01:50
timotimo sourceware.org/gdb/onlinedocs/gdb/...g-API.html
for some reason recompiling my moar made rakudo instacrash when starting 02:21
i just turned debugging down :\
i've made a bit of progress with an mvmobject pretty printer 02:33
MasterDuke nice. i've made a bit of progress fixing int-uint tests, getting the fudging correct for both moar and jvm was annoying 02:34
timotimo pointer to None (IterationBuffer) of repr VMArray<error reading variable: Cannot access memory at address 0x18> 02:35
doesn't that sound like a whole lot of fun?
MasterDuke address 0x18? not much lower to go 02:36
timotimo clearly an offset from a null pointer :\
i'll go to bed now, continue this tomorrow 02:39
MasterDuke later...
timotimo laters! 02:40
goo dluck with the tests :)
MasterDuke thanks, think they're done, just created PR so some more eyes look them over 02:46
02:48 ilbot3 joined 04:13 brrt joined 04:16 vendethiel joined 04:59 vendethiel joined 06:19 stmuk_ joined 06:52 domidumont joined 07:38 brrt joined 07:42 lizmat joined
brrt good * #moarvm 07:52
so.
I was wrong about a rather central assumption, haha
but 07:53
the good bit is
it's really simple to change
but…. i recently talked about having ARGLIST and CALL as sepeate or joined tiles 07:58
so now it turns out they've already been joined
question is, do i want to keep it that way
hmmm 08:24
no
i do want to distinguish between refs to ARGLIST and refs to CALL
08:30 zakharyas joined 08:44 brrt joined
jnthn moarning o/ 09:50
dogbert17_ o/ 09:52
dogbert17_ debugging an ENOCOFFEE problem
brrt moarning jnthn, dogbert17_ 09:53
dogbert17_ o/ brrt
10:02 brrt joined
dogbert17_ so what's on the agenda today, SEGV's, memory leaks or CONC? 10:05
brrt CALL/ARGLIST worries 10:06
jnthn Having got spectest6 now running reliably, I'm breaking off the SEGV/leak fixing for a bit and giving some attention, at long last, to the plight of .hyper and .race 10:09
dogbert17_ are they totally broken or only partially so? 10:11
jnthn Pretty broken 10:12
Essentially, my prototype to figure out API made it into Rakudo proper :) 10:13
nwc10 when is the Rakudo release?
jnthn Saturday 10:17
nwc10 excellent 10:19
and then Dogfood? :-)
jnthn Make spectest6 the default? Can certainly look in to that 10:21
nwc10 I think that that would be cool, but after the release 10:24
jnthn Yup
I should try it on Windows before that too, to be sure it's OK there 10:25
In theory it'll be faster than spectest5 there :)
nine timotimo: in Python you have to str(foo) all the time... 10:26
yoleaux2 14 Mar 2017 14:41Z <IOninja> nine: does 6.c-errata need any updates for require? t/spec/S11-modules/require.t is failing
15 Mar 2017 12:00Z <brrt> nine: if he had any plans on going to polyconf
15 Mar 2017 12:32Z <IOninja> nine: all the comments in this file say "return blah *type object* if none..." but all the codes return Nil. Is the code wrong or are the comments wrong? github.com/rakudo/rakudo/blob/nom/.../Handle.pm
15 Mar 2017 14:10Z <IOninja> nine: never mind. ugexe++ found the answer in github.com/rakudo/rakudo/commit/​8...1d673ae5a4 and I updated the comments
IOninja hah
nine woah
IOninja nevermind mine, it's no longer relevant
nwc10 jnthn: yes, as long as you have >1 core :-)
jnthn 4 real, 8 virtual
IOninja FWIW, spectest6 was slower on my 24-core VM. AND it still has the issue that its test count is off from stectest5, so I'd be cautious with making it the default just yet.
nine has at a 17 hour workday yesterday and is not entriely recovered yet 10:27
jnthn o.O
Ouch
IOninja: Slower on Linux isn't entirely surprising
IOninja nine: oh wait, one of mine is still relevant. There are 6.c-errata failures in t/spec/S11-modules/require.t
jnthn: but supposed to be faster on Windows? 10:28
nine IOninja: yeah, I need to look at those
jnthn IOninja: It's liable to be faster on Windows because spectest5 sets off tests in batches of TEST_JOBS, waits for those to complete, and then sets off the next batch, so if there's a slow test anywhere then it blocks progress on everything
IOninja Ah 10:29
10:46 geekosaur joined
samcv jnthn, but it only does that on windows? it's different on linux? 10:58
timotimo yeah, it's different
also good morn' 10:59
i think i want to build a "mvmbreak" command for moar-gdb.py; you'd "mvmbreak push_o" and it'd create a breakpoint inside interp.c at the right place
samcv that seems silly. there must be a reason i guess
timotimo hm. google also accepts "microsoft" when you search for "windows" 11:02
jnthn samcv: I think it's because it's hard, without having real async support, to do better on Windows 11:04
11:20 brrt joined 11:32 geekosaur joined 11:35 geekosaur joined
nwc10 reason - MS failed to implement socketpair() the way Berkley did 12:05
or more, close 12:06
er, implicit close
implicit close at process exit (on any other platform I think) will cause pending socket data not yet sent to be sent
Win32 drops it on the floor 12:07
this behaviour is not orthogonal.
this behaviour is not the way the de-facto spec (BSD) did it
this behaviour is very annoying
timotimo yoloooooo~ 12:08
nwc10 oh, and on Win32 perl *can* use select on sockets (and hence the product of socketpair)
but not on the IPC mechanism (pipes, I think) that doesn't have this "drop it on the floor" feature.
geekosaur welcome to the utter hack that is winsock 12:09
BSD sockets are kernel space
nwc10 aha, I can see where this is going
and I didn't this
but it explains a lot
geekosaur winsock is user space, the windows process termination functions do not in general permit a process to block termination or perform cleanup
nwc10 didn't $verb this where $verb was roughly "Realise"
aha. yes 12:10
because on *nix (and I think VMS, MVS, and therefore onwards RISC OS, AmigaOS ...)
the sockets are in the kernel
geekosaur yep
nwc10 so the user process exits
geekosaur Windows had to be different :/
nwc10 and the kernel "deals" with it
(sure, your data might never actually arrive - you quit, your fault you don't get to see any error message)
geekosaur heck, until NT4.0 killing a process could cause it to permanently leak memory 12:11
(that's right, the kernel did not automatically clean up process memory, the process had to clean up itself --- and this again failed if forcibly terminated...)
ilmari wut 12:12
timotimo huh, that's ... interesting 12:13
we have to make sure moar on pre-nt4 will --full-cleanup by default! 12:14
geekosaur that was understandable in old-style windows. NT3.1 and 3.5? waaaat?
(not that anyone cares about those any more. I dearly hope at least. I have heard there's still some win2000 deployed in various places, but that was at least mostly sane....) 12:15
ilmari perl5 supports win2k and later 12:16
timotimo i was quite amused recently when i read about the "three_finger_flag"
"If you are running under DOS this is the three finger salute, ctrl+alt+del. Most multitasking OS's will trap this combination before it reaches the Allegro handler, in which case you can use the alternative ctrl+alt+end." 12:17
dogbert17_ also note that harness6 is broken (leaks sockets) if you run it with TEST_JOBS=1 (the default) 12:29
m: say [+] (1..100).race # RT #130576 12:31
synopsebot6 Link: rt.perl.org/rt3//Public/Bug/Displa...?id=130576
camelia 0
dogbert17_ m: say [+] (1..100) 12:32
camelia 5050
geekosaur mm, actually i think the NT situation was a bit more complex. like you could register a cleanup handler, but if TerminateProcess got invoked while in the cleanup handler, cleanup aborted and that was what leaked
and some ways of terminating a process didn't let the cleanup handler run before NT4
nwc10 and the poor machine would eventually die from memory starvationg? ... 12:34
gah, bad fingers
anyway,
nwc10 reboots ilmari to prevent starvation
ilmari nwc10: typo of the day: perldoc -> perldog. clearly a sign I should go to brewdog for lunch
nwc10 :-)
ilmari they unfortunately don't have my favourite burger, the melted ego, any more :( 12:35
but they have a new one I want to try
15:58 MasterDuke_ joined
MasterDuke_ jnthn: not to distract from hyper/race, but have you had a chance to look at github.com/MoarVM/MoarVM/pull/551 ? 16:03
yoleaux2 3 Mar 2017 15:29Z <brrt> MasterDuke_: it's a one day job
[Coke] misread that as a "day one job"
MasterDuke_ heh. and i need to go back to the logs to see what the question was 16:04
timotimo the good thing about GC is that the more objects are already dead when we GC, the faster it is 16:05
MasterDuke_ ah right, about getting his JIT bisect tool to work with the current JIT 16:07
timotimo oh?
MasterDuke_ i asked it was a 5min or 5day job 16:08
this was when we were still trying to figure out that div_i bug
timotimo it really depends on what we want to achieve with it 16:09
jnthn MasterDuke_: Yeah, it looked better. I need to look more closely again though. Also I'm a bit worried about merging it before the weekend release, just in case it has any unintended consequences.
timotimo if we're fine with per-frame granularity, it'd be possible in like 30 minutes
if we want per-instruction granularity ... good luck!
MasterDuke_ jnthn: cool, just wanted to make sure it didn't slip through the cracks 16:12
jnthn, timotimo: while on the subject of reviewing things. i think github.com/perl6/roast/pull/254 makes the (int|uint)(8|16|32|64) tests better (and i think i have a fix for one or two of the todos in the works), but a second opinion would be appreciated 16:14
timotimo jnthn: do you think it'd be cleverer to write some code for "we don't use a normalizer at all" or to write "a normalizer that does nothing"? 16:37
i'm in a bit over my head, it feels like 16:42
jnthn Probably the "normalizer that does nothing" will require a lot less special cases :) 16:43
timotimo thought so, too 16:45
is decoderconfigure the right place to put the choice of normalization form?
jnthn Uh...hmm 16:47
Where are you doing to use it?
Thing is, at the moment Uni is a VMArray
timotimo i wasn't finding ops i could use to go from blob to vmarray 16:48
decodetocodes and encodefromcodes are marked NYI 16:49
jnthn ah, right
timotimo maybe that's where i should be looking instead for the null normalization bit?
jnthn yeah, those want to be I'd :)
timotimo "IRN" :)
jnthn Yeah, though the decoders still want a normalizer iirc
timotimo OK
we already have ops to go from a decoder ot a vmarray?
for decodestreams i mean 16:50
"decodertakebytes"?
jnthn Yeah, but that's just going to give you back bytes 16:51
timotimo huh? bytes from the input buffer? 16:52
jnthn I think the big question we really need to figure out is whether MVMString wants to be able to support codepoint level things
timotimo because after that there are no "bytes" any more :P
jnthn Yes, it's there to support mixed-mode text/byte I/O when we re-work IO::Handle to use the Decoder REPR
timotimo oh
allowing MVMString to do stuff that's not it NFG sounds like a "too many things on the same peg" situation 16:53
i.e. every string op would start off with "okay do we have to emulate grapheme semantics ourselves now?"
except if they already work with a grapheme iterator
hm. we probably already have a grapheme iterator path for almost all ops because of ropes 16:54
jnthn Yeah, we need to try and make sure we can lift out those kinds of checks
At specialization time
(the level ones)
I've been pondering how we're ever going to get the regex engine to accept Uni for example 16:55
And...if it's a VMArray that's gonna be a small nightmare
timotimo oh, yeah, that's true 16:58
oh, if the string itself has to be normalized from the start, we won't be able to match subsections of a regex in a different normalization mode?
or do we re-normalize the whole haystack and move to the right offset with an iterator? 16:59
jnthn All of these questions are why we kicked the Uni can down the road some ;-) 17:00
timotimo i can imagine :|
jnthn I think various operations are going to have to be OK with needle and haystack being NFG when the needle is Uni or NFC or whatever though
There's stuff in S05 for saying what level you want to work at 17:01
timotimo mhm
jnthn so (example) /:nfd \c[COMBINING ACUTE]/ 17:02
timotimo okay so a possible way forward from here is: i implement a do-nothing-normalizer, it'll turn the Decoder into a mode where instead of Str it'll create Uni objects
jnthn We can try a path like that for now, yeah
timotimo and once we figure out if MVMString gets a non-nfg mode-of-operation, we'll be able to implement :norm(NF-anything-but-G) to create Str, too
jnthn I suspect we'll need to experiment a bit before we hit upon the right design
Well, it won't create Str 17:03
It's rather than Uni will become something with repr VMString or some such
Or a wrapper around one like Str is
timotimo oh
jnthn But I think I'd like to then encode the level at REPR-data level 17:04
So we can spesh on it
timotimo right
jnthn Well, normalization more than level
So yeah, if we want to really Do This Right then that might be a logical first step
timotimo what would you consider a sane enum value for NULL normalization?
0 is already NFD 17:05
(because of the bitmask-backed design)
jnthn Hm, we can always shift them right one or something 17:06
timotimo so 1 would be null normalization and it'd mask all other bits in comparisons and all others would have a 0 LSB? 17:08
jnthn #define MVM_NORMALIZE_COMPAT_DECOMP(form) (form & 1) 17:09
#define MVM_NORMALIZE_COMPOSE(form) (form & 2)
#define MVM_NORMALIZE_GRAPHEME(form) (form & 4)
So those would become 2, 4, 8
And then the LSB can be a normalize bit
So NFG would be 8 + 4 + 1
timotimo ah, so more like only normalize if LSB is set 17:10
jnthn Yeah
Righty, I should go rest a bit then make kebabs :) 17:13
timotimo mhhh kebabs 17:14
17:15 domidumont joined 17:20 domidumont joined 17:27 zakharyas joined 17:30 domidumont joined 17:31 lizmat joined
timotimo so, MVM_string_decodestream_create isn't public API, right? because it's not MVM_PUBLIC? so it'll be fine to put a normalization number parameter in 18:00
how come we don't have an "outside-world normalization form code" for NFG? 18:03
NFC, NFD, NFKC, and NFKD in that order. i'll put NULL at the front (where it can have 0)
i wonder what the NYI "encodenorm" op will do 18:05
19:03 agentzh joined 21:40 ggoebel joined
samcv oh so we're coming up with more normalization things. fun! 21:49
21:56 MasterDuke joined
jnthn samcv: Well, mostly figuring out how we can properly support codepoint-level strings 22:13
(Uni and the various subclasses)
For those with use-cases taht need 'em
22:18 TimToady joined
Geth MoarVM/null-normalization: 922d709f3c | (Timo Paulssen)++ | 11 files
WIP on adding a normalization that does nothing
 22:57
timotimo samcv: are you interested in picking up my WIP? i have no previous experience inside the whole encoding/decoding business
MasterDuke heh. like the red queen, so much work to do nothing 23:00
timotimo in what card game is the red queen supposed to do nothing? :P
MasterDuke off with your head! 23:01
23:54 Geth joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!