Parrot 3.2.0 released | parrot.org | Log: irclog.perlgeek.de/parrot/today | Parrot is accepted for GSoC 2011! Student application deadline is Apr 8
Set by moderator on 27 March 2011.
00:03 woosh left
whiteknight NotFound: new syntax looks awesome. I can't wait to start using it in Rosella. All we need now are multis... 00:04
soh_cah_toa does nqp have any perl modules i could use for reading pbc? 00:06
i've been reading some perl6 articles to get used to nqp syntax but now i need to know how i can use w/ parrot 00:07
sorear you should use the Packfile PMCs for that 00:08
soh_cah_toa okay
but i don't know how to declar pmc variables in nqp. all i have to go on is perl6 resources 00:09
*declare
i asked around last night and it sounded like there really aren't many tutorials for nqp 00:10
00:10 kid51 is now known as kid51_at_dinner
whiteknight soh_cah_toa: NQP does have a test suite that you can mine for details 00:16
soh_cah_toa: or you can ask here and we'll help you
soh_cah_toa well, how would i declare a pmc variable in nqp? my $pmc .= new PMC ? okay, well how do i find out what methods can i use? 00:21
00:25 mikehh left
soh_cah_toa i mean, are there any nqp/perl modules i can dig through? ext/nqp-rx/t and ext/nqp-rx/src haven't been too helpful 00:26
00:27 theory joined
whiteknight soh_cah_toa: two ways 00:28
if you define the type in NQP code, it has a protoobject. So if you have a class My::Class, you create a new one as My::Class.new()
bacek_at_work soh_cah_toa, my $pmc := pir::new("Foo")
soh_cah_toa, for parrot-only pmcs
soh_cah_toa alright
whiteknight if you have a type built-in to parrot, or defined in an alternate object model, do what bacek says
soh_cah_toa alright, suppose i say: my $pmc := pir::new("pmc"); how can i find out what methods i can call? 00:30
00:32 lucian left
soh_cah_toa i'm starting to doubt whether i can use nqp for the debugger b/c there's no decent tutorial for it 00:34
bacek_at_work soh_cah_toa, for pir::new("foo") semantic is purely parrot-driven. Which mostly mean RTFS of foo.pmc :( 00:37
00:39 mikehh joined
soh_cah_toa alright, so the pmc's i can use in nqp are defined in src/pmc? 00:39
bacek_at_work soh_cah_toa, yes. 00:40
soh_cah_toa alright, now that is what i was looking for. thanks
bacek_at_work soh_cah_toa, it's also applicable to pir, winxed, etc. 00:41
soh_cah_toa i was just going to ask that. okay, good 00:56
what's a singleton pmc? 00:57
plobsing what indeed? 00:58
it is an old strategy to implement a singleton pattern in PMCs (every instance is actually the same instance). It has recently come underscrutiny for being crufty. 00:59
soh_cah_toa i'm not surprised. i see no practical advantage for that 01:00
dalek rrot/jit_prototype: b20b8dd | bacek++ | compilers/opsc/src/Ops/JIT.pm:
Use less pir.
 01:03
sella: 7f61f41 | Whiteknight++ | s (2 files):
Add in an alternate method for making decorators. Inheriting from Rosella.Decorate.Decorator allows transparent find_method fallback to a decorated target object. In essence, it treates the target object like a super-object for find_method
 01:10
sella: 11e9b77 | Whiteknight++ | src/decorate/Decorator.winxed:
remove cruft from Decorator. Implement stubs for get_attr_str and set_attr_str. These don't work because we aren't walking the mro. Object.find_method walks the MRO, get/set_attr_str do not (go figure, inconsistency?). Refactor find_method, keeping in mind that we can no longer do direct attribute access
sella/gh-pages: 31e3323 | Whiteknight++ | index.md:
Add note about future-proofing being a design goal of Rosella
 01:13
01:15 kid51_at_dinner is now known as kid51 01:27 jsut joined 01:28 whiteknight left 01:32 jsut_ left 02:13 kid51 left 02:32 petdance joined 02:43 bubaflub left 02:44 soh_cah_toa left
cotto petdance, have you started playing with the new pcc 1.0 release? 03:09
petdance no
what's pcc?
cotto portable c compiler
their site seems to be down atm 03:11
mikehh petdance: I have tried to use splint a couple of times recently on parrot - failed - any suggestions? 03:13
petdance How do you mean you failed? 03:14
cotto mikehh, if it's a simple fix to make it build, feel free to commit. Otherwise, you can file a pcc bug.
apparently it chokes on libav too 03:15
mikehh petdance: every time I tried it reports something like cannot find some ffi item and exits
petdance Did you run "make splint"? 03:16
mikehh yes and tried other variants
petdance If you want me to help you, we can work on it. 03:17
mikehh include/parrot/parrot.h:82:19: Cannot find include file ffi.h on search path: ./include;./include/pmc;include/pmc;compilers/ast;/usr/include;/usr/include 03:18
Preprocessing error. (Use -preproc to inhibit warning)
petdance How did you install splint?
What OS are you on?
Please nopaste the entire output. Or make a gist. 03:19
mikehh with apt-get I think
nopaste "mikehh" at 192.168.1.3 pasted "output from make splint" (37 lines) at nopaste.snit.ch/39436 03:21
dalek Heuristic branch merge: pushed 900 commits to parrot/lorito by cotto 03:22
benabik That's a lot of commits. 03:24
cotto Happily that branch doesn't touch much. Git handled it like a champ.
mikehh petdance: that's on Ubuntu 10.10 i386 - perl 5.12.3 - gcc 4.5
cotto git++
petdance oh 03:25
you have to run a full make first, mikehh
03:25 jrtayloriv joined
mikehh I am pretty sure I did 03:26
but I will try again without extra options
petdance This suggests you did not: find: ‘src/{gc,io,packfile,string}/*.c’: No such file or directory
mikehh yeah, don't know what that means 03:27
dalek rrot: cce61fa | petdance++ | compilers/imcc/cfg. (2 files):
Made blocks_are_connected be static, and localized a variable
petdance It means it oculdn't find the .c files to check
mikehh date && time perl Configure.pl --test --configure_trace 03:28
sorry wrong window 03:29
petdance: I did make realclean, config as above, make world, make splint - got the same result as the nopast 03:37
petdance huh
what happens if you just run that find statement from the command line?
mikehh get a list of files 03:39
find src/*.c src/{gc,io,packfile,string}/*.c -type f 03:41
src/core_pmcs.c
src/datatypes.c
src/debug.c
thru
src/string/encoding.c 03:42
petdance 00 03:48
0HMM
ARGH
SO wonder why that doesn't work under make. 03:49
what if you run that command directly from the command line?
mikehh ok - I'll give it a try
petdance or in a shell script, for simplicity 03:50
I'm looking at pcc right now 03:52
cotto petdance, you can access their site? 03:53
petdance Nope
cotto ok. I thought my isp was insane for a moment.
benabik It might be the shell make uses doesn't recognize {a,b,c} 03:55
mikehh petdance: does not have that Cannot find bit but fails in the same way
include/parrot/parrot.h:82:19: Cannot find include file ffi.h on search path: ./include;./include/pmc;include/pmc;compilers/ast;/usr/include;/usr/include
Preprocessing error. (Use -preproc to inhibit warning)
Preprocessing error for file: /home/mhu/t.parrot/src/core_pmcs.c
< more preprocessing . In file included from src/core_pmcs.c:10
*** Cannot continue.
petdance benabik: That's pretty likely! 03:56
I'll get rid of it
benabik++
mikehh the file ffi.h is in /usr/include/i686-linux-gnu/ 04:00
dalek rrot: e938612 | petdance++ | config/gen/makefiles/root.in:
Don't use curly brackets
 04:02
petdance mikehh: try pulling that and ee if it works better
mikehh petdance: 'k I'll give it a go - I somehow need to get etc/ld.so.conf included in my build -> include /etc/ld.so.conf.d/*.conf 04:11
petdance That doesn't sound right, but ok 04:14
bbiab 04:15
04:15 petdance left 04:21 petdance joined
mikehh seriously need to sleep = it's getting close to 5:30 am here 04:26
04:38 ShaneC left
dalek rrot: 114d897 | petdance++ | src/embed/api.c:
Use STREQ() for string comparisons, instead of incorrectly using strcmp as a boolean
 04:58
05:37 petdance left 05:40 contingencyplan left
cotto dukeleto, ping 05:42
dalek rrot/lorito: a31a6bd | cotto++ | src/dynoplibs/ (3 files):
add M0 prototype dynops, now with 75% more sharp edges
cotto seen dukeleto
aloha dukeleto was last seen in #perl6 2 hours 38 mins ago joining the channel.
cotto msg dukeleto I'm in your branch, prototyping stuff. 05:49
aloha OK. I'll deliver the message.
benabik cotto: Are we just implementing M0 in PBC now? That'll be a fun bootstrapping step... PBC on M0 on PBC. 05:50
05:50 jrtayloriv left
cotto benabik, that's just to get a handle on how close to complete it is 05:50
benabik cotto: You're fibbing. It's PIR all the way down. :-D
cotto I could be. I'm never quite sure. 05:51
05:51 contingencyplan joined
cotto I just want something runnable for now. 05:52
benabik Prototypes are good. I was just reading a paper that said to plan to throw the first implementation away since you'll probably have to do it anyway. :-) 05:53
cotto benabik, exactly 05:55
eventually something has to be useful, but often you don't know what you want until you're written something else
06:01 theory left
NotFound Winxed is a counter-example. The first implementation of the stage 0 was just an example of parsing, and it kept evolving until today. 06:10
Stage 1, I mean.
Well, you can plan to throw it, but you can change your mind ;) 06:12
06:39 jrtayloriv joined
dalek rrot/lorito: 987fb06 | cotto++ | t/dynoplibs/m0proto.t:
add a test for m0 prototype ops
 07:00
rrot/lorito: 9543274 | cotto++ | src/dynoplibs/m0proto.ops:
various minor fixes to m0 ops
rrot/lorito: 371acc7 | cotto++ | src/dynoplibs/m0proto.ops:
switch to FFA for m0 ctx storage, commit various crimes against C to make it work
rrot/lorito: 31ce228 | cotto++ | t/dynoplibs/m0proto.t:
add some further tests for non-crashing
07:08 ShaneC joined 07:44 contingencyplan left 08:14 cotto left 09:25 woosh joined 09:38 woosh left 09:51 Eduardow left 10:07 lucian joined 10:15 bacek left, dafrito_ joined 10:17 dafrito left 10:20 bacek joined 10:32 ambs joined 10:36 ambs_ joined 10:41 ambs left, ambs_ is now known as ambs 10:57 ambs_ joined 11:02 ambs left, ambs_ is now known as ambs 11:24 plobsing left 11:29 ShaneC left 11:55 Patterner left, Psyche^ joined, Psyche^ is now known as Patterner 12:16 UltraDM joined 12:18 plobsing joined 12:23 whiteknight joined 12:27 mtk joined 12:28 ambs_ joined, benabik left
whiteknight good morning, #parrot 12:28
lucian good day 12:30
whiteknight hello lucian, how are you doing today?
lucian i'm supposed to be working on a dozen different things at the same time ... 12:31
12:31 ambs left, ambs_ is now known as ambs
dalek rrot/jit_prototype: c0b5399 | bacek++ | t/compilers/opsc/21-jit-files.t:
Don't use Parrot_api functions until I figure out what they are doing to ->api_jmp_buf.
 12:34
rrot/jit_prototype: f448688 | bacek++ | t/compilers/opsc/21-jit-files.t:
Comment-out unused variable.
 12:37
whiteknight bacek: having api troubles? 12:38
12:40 plobsing left 12:41 bubaflub joined 12:42 plobsing joined
bacek whiteknight, not really. It's mostly API + NCI trouble 12:43
whiteknight bacek: okay, I designed that api_jmp_buf system, so let me know if you have any questions
12:44 ambs_ joined
atrodo lucian> ping 12:45
lucian atrodo: pong
12:46 ambs left, ambs_ is now known as ambs
bacek whiteknight, check my last commit. Without it "prove t/compilers/opsc/21*" exits with status 1. Because ->api_jmp_buf get clobbered somehow 12:46
atrodo lucian> I saw your comments about Lorito in the log last night, I was curious about what your thoughts on it were 12:47
12:47 JimmyZ joined
lucian atrodo: well, so far I've found the M0 spec flawed 12:47
atrodo lucian> so not the concept, just what's on the table now? 12:48
lucian it strives to be as simple as possible, at the expense of performance
atrodo: mostly, yes
whiteknight bacek: api_jmp_buf is a single field in the interp, and is set on every API call
atrodo lucian> that makes sense. Do you have specific examples off hand?
lucian there were ideas to not expose jumps, and play around with the PC. i don't see how a jit can figure out it should emit a jmp from PC fiddling
whiteknight those api functions can only be called by embedders, not from inside parrot
bacek: I need to make it a linked-list for proper nesting 12:49
12:49 mtk left
lucian atrodo: also, maths operations are severely lacking, compared to the native platforms. again, i don't see how a JIT could figure out repeated sub is division (that's not exactly what was going on, but similar) 12:49
bacek whiteknight, may be. But for jit_prototype PackFile_load_bytecode is good enough. I just want to write as little C code as possible :) 12:50
atrodo lucian> so what you're saying is you'd prefer to see a few more ops that would help a jit know what the intentions were?
lucian atrodo: i think instead of being the least common denominator, i think M0 should be the "most" common denominator when it comes to the expresiveness of he assembly
whiteknight bacek: yes, that's a good first step
bacek: I'll make the api reentrant and nestable soon
lucian atrodo: yes, pretty much. i'd like to see M0 mirror actual JIT targets (llvm, nanojit, x86, arm) 12:51
bacek whiteknight, excellent. Than I can switch to less-magic-functions :)
atrodo (less-magic)++
12:51 ambs_ joined
whiteknight api_jmp_buf gives the interpreter someplace to go if we have an unhandled exception 12:51
atrodo lucian> I'll agree with that, there's a point when simplicity is too simple 12:52
lucian atrodo: so perhaps look at llvm, jvm, x86, arm and find the least common denominator subset (even that i bet is bigger than M0) 12:53
12:53 plobsing left
lucian then add ops that are impossible to JIT for one particular platform otherwise 12:53
12:54 ambs__ joined, mtk joined
lucian things like SIMD 12:54
i don't know how they'd look like in M0
but a JIT must have knowledge of SIMD to effectively and efficiently generate code for SSE, NEON and friends
12:54 ambs left, ambs__ is now known as ambs
atrodo lucian> that'd be hard to see, but it's a good point 12:55
lucian atrodo: i don't know, perhaps i'm entirely wrong. i'm no expert
but i have read of the pains others have endured writing JITs of all kinds
atrodo lucian> any other qualms with lorito/m0? 12:56
lucian atrodo: not really, no 12:57
12:57 ambs__ joined
lucian i'd just rather not see parrot get yet another jit system ripped out 12:57
12:57 ambs_ left
atrodo Yea, it needs one 12:57
lucian i agree with cotto_work++ that llvm should certainly not be the only target 12:58
but i disagree with him on the size of M0
atrodo certainly. I'd also like to see C not be the only possible target as well 12:59
lucian thousands of ops is insane. but a couple hundred for a fast jit seems ok
atrodo Aye. I'd actually like to see less than a hundred. Something a normal person can keep in their head pretty easily 13:00
lucian how many does ARM thumb have? i think it's a few hundred
atrodo good question 13:01
lucian but yeah, fewer is ideal
13:01 ambs left, ambs__ is now known as ambs, woosh joined
lucian so programmer convenience for M0 shouldn't matter at all, i agree with the goals on that 13:01
no one should ever write M0 by hand
but performance? i think that's matters very much for M0
also for security, M0 has to be expressive enough to be able to ban, say, all I/O ops 13:02
atrodo Not by hand no, but compiler writers and parrot hackers will need to be able to grasp it
lucian sure, not *too* hard
atrodo lucian> that's one thing I'm staunchly in agreement with. Unless m0 is completely inaccessible to the end user
lucian java's security system is a nice research project, but it sort of sucks 13:03
i've looked a bit into how PyPy does sandboxing, and it's interesting
untrusted code runs in a process that can make no system calls, just write stdout and read stdin 13:04
whiteknight parrot definitely needs a sandboxing system. Of course, we need a design for such a system before we start writing code for it
lucian but that's perhaps tooe xpensive
atrodo I don't like java's at all, but that python system sounds interesting
lucian well, java's isn't that bad. they ban classes of jvm bytecode and java classes 13:05
the first bit sounds interesting for M0
whiteknight I think the prevailing idea about Parrot is that M0 won't have any security whatsoever, but different dialects or subsets or M1 will disallow certain operations
atrodo except when you can escape the sandbox with a variable change
whiteknight so we could use an M1 dialect with no system calls for untrusted code
lucian whiteknight: what do HLLs compile to? not M0? 13:06
whiteknight lucian: depends. The currently compile down to PIR/PBC. That's just one possible dialect of M1
I see no reason why that wouldn't continue
atrodo whiteknight> That only works if M1 is the only thing talking to M0
lucian whiteknight: right. but what if others use M0?
atrodo as soon as you let others compile to M0, you have to include security
lucian i see M0 as analogous to hardware-level assembly, and M1 as a C equivalent 13:07
am i wring?
whiteknight lucian: safety features only work if people use them. If you write unsafe code, you can't expect Parrot will inject safety for you
or, if you call an unsafe library from safe code, you get the same effects. caveat emptor
lucian whiteknight: no, but i could expect to be able to mark a chunk of bytecode as "no system calls"
13:08 plobsing joined
whiteknight lucian: depends. Do you mark it at compile time? Do you verify that mark before execution? 13:08
lucian whiteknight: i'd guess so. if unsafe, only allow execution of non-I/O non-system-call bytecode 13:09
and have input and output ops to communicate with the controller
whiteknight lucian: the question is, do we want parrot to make an "is this safe?" check on every single opcode dispatch for every single program?
Or, as I would prefer, do we do the verification of safety ahead of time?
atrodo whiteknight> define when "ahead of time" is. Compile time? 13:10
lucian whiteknight: it needs to read the bytecode at some point. if the discrimination is done on ops alone (not arguments), then it can be checked in one go
whiteknight Either we generate code with known safety properties, and trust that those properties were handled by the compiler, or we add in a verifier that can look at the PBC and determine safety before execution at runtime
lucian whiteknight: i like the latter a lot more
whiteknight atrodo: any time before execution. Can be at compile time or runtime. Each make different degrees of safety guarantee
atrodo whiteknight> You can never trust what the user gives you. if you want to prevent malice, you must do it at run time 13:11
lucian but it could be at M0 compilation time, sure
whiteknight lucian: if we tell the compiler to generate safe code, then sign the bytecode to prevent tampering, we get a good effect with zero runtime cost
lucian whiteknight: which compiler? if it's the M0 assembler, i agree
whiteknight atrodo: true. So we can add an ahead-of-time verifier. For safety your application may only use the ops in M1-safe dialect, etc
lucian whiteknight: and what exactly is M1? 13:12
atrodo lucian> PIR
whiteknight lucian: any language that compiles directly to M0
lucian whiteknight: then it's inherently unsafe
whiteknight the idea is that M1 will be intermediate forms like PIR/PBC. Things that compile to M1 are called M2, etc
lucian whiteknight: i see. i was thinking of your M2 compiling to M0 13:13
whiteknight lucian: people can take whatever path they want, but if they avoid M1, they lose the benefits and properties of M1
lucian is that check so expensive? it only needs to happen when reading in bytecode
whiteknight all we can do is provide the tools
a check at runtime before execution is probably not bad, especially if we can inform the verfier with things like annotations from the compiler and digital signatures 13:14
lucian i can't see a compelling reason for having two levels
whiteknight per-op checks during execution is bad
13:14 bluescreen joined
lucian yes, that would be bad 13:14
atrodo whiteknight> I agree, it has to be at a segment-ish level
lucian especially because it couldn't JIT
whiteknight right
plobsing an M0 verifier will not be as capable as an M1 verifier and will likely need to be of sufficiently higher complexity (both code and runtime) in order to even come close 13:15
bubaflub i know that PL/Parrot has *some* security features
plobsing M1 -> M0 destroys information
lucian plobsing: is it important information if you just disallow system calls?
13:15 ambs_ joined
plobsing lucian: not all system calls are created equal 13:16
whiteknight plobsing is exactly right. M0 is designed more for performance and flexibility. safety is not a concern of the M0 design
M0 are the primitives in which more advanced systems are implemented
lucian plobsing: and sandboxed code shouldn't do any of them
plobsing futex and writev are very different from a security perspective
lucian: are you saying I can only have all or nothing? that's not a very attractive security model. 13:17
13:18 ambs__ joined
lucian plobsing: the approach i was thinking of was having an explicit "angel" guard untrusted code, marshalling all system calls 13:18
sandboxed code can output and input
of course, that's only one approach
13:19 ambs left, ambs__ is now known as ambs
lucian plobsing: then, you can be sure that by default sandboxed code can't do damage 13:20
no need for a complex verifier, the app developer decides what's safe
mikehh one of the first computers I worked on was the CDC 6000 series - it had a very interesting architecture
13:21 ambs__ joined, ambs_ left
lucian reads 13:22
atrodo mikehh> how so?
mikehh all CPU operations were essentially register based, there were basically 8 X registers, 8 A registers
lucian en.wikipedia.org/wiki/CDC_6000_series
mikehh Setting A registers A1-A6 loaded the appropriate X register from memory, setting A6,A7 saved the register to memory 13:23
all arithmetic and logic ops were performed between X registers 13:24
13:25 ambs left, ambs__ is now known as ambs
mikehh furthermore all I/O was performed by separate (10 in early models) PPs which 13:25
copied to and from memory - the CPU had no I/O
the op system was controlled from PP0 13:26
13:28 benabik joined
lucian mikehh: so as a parallel, parrot could have an explicit "interpreter" for I/O ops? 13:33
benabik G'morning, #parrot
lucian waves 13:34
moritz particles
Coke ENOPARTICLE
benabik o/~ The best part of waking up is #parrot in my comp. o/~ 13:35
13:36 JimmyZ left
mikehh lucian: yes, just a thought, it was very effieient in its day 40+ years ago 13:36
lucian mikehh: yes, i can imagine
mikehh: what i was thinking is allowing application developers to do that 13:37
your own application is trusted, implicitly
mikehh I never understood why that concept was never really used in modern architecture
lucian you might want plugins to be not trusted, so you have a guard to marshall all their calls 13:38
mikehh: because few care about security. look at C
plobsing: atrodo: my idea may be crap, my point is that perhaps there needn't be yet another layer on top of M0. PIR, winxed, rakudo, pynie could all compile down to M0 13:39
mikehh lucian: not just the security aspects, all I/O is separate, ops are mostly register based
atrodo lucian> I agree with you. I think M0 should be another type of bytecode that parrot can consume 13:40
lucian mikehh: sure, that's what i'm saying. the "guard" or "angel" is the only thing sandboxed code can do I/O with
atrodo mikehh> not all that unlike m0 in theory
lucian mikehh: or it could be at a lower level as you're saying
mikehh you can then have load ops, store ops and all arithmetic, logic etc. is register to register 13:41
and you would need separate address calculation ops 13:42
maybe/maybe not 13:43
lucian mikehh: it sounds deliciously modular, but may be impractical indeed
mikehh and then separate I/O processes, with a different interpreter 13:44
the I/O can handle most of the security aspects 13:45
lucian mikehh: or just memory-isolated threads
mikehh lucian: I was thinking of keeping I/O completely separate from core ops 13:46
lucian mikehh: in a separate process? 13:47
13:47 Eduardow joined
mikehh lucian: yes 13:47
lucian mikehh: can't it be done with a separate thread only? 13:48
what would the downsides be
moritz you'd need some kind of IPC 13:49
atrodo doing io in a separate thread would be tricky if you don't have more than 1 thread
moritz which would be implemented as IO, mostl likely :-)
atrodo IO all the way down 13:50
moritz (or how reliable does shared memory work arcross operating systems?)
lucian that's what i'm saying, memory shared between threads works reliably everywhere
mikehh have a shared/lockable memory area for I/O buffers 13:51
core threads can use shared memory (or not) - all I/O in I/O buffers - I/O threads deal with that 13:53
could be slow on a single core machine, but not many of them around now 13:54
atrodo mikehh> what about embedded devices? 13:55
mikehh even 'phone processors are going multi-core
most new ARM and Atom processors are moving to multi-core 13:56
13:57 PacoLinux joined
lucian mikehh: what about existing devices 13:58
the scheduling overhead might just be too much
mikehh is parrot ready enough for them?
and are we going to port parrot to all existing devices 13:59
moritz surely not all :-)
13:59 ambs_ joined
mikehh parrot is not really production ready yet, so we should look to the future :-} 14:00
lucian i have plenty of single-core machines
anyway, it's an interesting idea 14:01
but the I/O threads will be idling most of the time anyway
so perhaps cooperative threads can work too
14:01 alin joined
lucian CPS should help 14:01
moritz mikehh: when you always look into the future, you forget the presence
mikehh: remember that parrot does have users
independently of whether you call it "production ready" or not
14:02 frodwith left
mikehh moritz: yes, but how many are operating on single-core machines? 14:02
benabik mikehh: Multi-core, while common, is not universal.
moritz mikehh: I don't know. We don't collect machine statistics from all HLL users 14:03
mikehh moritz: I know I am not, and I don't think you are
moritz mikehh: what we get is usually feedback from developers, and that is biased towards stronger machines
14:03 ambs left, ambs_ is now known as ambs
benabik OTOH, most single-core machines can deal with multi-threading without a major loss of efficiency. 14:03
moritz mikehh: so the simple answer is "we don't know, and thus shouldn't assume"
mikehh moritz: kid51 has a old Mac he tests on, but how many perl6 users try out single-core machines - maybe laptops, but new ones are nearly all multi-core 14:04
moritz mikehh: again, we don't know. 14:05
lucian benabik: i challenge that assertion
mikehh: i just bought a single-core netbook, brand new
moritz mikehh: I know that sorear stopped hacking rakudo because compilation took too long (and too much memory). Doesn't sound like a huge many-core server.
mikehh lucian: does it run parrot? 14:06
lucian mikehh: yep 14:07
mikehh: ARM Cortex-A8, 800mhz
whiteknight lucian: surely one counter-example does not disprove the trend
for good or bad, multi-core is the direction almost all chip manufacturers are heading 14:08
moritz no. But neither do statements like "most modern..." prove it.
lucian whiteknight: of course not. but multi-core is for parallelism, not concurrency
mikehh moritz: yes it takes far too much memory at the moment, I don't think separating I/O would cause a problem there
lucian whiteknight: even on multi-core, idling threads just create more scheduling waste
if threads idle, they shouldn't exist 14:09
mikehh lucian: all operating systems have a degree of parallelism
whiteknight I'm not saying anything about threading. Only pointing out that, whether we want it or not, multi-core is the future of computing
mikehh lucian: depends on what resources they are using, eg event monitoring 14:10
whiteknight: yes, but even single-core requires some parallelism 14:11
benabik lucian: Which assertion? The loss of efficiency? 14:13
lucian mikehh: well, in most cases thread scheduling overhead is deadly at around 2000 threads 14:14
benabik: yes
mikehh lucian: was not looking at that many, if you have that many you should have a separate core or more dedicated to scheduling 14:15
lucian mikehh: or not have them in the first place :)
benabik lucian: Implementing async IO as a thread using blocking IO should cost (IO + 2 thread switches), and a thread switch is generally lighter weight than a context switch. 14:16
lucian threads waiting for I/O has been repeatedly show to be a bad idea
mikehh: but i really like the idea of a semantic "process" handling I/O
mikehh lucian: well I tend to think you need to go to the erlang model or something
14:16 UltraDM left
lucian mikehh: what do you want people to do with threads? 14:17
mikehh: in general, on parrot
benabik lucian: But depending on how you use threads, you can turn all the thread switching into just function calls. "create_worker_thread(function)" becomes "function()", just run the IO functions in thread every time "add_IO_request" is called, etc.
lucian benabik: but the scheduling overhead of that one thread can be felt with lots of threads 14:18
mikehh lucian: yeah - I have been thinking about something like that for while (separating I/O)
lucian: and no - I have been reading a lot of anti-thread articles recently - different modelks are available 14:19
benabik lucian: Yes. I will agree creating many threads will be very very bad. But one or two event handling threads rarely kills performance.
lucian benabik: i see that as a workaround for the lack of proper async I/O, and it's ok for that
mikehh: heh. well, threads are good for CPU parallelism. they're a low-level feature for that 14:20
but the I/O guard processes need not be threads, i think
mikehh lucian: some of the $work I have been doing is on HPC - mostly using MPI
lucian benabik: GLib's gio does that, and it works out alright if you otherwise strictly use threads for parallelism (not concurrency) 14:21
mikehh lucian: which is why I am advocating separating out I/O, however we do it - monads anyone 14:22
14:24 plobsing left
lucian mikehh: of course, attention should be kept to I/O perf 14:24
mikehh ok - I am going to switch to 11.04 beta and see if I can figure out how to build perl 5.12.3 14:27
libraries are in different places and config does not seem to pick them up
14:27 ambs left
mikehh bbiab 14:27
14:28 JimmyZ joined, plobsing joined 14:29 mikehh left
Tene lucian: Sorry, this weekend didn't work out at all for me. I'll try again tonight. :( 14:41
14:41 theory joined
lucian Tene: what for? 14:41
reading or talking?
Tene lucian: responding to your 6model/python inquiry
14:42 cotto joined
lucian Tene: ok 14:43
Tene finally goes to sleep, about 8 hours later than planned, for a short nap before going to work. :/
14:46 bubaflub left 14:48 rohit_nsit08 joined
rohit_nsit08 whiteknight: hello, good morning 14:49
whiteknight hello rohit_nsit08
rohit_nsit08 whiteknight: what is finite automation in lexical analysis?
whiteknight that's a very big question!
rohit_nsit08 i read it's something similar to finite state machine we use in digital design but couldn't find any good reference on it. 14:51
whiteknight rohit_nsit08: it's mostly implicit. You don't actually have a "state", but you enter different flow paths when you get different inputs 14:52
rohit_nsit08 whiteknight: hmm didn't get that, u mean there are no states? 14:53
whiteknight: i read about in the context of lexer that, when it reads the source code and converts it into tokens , this conversion is done with the help of finite automation. 14:54
whiteknight rohit_nsit08: let's say we have two tokens in a language "foo" and "function" 14:55
rohit_nsit08 whiteknight: okay
whiteknight if we are lexing and we find an "f", we are in a state "foo or function"
then if we find "fo", we are in state "foo", and need to finish it
rohit_nsit08 hmm.. so they are like parallel paths and lexer switches between them based on the next input? 14:56
whiteknight so we have a state, based on the letters we have already found
right
rohit_nsit08 hmm..the concept is very similar to finite state machine 14:57
whiteknight: i updated my blog yesterday :-) i thinking to use it to keep track of my current work. 14:58
whiteknight rohit_nsit08: link?
rohit_nsit08: blogging is an important part of GSoC 14:59
rohit_nsit08 whiteknight: rohitnsit08.blogspot.com
whiteknigt: i'm writing my proposal in the form of blog post loll
whiteknight excellent
rohit_nsit08 thanks
15:11 mikehh joined, bubaflub joined 15:18 dmalcolm joined
rohit_nsit08 whiteknight: i submitted my proposal yesterday, u said there is some ranking system. Can students know about their ranks? 15:18
moritz rohit_nsit08: I don't think so. And the ranking only starts after the submission deadline 15:19
whiteknight rohit_nsit08: I don't know the details, but I think the ranking is just for keeping notes
mentors give points to applications, but those points aren't the only thing that matters. It's just to keep track 15:20
moritz aye; in the end a consensus is needed
whiteknight rohit_nsit08: We will be giving feedback as the process continues
moritz is correct
rohit_nsit08 moritz: Hi, okay.
whiteknight rohit_nsit08: and even if we have a great proposal, we still need to find a mentor, and we still need to have an available slot
rohit_nsit08 whiteknight: ya i understand that :-)
cotto_work ~~ 15:26
Cool. I have some interest in my trac/github fork. 15:36
15:41 plobsing left
bubaflub cotto_work: link? 15:44
cotto_work bubaflub: github.com/mickelangelo/github-trac
bubaflub: it looks like he's trying to integrate a bunch of forks of that plugin. 15:45
bubaflub cotto_work: great 15:48
15:49 frodwith joined
jnthn_ In case it's of interest to anyone here: I did a talk at OSDC.TW that looked at building compilers with NQP and some basic 6model bits; slides: www.jnthn.net/papers/2011-osdc.tw-compiler.pdf and code: www.jnthn.net/papers/2011-osdc.tw-c...r-code.zip 15:55
whiteknight jnthn++ 15:58
jnthn_ I'll try and sort out the 6model documentation situation in the next week or so. 15:59
15:59 frodwith left
whiteknight jnthn_: We would really appreciate it. At least one GSoC project may be using 6model intimately 16:00
jnthn_ I'm aware there's plenty of confusion about how to use it, and what it actually provides/intends to provide.
OK, which one?
I know somebody wants to work on PDL-ish stuff (the S09 bits).
whiteknight lucian's Python compiler
jnthn_ Ah, OK
whiteknight if he decides to go that route, which I do not know he will
jnthn_ I saw that post. It's one of the things that told me "I really must document this stuff more" :)
whiteknight Tene will be using it in Cardinal too, though I don't know a timeframe for that 16:01
16:01 frodwith joined
whiteknight We're going to want an improved object model for JavaScript too, though we probably won't do anything object model related this summer 16:01
jnthn_ Yes, I've already answered a few of Tene++'s questions and added a repr that was suitable.
I think the repr I added for Cardinal will - at first blush - be fine for Python too. 16:02
whiteknight I suspect the JavaScript case will be special, because we can just use Hash instead of Object
jnthn_ And probably also for JavaScript.
PerlJam hopes that guy succeeds with making Perl 6 more PDLy
jnthn_ (Hash instead of Object) That misses the point of what 6model exists to do :)
whiteknight although we will probably want a wrapper which can take a JavaScript "object" and make it more suitable to use from a language which separates methods and attributes
jnthn_ Using a hash for storage is a representation decision 16:03
You still want the think to act like an object otherwise.
So it's just choosing the correct representation.
s/think/thing/ 16:04
Rakudo's default representation will be P6opaque, but that's not suitable for various other languages that really just want a dictionary of attributes that looks hash-ish. 16:05
You can even have representations that do hybrid things: optimized storage for compile-time kmown attributes, and a fallback, lazily allocated hash for ones that come into existence at runtime. :)
16:06 frodwith left
moritz iirc that's what V8 uses as an optimization 16:06
jnthn_ IIRC, V8 is a little more complex than that, but yes, it's along those lines.
16:07 frodwith joined
jnthn_ V8 also tracks transitions an object make and can "rebless" 16:07
We don't have the mechanism for that in 6model yet.
Probably in a month or two we will.
16:07 frodwith left
jnthn_ I'm trying to see if I can get the same primitive to handle that and mix-ins. 16:07
BTW, languages where the methods are just attributes are easy. They just have a REALLY straightforward implementation of the add_method meta-method that sets an attribute. :) 16:08
16:08 frodwith joined
jnthn_ If JavaScript, your meta-object is probably a singleton. 16:09
16:09 darbelo joined
jnthn_ To a first unoptimized approximation, anyways. 16:09
moritz do you compile method calls to <attribute_lookup><invoke>? 16:10
or to a method call, and add a custom dispatcher?
16:10 frodwith left
moritz the latter sounds more cross-HLL robust 16:10
whiteknight moritz: it shouldn't matter. find_method will look for an attribute with the same name, and return null if that attribute isn't invokable
16:11 frodwith joined
jnthn_ moritz: The term "custom dispatcher" really needs to go away. 16:11
It's just a "dispatcher" :)
moritz whiteknight: ah, so it's just find_method and not a whole dispatcher
whiteknight right
jnthn_ find_method *is* a dispatcher pretty much.
Well, or delegation to one. 16:12
whiteknight right, minus the part where we actually execute what we find
jnthn_ Yeah
6model largely follows the Parrot model for now, fwiw
moritz "no trial, no jury, straight execution" :-)
jnthn_ find_method chooses what we'll dispatch to
whiteknight that "for now" part sounds ominous
jnthn_ And then actually invoking that is a separate step.
16:13 frodwith left
jnthn_ whiteknight: Oops, I didn't mean for it to. 16:13
whiteknight: fwiw, 6model in the CLR follows the Parrot model of separate find and invoke too.
whiteknight: I haven't really felt strongly that this needs to change. 16:14
You have to do both steps *somewhere*.
whiteknight okay, good. Not because I want more things to be the parrot-way, but because that separation is generally good design, methinks
jnthn_ Yes, I'm fairly comfortable with that design.
whiteknight Part of me wishes we had a "port Parrot to 6model" GSoC project in the works 16:15
jnthn_ It just makes it harder to write deferal-y stuff.
whiteknight but I worry that such a project would be far too huge an undertaking
jnthn_ It would be.
We should slip 6model in gradually.
Not in one big bang.
whiteknight jnthn_: is there a gradual way? Could you put together even a general outline of how that kind of conversion would take place?
our current system is so tangled, I worry it's going to be impossible to excise parts of it 16:16
darbelo Given a sufficiently large axe, excision is always possible. 16:18
jnthn_ whiteknight: Something like, 1) make the 6model stuff available in Parrot at all, 2) create meta-objects that provide Parrot's current class semantics so it's a small-ish step to switch over much code, and hope many people take that path, 3) get PMCs to be 6model objects, 4) hope Class and Object PMCs can go away by this point, or just turn them into simple things that delegate to 6model to give people more time to switch away. 16:20
Somewhere along the line, the methods/namespace relationship and the class registry need to go away. 16:21
imho, anyway
:)
I think the biggest mis-understanding at the moment is that people seem to think 6model is an implementation of a bunch of OO-ish stuff. In reality, it's a bunch of building blocks to let you build your languages view of OO. 16:23
*language's
PerlJam It wouldn't hurt to add that outline and a short description to the gsoc ideas page. 16:25
(especially if you broke it out into 2 or 3 projects)
jnthn_ You can see it like HLL::Grammar/HLL::Actions. It doesn't try to define the grammar for your language, but it does give you ways to parse identifiers, an OPP, quoted construct parser, etc.
whiteknight jnthn_: okay, so basic strategy is to get 6model into the core first, then start removing Object/Class? 16:26
I like that plan well enough
PerlJam how many gsoc proposals has PaFo gotten? 16:27
whiteknight That's a good question
lucian jnthn_: you might be interested in what PyPy does, similar to v8 in some respects morepypy.blogspot.com/2010/11/effic...jects.html
whiteknight PerlJam: I think we had 6 proposals submitted as of this morning
PerlJam cool
whiteknight but I know of other students who have show interest but haven't put anything up on the google-melange site 16:28
16:28 plobsing joined
PerlJam TPF only appears to have 2 submitted though I've heard tell of several others in the works. 16:28
darbelo We should encourage students to submit to melange as soon as possible. 16:29
jnthn_ lucian: Thanks, looks like a good article
lucian jnthn_: i've allocated the 1st week of gsoc to investigating 6model, or falling back to a custom map-for-attributes option if that fails
jnthn_ whiteknight: Well, the PMCs become real objects part is important too :)
whiteknight yes, 6 submissions right now: rohit_nsit08, lucian, soh_cah_toa, bubaflub, benabik, and joehms22
jnthn_ lucian: OK, when is that "first week"?
lucian jnthn_: i won't necessarily do it then, might be earlier 16:30
benabik needs to severely redo his proposal.
PerlJam lucian: I think jnthn is asking so he knows by when to have the docs done (or mostly done) :)
whiteknight WE NEED THEM YESTERDAY!
lucian PerlJam: i see
PerlJam or even started :)
lucian whiteknight: +1
whiteknight or, whenever
:)
jnthn_ PerlJam: Yes, that :) 16:31
benabik whiteknight: Not very funny... I'm really working hard to fit time to do proposals in between homework. :-/
lucian jnthn_: www.google-melange.com/gsoc/events/...e/gsoc2011
either community bonding period (might be busy with my dissertation then) or after start
16:31 jrtayloriv left
PerlJam jnthn: Have you signed up to be a mentor for TPF gsoc? 16:31
lucian jnthn_: so late april/may
16:32 benabik_ joined
PerlJam jnthn_: or for PaFo gsoc 16:32
?
jnthn_ PerlJam: Not yet
16:32 benabik left, benabik_ is now known as benabik
jnthn_ brb, laundry 16:32
16:33 bubaflub left
whiteknight benabik: I was talking about jnthn and his documentation 16:34
benabik Ah. Juy poor timing on my part then. Nearly had a heart attack. 16:35
whiteknight: ^^
jnthn_ back 16:37
Coke 6 proposals to PAFO, 2 to TPF
dukeleto ~~
jnthn_ PerlJam: I got interest in doing PDL stuff in Perl 6 and offerred to mentor it.
s/got/saw/
Coke dukeleto, whiteknight: I assume "hold off voting" until you say how it's working this year? 16:38
dukeleto Coke: i expect those numbers to increase this week. It is all about procrastination
PerlJam wonders if he should even apply to be a mentor for PaFo gsoc
dukeleto PerlJam: yes, please.
PerlJam: we need as many mentors as we can get
PerlJam I just feel so out of touch with Parrot these days. 16:39
benabik dukeleto: Or prioritization. I have very little time for things not due tomorrow.
cotto_work PerlJam: mentors don't need to be omniscient
Coke PerlJam: one of the proposals involves POD, so there you go.
PerlJam okay, okay. I just applied :) 16:40
dukeleto PerlJam: awesome!
16:40 davidfetter joined
dukeleto adds gsoc ideas to our ideas page 16:47
whiteknight jnthn++: That presentation was very nice 16:49
jnthn_ :)
It's *so* nice that you can write the whole compiler in a HLL now :)
whiteknight dukeleto: what ideas? 16:50
dukeleto whiteknight: i am stealing some GCI ideas and making them bigger 16:54
whiteknight: for instance, embedding parrot in a browser
whiteknight oh, nice 16:55
dukeleto: I don't know if any of them will be used this year, but I think we should keep a list like that available all year
we should merge that into our BigProjectIdeas page
dukeleto whiteknight: yes
whiteknight: you have any other ideas bouncing around? 16:57
whiteknight nothing in particular, I haven't been thinking about it really 16:58
I was brainstorming ideas pretty hard, until we got huge turnout
16:58 benabik_ joined, benabik left, benabik_ is now known as benabik 17:00 bubaflub joined
whiteknight msg cotto_work I'm looking at Parrot-Instrument now. It appears all tests are segfaulting in the same place. Looks GC related (something on the free-list is clearly not free). I suspect GC write barriers are the cause 17:01
aloha OK. I'll deliver the message.
whiteknight or, lack thereof
dalek tracwiki: v41 | dukeleto++ | GSoc2011
tracwiki: trac.parrot.org/parrot/wiki/GSoc201...ction=diff
tracwiki: v42 | dukeleto++ | GSoc2011
tracwiki: trac.parrot.org/parrot/wiki/GSoc201...ction=diff
dukeleto whiteknight: should docs.parrot.org/parrot/devel/html/d...w.pod.html exist ? 17:02
whiteknight: docs.parrot.org/parrot/devel/html/d...d.pod.html seems to think it should
whiteknight there is an embed_new.pod file in the repo
either the link is wrong or the new version isn't being htmlified
Coke new docs are not automatically html-ified. 17:03
whiteknight well there's your problem
Coke (depends on the directory it's created in. new pdds should be, e.g.)
checking...
dalek rrot: f64d532 | Coke++ | docs/index/developer.json:
Generate html for embed_new docs.
 17:05
whiteknight Coke++
Coke probably want to fix the reference in embed.pod so it's actually a link.
(note that I didn't fix the website, just 'make html' -it'll show up after the next release.) 17:06
17:07 alin left
cotto_work whiteknight: I'm here 17:09
17:09 spinclad left
cotto_work That sounds as likely as anything. 17:10
whiteknight cotto_work: I'm idly looking over the source for the .pmc files now, looking for places to insert write barriers
cotto_work whiteknight: I'm sure you'll find some. parrot-instrument has some very unusual PMCs.
whiteknight there are certain parts where I'm pretty certain I do not like the design 17:11
but that's neither here nor there
important is getting the damn thing to work
then making sure it's capable of supporting a new debugger 17:12
cotto_work yes
17:12 benabik left, JimmyZ left
dukeleto begs people to add more gsoc ideas 17:14
whiteknight do we not have enough?
dukeleto whiteknight: no, we don't
whiteknight: more is better
whiteknight are there students circling around in need of an idea?
dukeleto whiteknight: lots of students are trolling ideas pages right now, the few days before the deadline 17:15
whiteknight: ideas are still valuable, for gsoc or not, so I just like having lots of ideas
whiteknight ok
dukeleto whiteknight: there are plenty of crazy smart students trying to decide on an organization, and we want something that piques their interest
whiteknight okay, I'll see what I can come up with today 17:16
PerlJam dukeleto: how do those students even know to look at our pages?
whiteknight there's a link from the GSoC page
PerlJam (who's advertising?)
dalek tracwiki: v43 | dukeleto++ | GSoc2011 17:17
tracwiki: trac.parrot.org/parrot/wiki/GSoc201...ction=diff
17:22 mj41 joined
dukeleto PerlJam: i try to advertise, via twitter and identi.ca, and there are links from the main GSoC page 17:30
the more help I get, the better
dalek tracwiki: v189 | jimmy++ | WikiStart 17:33
tracwiki: add 6model link
tracwiki: trac.parrot.org/parrot/wiki/WikiSta...ction=diff
17:38 ShaneC joined
cotto_work Do we have a task to do something with 6model on Parrot? Moving wholesale is too big for a summer, but it seems like a useful subset of the task (e.g. initial implementation and poc) could be completed. 17:41
dukeleto cotto_work: perchance 17:42
cotto_work: 6model has fuzzy edges
cotto_work It's in nqp already. 17:43
dukeleto cotto_work: sure, but can you come up with a student project that is completable in 3.5 months and is well-defined ? 17:47
dalek rrot-instrument: 3c33584 | Whiteknight++ | src/dynpmc/instrument (6 files):
sprinkle in some GC write barriers in places that look important. Appears to have no effect on test suite
 17:48
tracwiki: v1 | jimmy++ | 6model 17:49
tracwiki: init 6model stuff
tracwiki: trac.parrot.org/parrot/wiki/6model?...ction=diff
tracwiki: v2 | jimmy++ | 6model
tracwiki: trac.parrot.org/parrot/wiki/6model?...ction=diff
tracwiki: v3 | jimmy++ | 6model
tracwiki: trac.parrot.org/parrot/wiki/6model?...ction=diff
tracwiki: v4 | jimmy++ | 6model
whiteknight cotto_work: I was just talking to jnthn about this. He suggests a good course of action is to get 6model code into Parrot along side our existing object model. Then we can start migrating Object and Class guts to it
dalek tracwiki: trac.parrot.org/parrot/wiki/6model?...ction=diff
cotto_work whiteknight: that's a sensible approach that's served us well in the past
whiteknight cotto_work: the more I think about it, that's probably the best. We have to figure out how to expose 6model to PIR code without conflicting with Class, however
I wouldn't want to introduce a "6modelClass" type, for instance. It should have a name that we would be happy to keep long term 17:50
cotto_work whiteknight: can you think of a way to expose m0-related constants to pir in a way that's less noisy than "m0_add_i ctx, .M0_I1, .M0_I2, .M0_I3"? 17:53
ideally "m0_add_i ctx, i1, i2, i3" 17:54
whiteknight what do you mean "m0-related constants"? 17:55
cotto_work names for m0 registers 17:57
I'm prototyping M0 in pir dynops. 18:00
whiteknight oh, okay
if you're trying to prototype it in PIR, you're just going to have to deal with some pain
.M0_I1 looks fine to me
cotto_work or will I? 18:18
18:28 dodathome joined 18:31 davidfetter_ joined 18:34 soh_cah_toa joined 18:35 davidfetter left, davidfetter_ is now known as davidfetter 18:37 cognominal left 18:43 mtk left 18:45 cognominal joined
lucian 127 ops hmm developer.mozilla.org/En/Nanojit/LIR 18:47
18:49 mtk joined
lucian i can't figure out whether nanojit can be used for function jiting 18:53
whiteknight like Parrot function jitting?
lucian whiteknight: sure. from what i'm reading, it may only be any good for tracing jits 18:56
which ain't bad at all, but with possibly less potential 18:57
(most js jits use both)
whiteknight I see no reason why NanoJIt couldn't do Parrot subs, based on what I am seeing here
it has call ops
it has the jump ops that Parrot CPS would use 18:58
nanoJIT has everything we would need
lucian hmm. webkit's squirrelfish extreme at first did "context threaded jit", which is a pasting jit essentialy 18:59
whiteknight it does have the extra barrier guards too. Parrot could use that to JIT individual subs, then jump out when the sub exited
lucian needs to read some more
whiteknight: there's something i read, about moz moving to webkit's codegen, or similar 19:03
whiteknight: ah blog.mozilla.com/dmandelin/2010/02/...germonkey/ 19:04
whiteknight: from what i'm reading, it appears they'd picked most of the low-hanging fruit with interpretation 19:05
i don't understand how JSCore's assembler is better, though 19:06
whiteknight: right, should've read through the end 19:08
JSCore does no optimisations at all
but it's very fast
i still don't get why nanojit couldn't have the optimisations turned off 19:13
19:34 theory left 19:35 theory joined 19:36 frodwith joined 19:37 bubaflub left 19:38 spinclad joined 19:47 ambs joined
dukeleto frodwith: welcome! 19:47
frodwith hey dukeleto. 19:48
19:49 mj41 left 19:50 bbatha joined
dalek rrot/lorito: 644a53b | cotto++ | / (3 files):
add a header with an m0 register name->number enum
 19:57
rrot/lorito: 339c95f | cotto++ | / (2 files):
add the necessary magic to generate pasm consts from an enum

Right now this results in things like q<m0_add_i ctx, .M0_I1, .M0_I2, .M0_I3> which is a bit too noisy for non-generated code.
rrot/lorito: 614523a | cotto++ | / (2 files):
use a bit of macro hackery to make M0 ops look nicer
rrot/lorito: 2bb961a | cotto++ | t/dynoplibs/m0proto.t:
make register use a little more consistent
20:03 bubaflub joined 20:04 bluescreen left
dukeleto cotto_work: rock on! 20:14
it is nice to see somebody else hacking on the lorito branch
20:15 hercynium joined
cotto_work dukeleto: it's nice to get something that can run, incomplete though it is. 20:16
20:18 bluescreen joined
dukeleto cotto_work: as my math advisors used to tell me: "something is better than nothing" 20:19
cotto_work: i have been living under a rock for the last few days, and have not had much time to hack on parrot or gsoc stuff
cotto_work: but it looks like we have an array of good proposals. Of course, I want more :) 20:20
cotto_work dukeleto: of course.
I'm a little surprised that we have as many as we do by this point. I know that if I were participating, I'd have only a vague idea of what I wanted to do by now. 20:21
lucian cotto_work: you'd be surprised 20:22
my first GSoC i was totally lost
ended up writing a proposal even i was impressed with 20:23
whiteknight the feedback process is extremely helpful
lucian that too 20:24
moritz TPF only has two proposals by now
lucian also, one gets extremely excited when offered a free hand and money 20:25
uh, i may have gotten that idiom wrong
whiteknight the money is definitely nice 20:26
20:27 whiteknight left
lucian yep. i've been pleasantly surprised however to find most fellow students more interested in pursuing their beloved project than the money 20:27
bubaflub lucian: the money is nice, but contributing to open source and advancing the project is much more fun 20:28
in my eyes the money just makes play-time possible
lucian bubaflub: exactly. and i'm glad most GSoC folk actually feel that way
i'd expect gold diggers to get disqualified fast, too 20:29
enthusiasm reeks
bubaflub lucian: we are violent agreement 20:30
20:31 plobsing left
dukeleto the money seems to be mostly un-important to students from wealthier countries, but $5K USD is a lot of money in some countries. It is more than the average yearly salary of most citizens. 20:31
bubaflub: how are those GMP bindings? 20:32
dukeleto cracks the whip and starves the hounds
bubaflub dukeleto: working, though i can't get the mpz_get_string quite right
lucian dukeleto: it's quite a lot for me too, but i guess i'm not a typical case
bubaflub dukeleto: i'm thinking about doing it in Winxed and Rosella
dukeleto lucian: it is a lot for any student in any country. "un-important" is quite right 20:33
bubaflub: interesting
bubaflub: there are performance considerations
bubaflub: will it be a mixture of PIR and Winxed ? 20:34
bubaflub dukeleto: it might be. the ncidef2pir could give us the basic PIR bindings
dukeleto: and then the object stuff in Winxed
dukeleto bubaflub: create a github issue for your problem with get_string and add any error output, stacktraces that you have 20:35
bubaflub: if you are talking about having a low-level PIR access and higher-level Winxed access: Yes.
lucian i've sort of come to the conclusion lately that you can't really do proper C bindings without a C compiler
dukeleto bubaflub: i just don't want to force the performance penalty of Winxed on people if they care about speed
lucian i'm too lazy to write a second proposal, though :)
bubaflub dukeleto: i could also have Winxed compile down to PIR
20:36 woosh left
dukeleto bubaflub: yes. but generated PIR is not as efficient, yet. 20:36
bubaflub dukeleto: ok. i think it's best to make the definition file, use ncidef2pir to generate the lowest level bindings possible in PIR
dukeleto bubaflub: but if you generate bytecode, i guess that comes out in the wash
bubaflub dukeleto: consumers can then use those functions as PIR 20:37
dukeleto bubaflub: yes, i like that
bubaflub dukeleto: or they can use the Winxed GMPz object for conveience
dukeleto bubaflub: i think you need a very low-level interface to GMP in PIR, and then you can build up a nicer/friendlier/slower interface from any HLL that you want
bubaflub: but concentrate on the low-level interface first. You need a foundation that can be built upon. 20:38
NotFound dukeleto: What winxed performance penalty?
dukeleto NotFound: i don't know specifics, I just assume. Perhaps I am making an ASS out of U and ME.
20:38 bbatha left
dukeleto NotFound: there wouldn't be a penalty for loading bytecode, but there would be for running Winxed scripts that interface to GMP 20:39
NotFound: and I just want both to be an option
NotFound dukeleto: winxed is a compiler, you don't need to run scripts if you don't want to. 20:40
dukeleto google-opensource.blogspot.com/2011...oogle.html 20:44
i just found that. Don't know how I missed it.
NotFound: i buy that
bubaflub: i would say that just keep the actual loading of stuff via NCI in PIR, and you can use winxed or whatever to hack on higher-level bindings 20:45
bubaflub dukeleto: sounds great.
dukeleto: i'll make a note of it in my proposal on melange tonight
dukeleto bubaflub: but create a gh issue for the get_string bug you are running into 20:47
NotFound BTW I'm not a mentor but I'll be happy to help with any winxed related task or doubts.
bubaflub dukeleto: okey dokey.
NotFound: you going to be around over the summer? 20:48
dukeleto bubaflub: this is the meta-plan: Everytime you get stuck for more than 15 minutes on one thing, create a github issue and describe it in detail, then hack on something else
atrodo NotFound> Why is winxed not a GSoC project?
20:48 dodathome left
bubaflub dukeleto: roger that. 20:48
lucian dukeleto: i thought winxed was supposed to be exactly at the same level of abstraction as pir
NotFound bubaflub: I'm always areound.
dukeleto bubaflub: when you get stuck on code, switch over to docs/tests, and vice versa
lucian: no
lucian: winxed is javascript-ish
lucian: i would say the syntax sugar is sweeter in winxed 20:49
bubaflub lucian: i'd like to use Winxed because object stuff is easier than PIR
lucian: also, syntax is a bit nicer
lucian dukeleto: sure, syntax. but i don't see how it's slower
i don't see where the overhead is
dukeleto lucian: yes, we already figured that out. It comes out in the wash when you compile to bytecode
tadzik I wish pct had code generation as good as winxed 20:50
NotFound atrodo: I've not imagined any Winxed task able to be a GSoC project.
Also, Winxed is my toy, I don't want other kids playing with it ;) 20:51
lucian NotFound: rewrite everything that's PIR to Winxed?
bubaflub NotFound: ah, that's why it's on SVN (-:
atrodo NotFound> Well, that makes all my arguments to make it GSoC moot :D
NotFound bubaflub: that's just an historic accident. These guys convinced me to use svn, then they switched to git :D 20:52
bubaflub hahaha. nice.
dukeleto NotFound: if you think of a gsoc idea, I can write it up
cotto_work NotFound: I'm not sure how you could see our use of svn as a convincing argument for its use.
dukeleto NotFound: surely there is some low-hanging-fruit that can be expanded into a 3.5 month project?
NotFound cotto_work: just because it was better than cvs. 20:53
atrodo I'm sure winxed docs could use some love
bubaflub NotFound: if you'd ever like help converting to Git on GitHub, let me know
cotto_work NotFound: ok. It's hard to argue against that.
NotFound dukeleto: the problem is that there is no long term plan, I design features while implementing them. 20:54
20:55 bbatha joined
NotFound But if you desperately want a suggestion for a task, here is one: write a Winxed (in its actual state) compiler targeting jvm. 20:56
cotto_work NotFound: be careful. That's what happened to PHP.
dalek tracwiki: v44 | dukeleto++ | GSoc2011 20:57
tracwiki: trac.parrot.org/parrot/wiki/GSoc201...ction=diff
lucian NotFound: that'd involve porting lots of parrot guts, wouldn't it?
NotFound cotto_work: I don't know much PHP, but my impresion is that they designed features *after* implementig them ;)
lucian: most probably. 20:58
lucian NotFound: no, they just didn't bother
cotto_work NotFound: or that
lucian NotFound: what does this do in php: $a = array(1, 2, 3)[1];
NotFound lucian: I guess $a = 2 ? 20:59
lucian NotFound: no. it's a syntax error
expressions aren't always expressions in php
NotFound Oh, nize.
lucian you have to do $v = array(1, 2, 3); $a = $v[1]; 21:00
when someone first told me that, i didn't believe them till i tried myself
NotFound I think winxed stage 0 still has some pitfalls like that.
And probably will ever have, but it's only a bootstrap tool. 21:01
$ winxed -e 'say([1, 2, 3][1]);' \\n 2 21:03
$ winxed --stage=0 -e 'say([1, 2, 3][1]);' \\n ##eval##:1: Syntax error: Expected ')' (found: '[' ) 21:04
Yeah, the same :D
lucian NotFound: heh. congratulations, your throw-away compiler is as good as php's 21:05
NotFound He
lucian i'm not sure if it's praise or not 21:06
NotFound At least it can disagnose the problem. 21:07
lucian indeed. tried in php? 21:08
bubaflub lucian: php
<?php $a = array(1, 2, 3)[1]; echo $a; ?>
Parse error: syntax error, unexpected '[' in - on line 2
NotFound I don't think I have a php cli installed.
cotto_work probably for the best
lucian bubaflub: yeah, a bit worse 21:10
dalek tracwiki: v45 | dukeleto++ | GSoc2011 21:12
tracwiki: trac.parrot.org/parrot/wiki/GSoc201...ction=diff
21:14 spinclad left
dukeleto just did a lot of gsoc stuff 21:18
bubaflub dukeleto++ 21:20
21:28 rohit_nsit08 left 21:31 ambs left
dalek rrot/jit_prototype: 16e2eec | bacek++ | t/compilers/opsc/21-jit-files.t:
Refactor test to expose "runloop" explicitely.
 21:31
rrot/jit_prototype: dc87ca6 | bacek++ | / (2 files):
Add Ops::JIT.module to access generated LLVM module. Rework test to initialize engine once only
bacek aloha humans
21:34 davidfetter is now known as humans
humans aloha bacek 21:34
21:34 humans is now known as davidfetter
bacek :) 21:34
21:37 bluescreen left 21:54 bubaflub left 21:56 bbatha left, perlite left 21:58 kid51 joined, perlite joined 22:00 bbatha_ joined
soh_cah_toa so i've been doing some nqp hacking to see if it's suitable for the debugger project and i have a few questions... 22:00
22:01 PacoLinux left
dukeleto soh_cah_toa: ask 'em 22:01
soh_cah_toa nqp doesn't have a traditional perl module system. you don't say "use Foo::Bar; my $for := Foo::Bar.new();" instead the objects available are the pmcs in src/pmc right?
dukeleto soh_cah_toa: in general, you don't need to ask to ask in here :) 22:02
soh_cah_toa agh, meant to say $foo
dukeleto soh_cah_toa: not quite. NQP does have a basic kind of object
soh_cah_toa: i would read the source of Plumage (which is written in NQP) to see a nice application written on top of NQP
soh_cah_toa: you can have "modules" in NQP
moritz nqp-rx doesn't have 'use' though 22:03
(nqp does)
dukeleto soh_cah_toa: which are basically just nqp files that you "use", but it is called something different
moritz: touche
moritz: in either language, tho, you can "drop down" to PIR
soh_cah_toa: ^^^
soh_cah_toa what's the difference between nqp and nqp-rx?
dukeleto soh_cah_toa: Q:PIR in nqp, iirc
i am saying nqp to mean the newer flavor, nqp-rx 22:04
soh_cah_toa: nqp-rx is a newer implementation of nqp that has nice features like regexen
lucian soh_cah_toa: nqp is un-cool and square
soh_cah_toa okay
dukeleto soh_cah_toa: rx stands for "regex"
soh_cah_toa: you want to use nqp-rx
lucian soh_cah_toa: nqp-rx isn't
btw, i'm finding i'm almost liking perl6. shocking, to me
tadzik well, the New nqp is nqp, not nqp-rx
dukeleto lucian: OH THE HUGE MANATEE 22:05
tadzik: it is all very confusing
tadzik yeah
soh_cah_toa yeah me too. it still feels like a different language though
dukeleto nqp was called nqp originally, then the newer nqp was called nqp-rx, but now we are calling nqp-rx nqp and you have to say "the old nqp" to mean the nqp before nqp-rx. Can't we afford a new name?
soh_cah_toa it almost feels like another languages based off of perl
dukeleto soh_cah_toa: NQP = Not Quite Perl 6, i.e. a desugared subset of Perl 6 22:06
soh_cah_toa too many nqp's!
cotto_work "snqp"
dukeleto soh_cah_toa: so yes, it should feel perlish :)
cotto_work still not quite perl 6
lucian ENQPSTACKOVERWLOW
dukeleto cotto_work: i like where you are going with this
yet another nqp = yanqp 22:07
soh_cah_toa another thing...the pir::new() statement. am i right in assuming that anything prefixed w/ pir:: is a pir opcode?
dukeleto soh_cah_toa: yes
lucian ENOTQUITEERROR
soh_cah_toa okay. is that just the syntax or is there a pir namspace and pir::something is it's fully qualified name? 22:08
dukeleto soh_cah_toa: it is syntax to access PIR 22:09
cotto_work It could progress. not quite perl 6 -> still not quite perl 6 -> not entirely unlike perl 6 -> pretty close to perl 6 -> ...
dukeleto soh_cah_toa: Q:PIR { ... } means that ... is PIR and pir::xxx means you are calling the xxx PIR opcode
lucian cotto_work: scheme's towers come to mind
dukeleto my god, it's made of onions! 22:10
cotto_work dukeleto: I don't know how I feel about that op. We had enough trouble with hcf.
soh_cah_toa okay. by the same respect, is there a Q:PASM { ... } ?
dukeleto apologizes to Arthur C. Clarke
soh_cah_toa: no
soh_cah_toa okay
dukeleto soh_cah_toa: humans shouldn't write PASM unless you are an unlucky person
cotto_work dukeleto: +1 22:11
soh_cah_toa everyone says that but i don't think it's that bad really
i enjoy low level stuff
dukeleto soh_cah_toa: don't ask the definition of an unlucky person, unless you want to become one :)
soh_cah_toa haha
dukeleto soh_cah_toa: that ain't the point
soh_cah_toa: PIR is low enough. PASM is just desugarred PIR
soh_cah_toa: it is unnecessarily painful for the developer
soh_cah_toa: you should always code in PIR and above 22:12
soh_cah_toa: makes sense?
soh_cah_toa aw man, so i can't do my gsoc proposal in binary?
lucian of course, personal opinion differs. i'd remove pir from the list of reasonable choices
dukeleto soh_cah_toa: for instance, you have to manually keep track of registers in PASM. You don't want to have to do that.
dukeleto goes to the post office
lucian soh_cah_toa: however, PIR is still assembly. the fullness of that fact only hits you when you try to do some flow control 22:14
soh_cah_toa yeah, what the heck...goto's?
anywhere that goto is necessary is assembly 22:15
lucian yeah, you get goto. want something else? tough
tcurtis doesn't consider C to be quite assembly. 22:16
tcurtis does consider goto to be occasionally necessary for proper error handling in C. 22:17
lucian tcurtis: sure. that's C's fault :)
soh_cah_toa tcurits: yeah but goto's aren't necessary in c. they only acceptable place for them is in nested loops
lucian soh_cah_toa: actually, they're very good for interpreters
davidfetter where else would you put one? :P
soh_cah_toa really? well, i don't know much about writing interpreters
but that's why i'm here 22:18
lucian soh_cah_toa: it's intuitive. opcode dispatch means jumping around into your interpreter code, which is C
you could do a switch, but it's not as straightforward or fast 22:19
lucian doesn't know much about writing interpreters either
soh_cah_toa though i did make a c interpreter a long time ago. it was far from efficient though 22:20
tcurtis They're nice for error handling in general. Much nicer that nested ifs.
s/that/than/
soh_cah_toa tcurtis: you mean the goto statement? 22:21
tcurtis soh_cah_toa: Yes.
darbelo As an aside. The efficient 'clever goto' technique of interpreter implementation isn't actually standard C.
soh_cah_toa hmmm...i suppose. just one if w/ several goto's inside 22:22
ahh...the age old flame war. to goto or not to goto?
darbelo It's based on 'computed gotos' which are about 30% more harmful to control flow purist :)
lucian shoves fingers in ears and starts yelling at darbelo 22:23
soh_cah_toa ha 22:24
darbelo Although if you really want to traumatize people you can spring some of gcc's more esoteric C extensions on them. 22:25
lucian darbelo: it's in CPython as a configure option, so i mark it as ok in my head
soh_cah_toa b/c cpython said so? ;) 22:26
lucian soh_cah_toa: YEAH
actually, because lots and lots of people use it in practice and it's fine
darbelo In my experience, most's people define 'portable C' as 'GCC compiles it on several platforms' :) 22:27
lucian darbelo: yep. enough for me :)
soh_cah_toa ha. yeah me too 22:28
22:28 whiteknight joined
lucian CPython does this the right way, which is nice 22:28
darbelo: but in fact, i think on many compilers you get a speedup with gotos anyway 22:29
even if it's not precomputed goto
soh_cah_toa all this talk of python...maybe i'll pick up a python book after gsoc.
22:30 wknight8111 joined
tcurtis soh_cah_toa: for an example of what I was talking about with C error handling using goto: gist.github.com/902605 22:30
lucian soh_cah_toa: i'm afraid i exude the stuff
darbelo Probably. IIRC cotto had a paper which did some comparisons on various dispatch mechanisms, computed goto was the fastest, and dumb switch was the slowest.
dalek rrot/jit_prototype: c1dbe0f | bacek++ | t/jit/jitted.ops:
Add PCC ops to jitted.ops.
 22:31
rrot/jit_prototype: ac5d73b | bacek++ | t/compilers/opsc/data/07.pir:
Add test for simple PCC calls.
rrot/jit_prototype: 50e04be | bacek++ | runtime/parrot/library/LLVM/Type.pm:
[llvm] Add Type.opcode_t
rrot/jit_prototype: dbdc85f | bacek++ | runtime/parrot/library/LLVM/Module.pm:
[llvm] Add stubs for handling Module.global
rrot/jit_prototype: 8ab0fe5 | bacek++ | compilers/opsc/src/Ops/JIT.pm:
Add few more variable and function aliases to Ops::JIT
rrot/jit_prototype: f5806fb | bacek++ | compilers/opsc/src/Ops/JIT.pm:
Implement calculation of special ops size
rrot/jit_prototype: 1116095 | bacek++ | compilers/opsc/src/Ops/Trans/C.pm:
Add CUR_OPCODE define back. It's acually used
darbelo All of which is mostly irrelevant to parrot, since we are too slow in other places for this kind of optimization to give any sort of benefit :)
soh_cah_toa tcurits: after your labels, all you have is a function call. wouldn't it be easier to just call the function within the if block? 22:32
lucian darbelo: true. and depending on JIT style, fast interpretation might not matter
dalek rrot/opsc_llvm: 5a77625 | bacek++ | / (3 files):
Revert "link libLLVM only to llvm_engine.so, not to libparrot"

This reverts commit b704d9e17d9ac71c26e17e51caf43167e3b62f73.
rrot/opsc_llvm: 8cf3d99 | bacek++ | runtime/parrot/library/LLVM/Type.pm:
[llvm] Add Type.opcode_t
rrot/opsc_llvm: 3ba281e | bacek++ | runtime/parrot/library/LLVM/Module.pm:
[llvm] Add stubs for handling Module.global
tcurtis soh_cah_toa: I could, but I'd have to call foo_cleanup() in three places and foo_bar_cleanup in two. 22:33
22:33 pjcj left
lucian soh_cah_toa: think of it as try/catch/finally, but crap 22:33
tcurtis And in actual cases, you sometimes get more than two things that could go wrong in a single function. 22:34
darbelo Everything can go wrong in every single function :)
soh_cah_toa alright, i see your case. but then it comes back to the old "performance vs. readability" trade off 22:35
i prefer readability and let the compiler handle any optimizations 22:36
to each his own though... 22:37
on a completely different and previous subject...does anyone know what the Q in Q:PIR { ... } stands for? 22:38
tadzik Quote probably 22:40
soh_cah_toa i don't like using constructs because "that's the way it is". i like to know what i'm typing
22:40 pjcj joined
soh_cah_toa tadzik: that was my first guess 22:40
22:44 wknight8111 left 23:05 hercynium left
tcurtis soh_cah_toa: well, actually, the goto-style shouldn't be a significant performance difference. It's useful because it avoids duplicated code which is a maintainability and readability win. 23:10
23:11 bubaflub joined
tcurtis may be excessively biased against nested if's by his use of 8-space tabs for indentation on most personal code. 23:21
sorear soh_cah_toa: yes, it stands for 'quote' 23:23
cotto_work 8-space tabs? You think you know a guy...
;]
bacek_at_work ~~
23:23 dmalcolm left
tcurtis cotto_work: Truthfully, it's mostly just that I've been using vim without any configuration as my primary editor lately. :) 23:25
lucian tcurtis: that's a double-sin! 23:26
first you use 8 spaces, then you don't customise vim
you may require penance 23:27
tcurtis lucian: But the 8-space tabs are a consequence of the unconfigured vim. 23:28
lucian tcurtis: exactly! 23:29
you're even breaking causality! :)
tcurtis :) I almost got away with it, too. 23:32
dalek rrot/lorito: 6d0e704 | cotto++ | src/dynpmc/m0ctx.pmc:
make register use a little more consistent
 23:33
rrot/lorito: 781dd89 | cotto++ | src/dynpmc/ (3 files):
make m0ctx "work"
rrot/lorito: bf688af | cotto++ | src/dynoplibs/m0proto.ops:
add some TODOs
 23:34
23:37 kid51 is now known as kid51_at_dinner 23:39 petdance joined 23:41 darbelo left 23:47 lucian left
petdance so how do I look at whiteknight/imcc_compreg_pmc ? 23:48
whiteknight all squinty-like
petdance It's a remote branch 23:49
and I don't know how to look at that
whiteknight if you have git, do a git checkout whiteknight/imcc_compreg_pmc
petdance Right, I have git
and I'm normally on origin/master
ah, ok 23:50
I'm gonna splint it
whiteknight awesome
petdance++
mikehh petdance: I finally managed to get make splint to run - needed to set C_INCLUDE_PATH=/usr/include/i686-linux-gnu to pull in ffi.h 23:54
petdance huh, ok
mikehh I am now testing Ubuntu 11.01 beta i386 - will let you know 23:55
petdance: also I had to load Perl::Critic::Deprecated to get Perl::Critic::Bangs to work (well tests - haven't run it yet) 23:56
petdance yeah, that needs to get shipped. 23:57
that's been fixed.
mikehh bah Ubuntu 11.04 beta

Powered by Cro and the Raku® Programming Language.

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

Thank you!