<< 05-05-2014 >>
00:05:57dom96Varriount: why do you need the entire overlapped structure?
00:06:05dom96and what "choice bits" are you referring to?
00:08:55*milieu joined #nimrod
00:14:21Keletdom96: If I have some magic macro that can do var f = toProc("x^2 + 3x"), how can it know that f is of type var f: proc(x: float): float or something like that?
00:14:30KeletFor example is it possible to do something like
00:15:01Keletvar f = toProc("x^2 + 3y") and have it make a 2-sized parameter list based on order of introduction without explicitly knowing it at compile time?
00:15:10Keletah
00:15:16Keletall this macro stuff has to be done at compile time
00:15:20Keletdoh.
00:15:55Keletsuppose the best you could do is something like var x = calc("x^2 + 3x", 5) which is meh
00:16:10Kelet(to calculate f(5))
00:16:35dom96No, it's pretty simple
00:16:46dom96as long as all your params are floats and the return type is a flaot
00:16:47dom96*float
00:17:02dom96you may be able to use 'auto' but type inference still isn't perfect
00:17:07Keletyeah, that's a fair assumption
00:17:46KeletBut how is this done if say I make a math-evaluating REPL?
00:17:52KeletI thought macros were compile time only
00:19:30dom96they are.
00:19:37EXetoCyes they are, but an AST is then returned which is compiled
00:19:40dom96If you want a REPL then just write a normal app
00:20:16reactormonkbut how would you execute the repl?
00:20:51EXetoCis it a REPL?
00:21:58EXetoCbut yeah, you just parse a string at runtime. no need for any magic
00:29:32fowlmmmm digit := chr({'0'..'9'}).save((match:string) -> Node => IntNode(i: parseInt(match)))
00:29:59*brson joined #nimrod
00:34:39dom96good night
00:44:16KeletEXetoC: Can I make a proc at runtime?
00:55:42*Demos joined #nimrod
01:00:25EXetoCKelet: no. parse it and decide what to do at run time
01:01:14*brson quit (Quit: leaving)
01:09:06EXetoCa proc can be loaded from a share lib, but that's only necessary for plugin systems and such
01:13:04EXetoCwhat you do is, you parse it and then you turn that into a tree, which you then have to validate and such
01:13:42EXetoCmight be complicated. good luck ;) bye
01:24:01Demoscan I put a nimdoc comment right before a type section that contains only one type?
01:50:46*rustypipe joined #nimrod
01:51:28rustypipebesides the tutorials on the website, are there any other tutorials for learning how to program with Nimrod?
01:57:09Demoswell the manual is quite good, and there is that "nimrod for C programmers" document that Skrylar wrote
01:57:27*brson joined #nimrod
01:58:11EXetoCskyfex?
01:59:14Demosoh maybe
01:59:36Demosyeah it was skyfex, sorry skyfex
02:21:40*Zuchto quit (Remote host closed the connection)
02:30:53*Skrylar quit (Ping timeout: 264 seconds)
02:34:29*q66 quit (Quit: Leaving)
02:37:32*rustypipe left #nimrod (#nimrod)
02:50:05*Demos quit (Ping timeout: 264 seconds)
03:05:07*perturbation joined #nimrod
03:05:35perturbation?
03:06:03perturbationsorry all, disregard... will be back later
03:06:04*perturbation quit (Client Quit)
03:10:31*brson quit (Quit: leaving)
03:13:38*Demos joined #nimrod
03:14:05*Skrylar joined #nimrod
03:38:05Demoswhat platform defines do we have?
03:38:09Demosany for OpenBSD?
03:38:14DemosBSD in general?
03:47:19VarriountDemos: What do you mean by defines? Like, defined(Windows), defined(Linux), etc?
03:48:08VarriountDemos: Also, remember you can dump defined symbols to output using "nimrod dump file.nim"
03:48:21Varriountdom96: You still there?
03:49:27Varriountdom96: I need access to the Event member, or otherwise some way of attaching extra data to the overlapped structure. Specifically, I need to associate a buffer pointer with the overlapped structure.
03:49:45Demosyeah
03:49:56Demoswait I am not dom96, as it turns out
03:50:06Demosand yeah I mean that kind of define
03:50:16Demoswhen not defined(OpenBSD) seemed to work
04:07:19*TylerE quit (Ping timeout: 252 seconds)
04:08:28*TylerE joined #nimrod
04:37:38Varriountdom96: I think I found away around not having direct access to the Overlapped structure with the callback (however you still might want to think of ways that others will want to extend asyncdispatch)
04:39:52*Demos quit (Quit: WeeChat 0.4.2)
04:45:53reactormonkok, apparently we still need moar documentation.
04:46:45Varriountreactormonk: What gave you that silly idea? :3
04:58:54*isenmann joined #nimrod
05:04:06VarriountHi isenmann
05:47:55*Skrylar quit (Ping timeout: 250 seconds)
07:00:03*Skrylar joined #nimrod
07:13:07*vegai left #nimrod (#nimrod)
07:29:17njoejoehow to seek() and tell() on a PFileStream (and also on TFile)?
07:35:25*nande quit (Remote host closed the connection)
07:35:59Araqnjoejoe: I don't think this is possible for PFileStream, dunno about TFile
07:36:17AraqI usually use memfiles when I don't read sequentially
07:38:22njoejoeok, i will look into memfile. have multi GB files and was hoping for no mmap.
07:42:02Araqwell you can easily wrap seek and tell if they are missing
07:42:11Araqnimrod's TFIle is C's FILE*
07:42:55*bjz quit (Ping timeout: 276 seconds)
07:58:25*Zuchto joined #nimrod
08:02:41njoejoeOh I found TFile::setFilePos() which i think will do it. hopefully if I setFilePos() on the underlying file of the PFileStream, things will work out.
08:04:18SkrylarAraq: is there already a convenient jmp_buf and setjmp in the stdlib? i was poking around the other day and png needs
08:04:31Skrylari fiddled it in to building, wasn't sure if we already had it
08:08:57njoejoeyes, that worked. but also i found PFileStream::setPosition() which also appears to work. So I'm using that.
08:10:18*wan quit (Ping timeout: 240 seconds)
08:11:39Araqnjoejoe: ah good. I shouldn't answer questions to the language or the stdlib anymore ...
08:11:52Araqmore often than not, I'm wrong
08:12:15njoejoeyou've forgotten more than most will ever know :-)
08:34:41*DAddYE quit (Remote host closed the connection)
08:35:17*DAddYE joined #nimrod
08:39:30*DAddYE quit (Ping timeout: 240 seconds)
09:15:48*faassen joined #nimrod
09:33:54*zahary1 joined #nimrod
09:35:54*zahary quit (Ping timeout: 240 seconds)
09:37:19*BitPuffin quit (Ping timeout: 276 seconds)
09:38:11*BitPuffin joined #nimrod
09:38:44*faassen quit (Quit: Leaving.)
09:43:03*faassen joined #nimrod
09:45:25Skrylarneat
09:45:29Skrylargiflib's API is reasonable
09:45:44*faassen left #nimrod (#nimrod)
10:26:42*wan joined #nimrod
10:53:19AraqSkrylar: let us know when every C library in existance has a Nimrod wrapper :-)
10:53:52EXetoCthe 2000 most popular once will do
11:08:14*zahary joined #nimrod
11:10:14*io2 joined #nimrod
11:11:04EXetoC*ones ..
11:11:15*zahary1 quit (Ping timeout: 265 seconds)
11:14:07dom96Varriount: That is how you're meant to do it. You can capture anything inside the callback, there is no need to extend the overlapped structure
11:23:42*slacko15949 joined #nimrod
11:30:38Araqhi slacko15949 welcome
11:36:14*slacko15949 quit (Ping timeout: 240 seconds)
11:49:29*wan quit (Ping timeout: 255 seconds)
12:03:45*wan joined #nimrod
12:08:57Skrylarlol
12:09:08Skrylari'm not working on *that* many
12:10:40milosnhey hey
12:14:58*wan quit (Ping timeout: 240 seconds)
12:21:26*kshlm joined #nimrod
12:37:04*q66 joined #nimrod
12:37:04*q66 quit (Changing host)
12:37:04*q66 joined #nimrod
12:46:36Araqhi kshlm welcome
13:07:17*io2 quit (Quit: ...take irc away, what are you? genius, billionaire, playboy, philanthropist)
13:42:53*darkf quit (Quit: Leaving)
13:43:22*menscrem joined #nimrod
14:03:14milosnAraq: hi, you got spare moment? you mentioned that i can do python style setattr() and getattr() in nimrod ... care to give me couple of one-liners i can start from? :)
14:03:32milosnor proc names in typeinfo module, and i can take it from there myself
14:50:13*Jesin quit (Quit: Leaving)
14:55:17*bjz joined #nimrod
15:01:08fowlmilosn, if you consider the case where you do not use a dynamic field
15:01:26fowlmilieu, setattr(obj, "x", 42) is obj.x = 42
15:13:18*milieu quit (Ping timeout: 240 seconds)
15:21:55*brihat joined #nimrod
15:22:00*brihat quit (Client Quit)
15:24:00*chemist69 joined #nimrod
15:33:09*Matthias247 joined #nimrod
15:35:14*chemist69 quit (Quit: leaving)
15:43:23*noam_ joined #nimrod
15:43:43*bjz quit (Ping timeout: 252 seconds)
15:47:18*noam__ quit (Ping timeout: 240 seconds)
16:03:17*untitaker quit (Ping timeout: 264 seconds)
16:08:45*untitaker joined #nimrod
16:11:37EXetoCshould this work? var arr = cast[ptr array[4, int8]](addr someObj)[]
16:11:59EXetoCnot that all those indirections are needed, but it's the first thing I tried
16:15:52*Matthias247 quit (Read error: Connection reset by peer)
16:18:43fowlEXetoC, yes
16:19:55EXetoCyeah looks right. reporting it
16:26:33fowlwhats wrong
16:27:03EXetoCinvalid C code is generated
16:27:11fowlEXetoC, you are casting the data TFoo to a pointer
16:27:29fowlEXetoC, use cast[array[..]] or cast the address of foo/i
16:27:50EXetoCtypo. doesn't matter
16:30:01EXetoCit should result in a runtime error in that case
16:30:10EXetoCunless one is really lucky :>
16:30:48fowlwhat u need this for
16:30:58fowlcast[ptr array[4,int8]](foo.addr) works for me
16:32:38EXetoCI upgraded to GCC 4.9 recently. it might have something to do with that
16:39:45*Jehan_ joined #nimrod
16:46:00EXetoCfowl: what are you using?
16:51:44fowl4.8
16:53:26*Jesin joined #nimrod
16:55:04*lesshaste joined #nimrod
16:55:06lesshastehi
16:55:16fowlhello
16:55:18lesshastewhat's the status of using C libraries in nimrod?
16:55:23lesshasteI mean libraries written in C
16:55:44lesshasteor just generally the use of libraries at all
16:55:47fowllesshaste, highly possible
16:55:54Jehan_It's fairly straightforward for the most part.
16:56:09lesshastewhat do you have to do? I do a lot of scientific computing
16:56:10fowlmany examples are available on nimrods package manager, babel
16:56:19lesshasteand I am very impressed by the speed you get from nimrod
16:56:20Jehan_It's generally a question of whether c2nim can handle it automatically or whether you have to parts manually.
16:56:35lesshasteoh you mean actually translate the C?
16:56:40lesshastefowl, thanks.. I should look there
16:56:47Jehan_Well, Nimrod's speed is more or less C's, modulo how much you use GC.
16:56:57Jehan_Since it compiles to C.
16:56:59fowllesshaste, for a library you just need to bridge it to nimrod by defining the interface
16:57:02lesshasteJehan_, right but some C libraries are seriously tuned. BLAS for example
16:57:09lesshastefowl, ok thanks
16:57:13Jehan_Yes, of course.
16:57:37fowlit looks like this; proc SDL_SetVideoMode (w,h,bpp: cint): PSurface {.importc, dynlib:"libSDL.so".}
16:57:44lesshasteah ok
17:00:40*DAddYE joined #nimrod
17:12:18*BitPuffin quit (Ping timeout: 265 seconds)
17:18:30milosnfowl: ah setattr() ... is there an getattr()?
17:18:47milosnill try setattr() when i get to that bit
17:19:04Araqjust read the whole module
17:19:09Araqit is not big
17:19:54Araqbasic approach: convert your thing to a TAny
17:19:58fowlmilosn, to retrieve field X from obj Y, the syntax is Y.X
17:20:17Araqfowl: that is not what he means
17:20:56Araqbbl
17:21:13milosnAraq: reading type info docs as we speak ... there is no mention of setattr() nor getattr() :)
17:21:19milosnok catch you later
17:23:19EXetoCprobably not, but see marshal.nim for TAny in action.
17:25:22*Jesin quit (Quit: Leaving)
17:25:41EXetoCmilosn: and here's a similar construct: http://build.nimrod-lang.org/docs/manual.html#dot-operators
17:27:54*Skrylar quit (Ping timeout: 276 seconds)
17:32:59EXetoCdom96: how do you indicate failure with async? 'quit' I guess
17:33:42EXetoCbut everything is wrapped in async atm and I don't need really need to take that approach in this case
17:35:05*skyfex quit (Quit: Computer has gone to sleep.)
17:35:39*skyfex joined #nimrod
17:45:36dom96EXetoC: raise an exception
17:54:14reactormonklesshaste, you could code BLAS in nimrod, it is powerful enough. And if needed, include ASM blocks. nimrod has an `asm`, but that is for C code, so double asm if you need some assembler.
17:57:01*boydgreenfield joined #nimrod
18:00:09EXetoCdom96: it should work now?
18:00:16dom96yes
18:02:49milosnhmm
18:02:50milosnbase/Base_domains.nim(34, 23) Error: conversion from TAny to TBase_domains is invalid
18:03:06milosndoes anything specific need to be done for type conversion to work?
18:04:40milosnbasically I have user defined type TDomain which inherits from TBase_domains ... in TDomain init function i need to set some important variables so the new TDomain object is passed to init function of TBase_domains
18:04:48milosnthats how i ended there :)
18:05:03EXetoCdom96: ok so it doesn't propagate all the way to the default exception handler, end the current exception is nil when doing "try: await stuff() except E_Base: ..."
18:05:23milosnlet me try without type conversion
18:05:25*Jesin joined #nimrod
18:05:40dom96milosn: store your object as TBase_domains not a TAny
18:05:59*BitPuffin joined #nimrod
18:06:00dom96milosn: or show us your code so we can help you better
18:06:15dom96EXetoC: Yeah, getCurrentException won't work.
18:07:04dom96EXetoC: but you should be able to catch EBase
18:08:52milosndom96: TBase_domains is auto-magicaly generated code that describes the table 'domains' in the database, user defines TDomain which inherits TBase_domains ... TBase_domains knows nothing about how the user is going to use it
18:09:40milosnand since i can not init fields of an object in the type description ... i need init proc
18:10:21milosnwhere i can only receive as an argument user defined type TDomain ... as TAny, on which i need to init important fields
18:10:24milosn:)
18:11:06dom96once again I will have to ask you to show me your code
18:11:43milosnoke, its couple of files, if i dont figure it out ... ill have to paste someplace
18:11:57EXetoCdom96: is it a temporary limitation?
18:11:58dom96gist
18:12:24dom96EXetoC: Yeah, should be a simple s/getCurrentException/future.error/
18:12:34dom96that the macro has to do
18:14:07lesshastereactormonk, thanks
18:16:42EXetoCthat works, assuming that it's called directly
18:20:04EXetoCare other approaches much more complicated?
18:21:40dom96IMO it would be nice to have a way to capture the exception in the 'except' block.
18:21:46reactormonklesshaste, do you use asm frequently?
18:21:51lesshastereactormonk, never
18:21:58dom96like: except varName: EOS :
18:21:59lesshastereactormonk, I just want to use scientific libraries
18:22:01dom96or something, but that looks weird
18:22:44reactormonklesshaste, yup, then wrap them and you should be good. c2nim the header files and you should be able to use them. However, c2nim isn't perfect, so you will have to remove some stuff from the header file and port it yourself
18:23:15reactormonkthere's some pragma pushing you can do to reduce the verbosity of the wrapper file, lemme see...
18:23:31lesshasteok thanks
18:23:37lesshastethis may be beyond me at the moment
18:23:38lesshastebut thanks
18:24:10dom96lesshaste: what kind of scientific computing do you do?
18:24:23lesshastedom96, mostly uses of linear algebra libraries
18:24:32lesshastecalculating eigenvalues, rank etc.
18:24:53reactormonklesshaste, you can push and pop pragmas if you want to apply them to multiple procs
18:25:01dom96lesshaste: perhaps this would be of use to you then: https://bitbucket.org/BitPuffin/linagl
18:25:16lesshastedom96, looks good!
18:25:22lesshasteI hope they are not reimplementing it all!
18:25:30lesshastethe amount of reimplementation of standard libraries drives me spare
18:26:08dom96although IIRC there may be some problems with that library due to compiler bugs :\
18:26:21dom96EXetoC: Actually, it looks like I can just set currException
18:26:44dom96EXetoC: if it was exported D:
18:28:40*BitPuffin quit (Ping timeout: 265 seconds)
18:28:42*nande joined #nimrod
18:30:28EXetoCdom96: so, set it in the beginning of the block, and then unset at the end? if it's even that simple
18:31:17dom96yea
18:32:50*OrionPK quit (Remote host closed the connection)
18:40:11*OrionPK joined #nimrod
18:42:36*Demos joined #nimrod
18:48:18*angus joined #nimrod
18:48:26angushello
18:49:36dom96hi angus
18:49:38angushow can I initialize an array with an index or elements of range type?
18:50:01angustype
18:50:01angus RangeA = range[1..3]
18:50:01angus RangeB = range[1..10]
18:50:01angusconst Arr: array[RangeA, RangeB] = [1, 2, 3]
18:50:23angusI get:
18:50:42reactormonkangus, well, you're not giving it an array of ranges
18:50:58anguswell I can't paste the error.
18:51:22angusreactormonk: ?
18:51:30reactormonkangus, I guess it's complaining about array of ints vs. array of ranges?
18:51:42angusI give it an array of integers inside the range
18:52:00angusreactormonk: yes.
18:52:14angusError: type mismatch: got (Array constructor[0..2, int]) but expected 'array[1..3, RangeB]'
18:52:29angusI can't also cast it
18:52:29dom96[1.RangeB, 2, 3]
18:52:31fowllesshaste, what library do you need wrapped? i'll take a look at it and see if it is easily doable wiht c2nim
18:52:43reactormonkdom96, ah cool
18:52:43EXetoCimplicit conversion only works for the elements after the first one, and the first element decides the type
18:52:51angusaha!
18:52:56EXetoCand as such you can do things like [1i32, 2, 3]
18:53:01reactormonkthought that wasn't possible ^^
18:53:13reactormonkis the range type checked at runtime?
18:53:36EXetoCit might perform the check at compile time in some cases
18:53:41dom96IMO angus' code should just work
18:53:43angusdom96: great, it works!
18:53:48angusthanks
18:53:51dom96without the need for 'RangeB'
18:53:52reactormonkdom96, well, that would be return type overload
18:54:03dom96but Araq says it's tricky
18:54:07dom96so this is necessary for now.
18:54:12dom96angus: np
18:54:32lesshastefowl, thanks! basically lapack
18:54:46lesshastehttp://www.netlib.org/lapack/
18:54:54reactormonkdom96, didn't know about the ranges as types O.o
18:55:19dom96reactormonk: ranges are indeed types :P
18:55:23EXetoCNatural and Positive in system are good examples
18:56:05EXetoCand maybe those should be generic, but with int by default. doesn't work yet though
18:56:26reactormonkEXetoC, then we get into typeclasses
18:57:10EXetoCyes
18:59:04angusI think I'm giving up on range types for now
18:59:25dom96angus: how come?
18:59:38angusI thought having a range parameter would help to check argument values automatically
19:00:02angusbut it forces me to cast the arguments to the range type
19:00:15*BitPuffin joined #nimrod
19:00:18EXetoCnot just the first?
19:00:18angusif I have proc x(a: RangeType)
19:00:39angusI can't call it as x(1). I have to write it as x(RangeType(1))
19:00:44EXetoCI've never needed to explicitly cast, except when dealing with tuples
19:00:55dom96sounds like a bug
19:01:13fowlthat is most definitely a regression or something
19:01:22EXetoCdo the underlying types match?
19:01:45anguswhere can I paste?
19:01:50anguspastebin?
19:01:57EXetoCgist?
19:02:13EXetoCsome people don't like pastebin
19:02:25dom96gist supports nimrod
19:05:27angusok, here it is: https://gist.github.com/anonymous/998daa3b3c150dad2915
19:05:48angusproblem is in the object constructor, not the proc invokation as I mistakenly thought
19:05:52Araqreactormonk: nimrod's "asm" is mapped to C's "asm" section, no need for 2 asms or something
19:06:10angus Error: type mismatch: got (int, TMonthRange, TDayRage)
19:06:16anguswhy?
19:07:42Araqbecause object constructors look like this:
19:07:54AraqTDate(year: year, month: m, day: d)
19:08:07Araqbtw hi angus welcome
19:08:08lesshastefowl, maybe someone is already working on lapack?
19:08:19lesshastefowl, given the linear algebra link someone pasted before
19:08:27angusAraq: same error, just with the labels
19:08:31lesshastehttps://bitbucket.org/BitPuffin/linagl
19:09:00angusAraq: wait! no
19:09:08angusthat works. I was using "="
19:10:00angusAraq: thanks :)
19:11:20*ehaliewicz joined #nimrod
19:12:08*Trimsty joined #nimrod
19:14:32reactormonklesshaste, ^ I stand corrected
19:15:04lesshastereactormonk, sorry.. about what?
19:15:09reactormonklesshaste, asm
19:15:21lesshastereactormonk, oh I see :)
19:15:29*Trimsty quit (Client Quit)
19:19:33*boydgreenfield quit (Read error: Connection reset by peer)
19:19:56*boydgreenfield joined #nimrod
19:21:21*boydgreenfield quit (Read error: Connection reset by peer)
19:21:53*BitPuffin quit (Ping timeout: 264 seconds)
19:22:14*boydgreenfield joined #nimrod
19:22:58Varriountdom96: Is there a way to schedule a repeating callback yet? or a timed callback?
19:24:00*boydgreenfield quit (Read error: Connection reset by peer)
19:24:54*boydgreenfield joined #nimrod
19:28:22fowllesshaste, lapacke.h looks like it should translate without much modification
19:28:31lesshastefowl, great!
19:28:46fowlfreaking big though, mine is > 16k lines
19:28:59lesshastefowl, yes it's a very important library
19:36:03*BitPuffin joined #nimrod
19:39:03*Jesin quit (Ping timeout: 250 seconds)
19:39:48*menscrem quit (Ping timeout: 240 seconds)
19:42:09*nande quit (Ping timeout: 245 seconds)
19:45:40dom96Varriount: no
19:46:04*Matthias247 joined #nimrod
19:49:59*Jesin joined #nimrod
19:52:10Araqhi Matthias247 wb
19:52:20Matthias247hi
19:55:09*BitPuffin quit (Ping timeout: 240 seconds)
19:59:55Araqmilosn: var foo: MyObject
20:00:09Araqvar a = foo.toAny()
20:00:28Araqa["fieldName"] = 14.toAny
20:01:33Araqfor name, val in a: echo name
20:03:06milosnAraq: thanks, i havent got to that part yet ... :)
20:03:26milosnill note it down as a comment in the code, so i dont have to ask again ;)
20:03:57Araqoh yeah, get rid of your DomainBase OO crap
20:04:11Araqit won't help you
20:04:25milosnyes, slowely i am coming to realise that myself
20:04:35milosnit just wont "bend" that way
20:05:30Araqyou should do it entirely differently
20:05:46milosnlike? i am open to suggestions :)
20:05:55Araqwe already have an SQL parser
20:06:17milosnyou know better how nimrod will "bend" ... but i am not making another ORM-ish DSL
20:06:20milosn:)
20:06:26*milosn dont belive in DSL that does SQL
20:06:37Araqyou should write a tool that connects to a database and asks it for the table descriptions
20:06:52*Jehan_ quit (Quit: My MacBook Pro has gone to sleep. ZZZzzz…)
20:06:54milosnhmm ok, and than?
20:07:09milosnthats what my "mygenbase" does, and generates .nim base files
20:07:10Araqthis tool then parses the SQL and generates *type-safe* nimrod wrapper glue
20:07:52Araqand then all you need to do is to ensure the tool works decently with nimrod's staticExec
20:08:06Araqyou could instead write it all as macro
20:08:43Araqbut I think the macro VM is still not up for this task
20:09:14Araqplus people don't really want to recompile the ORM tool whenever they recompile their application
20:09:36milosnhmmm thats cool, i could do that in an afternoon and have basic AR ORM in place (or couople fo afternnons :P) ... but we lack the most fundamental bit of AR style ORMs ... you cant extend your DB objects
20:09:45milosnadd useful procs etc.
20:09:57*boydgreenfield quit (Ping timeout: 240 seconds)
20:10:03milosnAraq: other option is ... DSL for SQL, i dont wanna do that :)
20:10:19Araqwell I agree
20:10:25AraqSQL already is a DSL
20:10:49Araqthe major problem is that it is not type-safe
20:11:01*boydgreenfield joined #nimrod
20:11:02Araqwhen you use sql"foo bar" in Nimrod
20:11:13Araqand that sucks
20:11:35*boydgreenfield quit (Client Quit)
20:11:39milosnhmmm
20:12:08Araqperhaps you should generate a domSql proc that checks the SQL at compile-time
20:12:24Araqer, I mean a macro
20:13:10Araqwell it depends on what you're after
20:13:22Araqyou said "extend the DB objects"
20:13:33AraqI said "type safety" ;-)
20:13:56milosnsomething easy to use :) ... and get the job done (imaginary job, i dont do anything in nimrod yet) :D
20:14:07milosnbtw what does this tell me: test.nim(15, 10) Error: expression cannot be casted to TDomain
20:14:28Araqdon't use 'cast'
20:14:39milosnuse conversion?
20:14:44Araqyes
20:15:00Araqand if type conversion doesn't work, you're doing it wrong
20:15:12milosnive tried, i cant do a single one ... is there some pragma or something which i am missing?
20:15:27*milosn looks at the manual again
20:16:46milosnoh spoke to soon ... my cast:
20:16:47milosn var base: TBase_domains = initTBase_domains()
20:16:48milosn var d = TDomain(base)
20:16:49*lesshaste quit (Quit: Leaving)
20:16:52milosnthat worked
20:16:55milosnerr conversion
20:16:58milosnnot cast
20:17:10milosnlet me see if test code works when i run it :P
20:17:37milosnAraq: if i manage to "bend" it the way i want it ... you are gonna hate me dude
20:17:41milosnbut thats life :P
20:18:08Araqyes. don't worry. I hate everything anyway.
20:18:16milosn:)
20:18:37AraqI'd focus on type safety and on generating prepared statments though :P
20:19:00Araqthese are clear goals and you end up with something better than the db_* modules
20:19:46Araqthe alternative is to go after buzzwords and people will continue to use the db_* modules because they work. :P
20:20:12milosn:D
20:20:47milosnyou got lucky ... conversion compiles, but dont actualy work
20:21:57Araqfor inheritance you should really work with 'ref' btw
20:22:07Araqeverything else is tricky
20:22:27Araq(plus there is still some compiler bug lurking in there...)
20:28:08njoejoevar buckets: array[0..9, string] gives "ambiguous identifier: 'string' -- use a qualifier" how to do it?
20:28:26Araqsystem.string
20:28:53njoejoethank you
20:33:49*boydgreenfield joined #nimrod
20:43:54njoejoeis there a way to tell array to initialize the strings to ""? if i buf[x] &= "appending" i get sigsegv unless I first loop through the array and initialize the strings. also, why system.string instead of plain "string" as type?
20:46:41EXetoCjust 'string' should be enough as long as it hasn't been exported in one of the imported modules
20:47:41fowlnjoejoe, you defined string?
20:48:17njoejoefowl: no. perhaps because i'm using 0.9.5 version of nimrod?
20:48:21EXetoCnjoejoe: there's no such shortcut I think, but add (&=) will basically behave like safeAdd eventually, where a nil target will be assigned the content of the second argument rather than resulting in a crash
20:48:45fowlnjoejoe, "ambiguous identifier: 'string' -- use a qualifier" means there is another string defined
20:49:17EXetoCsuch bugs might still exist
20:49:59fowlnjoejoe, proc init (arr: var openarray[string]) = for idx in 0 .. high(arr): arr[idx] = ""
20:50:17njoejoefowl: my entire program just consisted of the line: var buckets: array[0..9, string] so perhaps something with 0.9.5
20:50:22EXetoCit's not required at all times anyway. that'd break about a million things
20:51:06fowlnjoejoe, i cant reproduce
20:51:10EXetoCnjoejoe: I just compiled that compile. no errors
20:51:47EXetoCso no imports?
20:52:35njoejoeOMG. my testfile was called "string.nim" i think that was the problem... oops.
20:55:05Araqyeah a common gotcha
20:55:26Araqbut I don't know what can be done about it, explicit "module foo" lines are meh
20:56:50dom96why can't the compiler disambiguate that?
20:57:07Araqoh it can easily
20:57:14Araqbut I can't do everything
20:57:34Araqand disambiguation only is about overload resolution currently
20:57:44dom96then why the mention of explicit "module foo" lines
20:58:07njoejoeI don't think it's a problem. i'm pretty sure i'll never do it again.
20:58:14Araqbecause it's not always possible to disambiguate
20:58:44Araqnjoejoe: yeah but pretty much every newbie stumbles over that
20:58:52Araqbut oh well
21:01:33Araqoh btw ... I am now misusing my power to tell you guys: Foxit Reader made me install the Yahoo Search Bar accidentically.
21:01:52Araqand that's it. I'll never use it again.
21:06:13Araqthat thing is an impertinence sondergleichen
21:06:41dom96You're right. We should put that in Nimrod's installer.
21:06:45dom96To make some bucks.
21:06:46Araqeine unglaubliche Unverschämtheit
21:08:26boydgreenfieldhi all, quick question – is anyone using the memfiles module? And if so, to actually create mmap files, or just to write them? I get an EOS / bad file descriptor error whenever I try to create a new mmap file, but perhaps I’m misunderstanding the usage.
21:11:43Araqboydgreenfield: the compiler itself uses it but only for reading
21:11:54Araqmaybe writing is broken :-/
21:12:22boydgreenfieldAh ha. Well, if this usage is correct, then I think it probably is :) :mm = memfiles.open("/tmp/test2.mmap", mode = fmReadWrite)
21:13:44Araqlooks good yeah
21:13:59Araqplease fix it and make a pr
21:14:26boydgreenfieldYep… looking into it now.
21:18:17boydgreenfieldOk. Sorry it works fine. I made the mistake of using `mappedSize` instead of `newFileSize`.
21:18:29reactormonkboydgreenfield, o/
21:18:45boydgreenfieldI.e., mm = memfiles.open("/tmp/test2.mmap", mode = fmReadWrite, mappedSize = 1000) instead of mm = memfiles.open("/tmp/test2.mmap", mode = fmReadWrite, newFileSize = 1000)
21:18:49boydgreenfieldYes. Embarassing. Sorry.
21:19:01reactormonkboydgreenfield, write a comment in the documentation of the method
21:19:08reactormonkyou probably won't be the last one
21:19:59fowlaye
21:20:02fowlgive the module some examples
21:20:22reactormonkis it just me or should mappedsize and newfilesize the same in case of writing?
21:20:48reactormonkor mappedsize and newfilesize are for read/write, and not to be used together?
21:21:14Araqthey are orthogonal concepts if you ask me
21:21:49reactormonkAraq, in memory, the other on the disk?
21:22:05Araqyes
21:22:16reactormonkdoes the size in memory matter when you're writing?
21:22:55boydgreenfieldTo confirm: the proper branch to submit a PR to is `devel` yes?
21:23:06Araqdepends, you can write past its end to append to the file
21:23:08*flaviu joined #nimrod
21:23:21Araqboydgreenfield: right
21:23:22reactormonkboydgreenfield, yup
21:23:32reactormonkAraq, ok, point taken
21:25:19reactormonkAraq, found a downside to my new $ for floats. 5.0 is "5".
21:26:05Araqreactormonk: patch it afterwards
21:26:20Araqappend ".0" when necessary
21:26:23reactormonknot sure what's the right way to do it
21:26:34Araqit is, trust me
21:26:47fowllol
21:26:54fowlreactormonk, lets just see how ruby does it
21:27:16reactormonkfowl, no idea anymore where the code is :-/
21:27:29*Jehan_ joined #nimrod
21:27:36flaviuIs there a way to take a pointer and a length and call it a string?
21:27:59Araqonly with a copy, flaviu
21:28:03reactormonkflaviu, a cstring is just an array IIRC, so you should be able to cast it to one
21:28:06fowlreactormonk, lol, from mruby: > 5.00001
21:28:07fowl => 5.00000999999999
21:28:40fowlreactormonk, mruby has less code, i'll look there
21:29:21reactormonkfowl, try 5.0
21:29:42reactormonk... that's just float.
21:30:23fowlits "5.0"
21:31:37Jehan_Umm, IEEE 754 fp numbers can't represent decimal fractions accurately? That's nothing new.
21:32:17AraqJehan_: what's the best way to implement a barrier? I'm using a counter + a condition variable
21:32:20reactormonkfowl, huh
21:32:24fowlreactormonk, https://github.com/mruby/mruby/blob/master/src/numeric.c#L111
21:33:00reactormonkfowl, hmm
21:33:09Jehan_Araq: Umm. Second best is to handcode it in assembly. :)
21:33:11fowlhuge function
21:33:11reactormonkjust copy that one?
21:33:15Jehan_Best is not to use one. :)
21:33:52Jehan_Barrier usually means n-1 threads come to a halt => poor CPU utilization.
21:34:00Jehan_Also, thundering herd problem.
21:34:45Araqwell I need it for correctness and last I checked correctness is sometimes more important than speed.
21:35:45fowlreactormonk, yea but it doesnt look fun. give me about a week lol
21:36:59Jehan_I'll explain in a minute why I wasn't being facetious (or only partly). For now:
21:37:20Jehan_A condition variable is an acceptable temporary hack to provide the functionality.
21:37:39Jehan_It's likely not to be very efficient, though.
21:38:42Jehan_A semaphore-based solution is likely the better choice.
21:39:03Jehan_But we're now running into various implementation-specific questions.
21:39:39boydgreenfieldreactormonk / Araq: Comment on memfiles mappedSize / newFileSize. So there is an issue with them being orthogonal, because it following use case fails: mm = open(“/tmp/test.mmap”, newFileSize= 1024, mappedSize = -1)
21:39:52flaviuAraq: Sumatra PDF is small and super fast
21:40:00Jehan_First of all, some pthread implementations support only a subset of semaphore functionality.
21:40:26Araqmeh semaphores
21:40:39Araqit's been years since I last used one
21:40:40boydgreenfield(sorry bumped enter), which means that you can’t really use open() in readWrite mode without first checking if the file exists. Araq, from your perspective is this the desired behavior and one should just have a higher-level wrapper around the functionality? I’ll write up a longer docstring either way.
21:41:28Jehan_So you may have to write something different or write a fallback implementation.
21:41:32Araqboydgreenfield: I think this behaviour is fine, but if it's cheap to avoid, go for it
21:41:48AraqI only use locks or condition variables
21:42:01Araqor cas or atomic stuff
21:42:07Jehan_That's extremely portable, but not always sufficient.
21:42:10boydgreenfieldAraq: That’s fine, I’ll just document that one shouldn’t expect that to work.
21:42:51Jehan_In general: If you don't mind the speed hit from barriers, you may as well just have a thread wait for the remaining ones.
21:43:10Jehan_It's more likely to produce correct code (that's another problem with barriers).
21:43:47Jehan_If you do mind, what you actually want is the "sync results" and "wait for results to be complete" be separate steps.
21:44:47Jehan_So that the system doesn't grind to a halt each time you use a barrier and one thread is lagging a bit behind.
21:45:44Jehan_In general, that approach leads to a data flow architecture, though.
21:46:14Jehan_Because it's still one of the best ways to reason about the coordination of multiple threads in non-trivial situations.
21:46:59Jehan_If you just want to provide a basic barrier implementation, though, yeah, just use mutexes and condition variables. At least that's what I always did. :)
21:47:50Jehan_It's functional, and the implementation is robust without a number of special cases to consider and test.
21:47:56Araqwell I implemented my own "FastCondVar". wasn't faster at all, but introduced nice crashes
21:48:20Jehan_Heh. Been there, done that. :)
21:48:52Araqwas on windows though with my barely sufficient timing methods
21:48:54Jehan_The problem with rolling your own condition variables is that it's really hard to implement blocking efficiently and portably.
21:49:18Araqyeah but for that I uses the OS's cond var as a fallback
21:49:23Araq*used
21:49:47Jehan_For example, on POSIX it generally means that you're playing around with signals, meaning you're eating up a signal slot, plus the usual adventures in sigsuspend()/sigwait()/etc. semantics.
21:50:12AraqI pretend signals don't exist
21:50:27Jehan_They're unfortunately necessary to do certain things at all.
21:50:49Jehan_And wouldn't be as much of a problem if there were more signal numbers available (portably).
21:51:05AraqI hate all this Unix crap. It's incredibly bug prone.
21:51:54Jehan_I could say the same about some of the Windows stuff. :) Even though I don't share the usual Unix attitude about looking down on Windows.
21:52:04njoejoeAraq: what's your OS of choice?
21:52:32Jehan_Windows does some stuff amazingly well. Other things not so much. But that's true of all OSes.
21:52:42Araqnjoejoe: Haiku. (Political answer)
21:53:08njoejoe:-) nimos FTW!
21:54:12Jehan_But yeah, wait-while-stopping-the-world is a functionality that I wouldn't prioritize for wasting speedup efforts on.
21:54:13flaviuJehan_: The thing that windows does the best of all OSes is being compatible with windows
21:54:33Jehan_Chances are most of the system is going to sit idle, anyway, so who cares about speed?
21:55:32AraqJehan_: I don't really want to discuss this again, but
21:55:40Jehan_Heh. :)
21:55:42Araqlet foo = getFuture(input)
21:55:54Araqonly works when 'input' is on the heap
21:57:07Araqand that's something Groovy's dataflow variables don't have to consider
21:57:25Araqbut it surely matters *a lot* for Nimrod
21:58:54Jehan_True, but not all forms of dataflow have to involve futures.
21:59:27Jehan_Barriers are a pretty low-level operation regardless.
21:59:55AraqBarriers keep the stack frame from disappearing :P
22:02:05Araqalso iirc barriers are really cheap on a GPU
22:03:35Jehan_The problem with barriers is more that you're turning an N-core system into a 1-core system for a bit. :)
22:03:55Jehan_And then Amdahl's law hurts you.
22:06:23NimBotAraq/Nimrod devel d043854 Araq [+0 ±7 -0]: implemented builtin noncopying slice
22:06:23NimBotAraq/Nimrod devel a11ae92 Araq [+0 ±2 -0]: fixes #1140
22:06:23NimBotAraq/Nimrod devel 4055fd4 Araq [+0 ±1 -0]: some additions to winlean
22:06:23NimBotAraq/Nimrod devel fe690a2 Araq [+1 ±0 -0]: added test case for bug #1140
22:06:23NimBot1 more commits.
22:07:01AraqOrionPK: here you go. Finally I fixed it.
22:14:43AraqJehan_: what do you mean "dataflow without futures"?
22:15:21Jehan_Well, that dataflow is a concept that's not necessarily tied to specific programming language implementations.
22:15:30Jehan_As a very basic example, think petri nets.
22:16:16Araqsure
22:17:01Jehan_Futures are being used a lot these days because they have many attractive features in addition.
22:17:50NimBotAraq/Nimrod devel 77d5ab1 Araq [+0 ±2 -0]: bootstraps works again; get rid of compile-time 'rand' (too weird)
22:18:21*reloc0 quit (Read error: Operation timed out)
22:18:47Jehan_Barriers are generally more interesting for distributed systems, where this kind of synchronization is genuinely hard.
22:19:37*reloc0 joined #nimrod
22:19:51Jehan_For shared memory systems, I'd recommend to have a look at SNZIs (scalable non-zero indicators). These aren't barriers, but have related semantics that are also often useful.
22:42:35*xenagi joined #nimrod
23:00:38*nande joined #nimrod
23:01:27OrionPKAraq cool, I will test out the full templates lib later tonight
23:02:25boydgreenfieldAraq / reactormonk: Another question realted to memfiles. It appears that creating a new mmap-backed file only has the group read and execute permissions set. I’m looking through the system open() source code, and believe this is set here https://github.com/Araq/Nimrod/blob/devel/lib/posix/posix.nim#L1799. But at least this guide indicates that a mode constant has to be specified if using the O_CREAT flag, which
23:02:25boydgreenfieldI don’t see being done http://linux.die.net/man/2/open. Can anyone w/ more POSIX experience comment? It appears to me that the file permission flags aren’t being set properly becuase the call to open(filehandle, flags) should really be open(filehandle, flags, mode) since flags has O_CREAT set. Correct?
23:07:09*milosn quit (Ping timeout: 245 seconds)
23:10:01*Matthias247 quit (Read error: Connection reset by peer)
23:11:57Jehan_Araq: https://gist.github.com/rbehrends/2783e808550b1f11ae19
23:12:10Jehan_Some barrier code that doesn't suck entirely. :)
23:14:46OrionPKcompiler/semfold.nim(410, 5) Error: undeclared identifier: 'mRand' ?
23:15:42dom96yep, he broke the build.
23:20:02reactormonkboydgreenfield, int open(const char *path, int oflag, ...); according to the manpage, it seems like the wrapper is incomplete
23:20:05reactormonkboydgreenfield, just add your own.
23:20:46reactormonkboydgreenfield, oh wait, it's got the varargs pragma, so it should as you expect it to with posix.open(filename, flags, mode)
23:20:48fowlreactormonk, no, that is a varargs function
23:22:55dom96good night
23:23:27njoejoegood night dom96
23:25:11*milieu joined #nimrod
23:33:31boydgreenfieldreactormonk: Ah. Just picking this up and wasn’t familiar with the .varaargs. proc. I’ve patched it locally and will submit a PR for the memfiles module to do the same (doesn’t seem very useful to have the default behavior create a file you can’t re-open).
23:36:38*Skrylar joined #nimrod
23:49:40*Jesin quit (Quit: Leaving)
23:59:31*darkf joined #nimrod