00:21 vendethiel joined 04:25 llfourn joined 04:54 vendethiel joined 09:05 domidumont joined 09:07 patrickz joined 09:10 domidumont joined 09:55 virtualsue joined 10:31 dalek joined 11:06 vendethiel joined 12:13 virtualsue joined 12:26 leont joined 12:43 dalek joined 13:21 colomon joined 13:25 colomon joined 14:03 geekosaur joined
flussence has anyone here ever successfully built rakudo+moar with --has-libffi? 14:39
15:23 kjs_ joined 16:11 zakharyas joined 16:23 vendethiel joined
hoelzro o/ #moarvm 16:30
flussence: I have
flussence I can't get it to work on gentoo :( both moar and rakudo assume ffi.h is /usr/include/ffi.h 16:32
hoelzro flussence: have you tried CFLAGS=-I/lib/ffi/location perl Configure.pl --has-ffi ?
it probably won't compile the probe if -lffi isn't in your compiler's default search path, though 16:33
sadly, Configure.pl's probe doesn't respect LDFLAGS (although it may now)
flussence I've put a makefile patch for that on moarvm's github issues, can't find anywhere appropriate to do the same for rakudo though.
(anywhere in Makefile-Moar*.in, that is) 16:34
16:53 mohij joined 17:34 TimToady joined 18:41 brrt joined
brrt ping konobi 18:46
19:10 mojca joined
mojca does moarvm with jit work on 32-bit platforms at all? 19:11
flussence nope 19:13
brrt no 19:20
what kind of platform are you looking for?
ugh, my designed tile thingy is 152 bytes large even when leaving out the list pointers :-( 19:29
19:38 llfourn joined
mojca i386-darwin 19:38
in Configure.pl I see "if ($Config{archname} =~ m/^x86_64|^amd64|^darwin(-thread)?(-multi)?-2level/) { ... then enable jit"
but for darwin there doesn't seem to be any distinction between 32-bit and 64-bit
timotimo brrt: too big for a cache line or what's the matter? 19:39
brrt just big and ugly
timotimo hmm
brrt mojca, oh, wow
yes, that piece of code is Stupid, and my fault too
i didn't know any better way to detect the architecture in perl 19:40
i still don't, for that matter
mojca but I checked the configure parameters and there is no really straightforward variable that I could see
leont archname is not a very portable concept
brrt no, it isn't
mojca I have some very complex code to detect whether a mac is 64-bit capable 19:41
brrt mojca, the JIT works on amd64 posix and windows architectures, and nothing else
mojca ok, moderately complex
leont perl5's configure takes it from uname (at least on Unixy systems)
brrt hmmm
that's something, yes
but, not correct either
because it disregards architecture 19:42
mojca no problem, I don't think the user who was hit by this really cares about JIT, it would only be helpful if it didn't throw a bus error
brrt eh, wait, let me restate that
well, it's something that has to be reworked one day or another
mojca on OS X probably the most reliable way would be to check for compiler flags
sorry, not flags, defines
brrt leont: we shouldn't assume target and host architecture are equal, and we do 19:43
python setuputils also makes that mistake
mojca what's the best quick solution to disable jit before the configure "bug" gets fixed?
leont True
brrt --no-jit
mojca ok
oh, wait, one more thing
brrt ask away :-) 19:44
mojca what if a user builds universal binaries?
brrt hahahahaha
no
mojca also --no-jit just to be on the safe side?
brrt probably yes
mojca I can certainly build universal binaries
brrt i can't and never tried...
mojca but I'm not sure if I can run them :)
brrt if you make a universal binary, does it include two architecture shared objects 19:45
leont Basically, yes
brrt hmm
leont In theory, ELF supports that too, just no one implemented it much beyond proof of concept, because why would you? 19:46
geekosaur arch i386 somecommandhere
er
brrt then basically, architecture-dependent compiler definitions could be made to do config, and we'd never care about a Configure.pl script at all
geekosaur arch -32 or arch -64
brrt because it'd then be the job of the compiler to get the configuration right 19:47
well, with some exceptions like cgoto support
stuff that would have to be probed for
mojca I opened a new ticket: github.com/MoarVM/MoarVM/issues/324 19:50
Just curious, is powerpc supported? 19:51
I'm not asking about JIT of course
flussence people have built on arm, so I don't see why not... 19:52
mojca Just for information: LuaJIT works on i386-darwin, but I believe there is specific code that has to be written for perl
Maybe I'll try the ppc just for fun
nwc10 Power 8 worked last I tested it, 32 and 64 bit, but some tests failed 19:53
brrt i think powerpc is supported in principle
nwc10 IIRC jnthn fixed what he thought might have been the cause of those
mojca ok, thank you
nwc10 but I've not had a chance
to retest
brrt luajits support for powerpc or not is basically separate from moarvm :-)
we don't actually share any luajit code beyond the build process
and couldn't, even if we wanted
mojca luajit on powerpc-darwin is broken
I can compile luajit with some patches, but then random code crashes 19:54
(even though most of it works; but those crashes make it useless)
also, isn't it slightly wrong to use $Config{archname} or $Config{anyotherthing} to determine the architecture of the binary we are building? 19:59
I can imagine that one could easily use x86_64 perl to build MoarVM for PPC for example 20:00
[Coke] hey, we got a failure report for moarvm on os x 10.6 20:14
nwc10 and it wasn't from me :-)
[Coke] mojca++ again. :) 20:15
nwc10 also, apologies in advance. My client might be here, but I'm not hugely around currently. 20:16
brrt mojca, yes, it is terribly wrong; but the question is, what would be a better strategy 20:21
i do know the answer, by the way
compile a binary that invokes the compiler with the correct flags, based on what your compiler thinks is the right thing... except that doesn't work for cross-builds either 20:22
aargh
mojca in a way one should probably treat every Mac OS X compilation as "cross-compilation" just in case
brrt probably, yes
mojca not in the sense that one should expect --host and --target flags
I'm not sure what your compilation chain does 20:24
but in my opinion the proper way to determine the architecture on OS X would be to actually run the compiler and grep for some defines; something similar that autoconf does (I'm not saying I like autoconf) 20:25
brrt Configure.pl builds a Makefile from build/Makefile.in, based on probing and args from the environment, and the makefile than builds all the things 20:26
mojca then again ... if this is just for JIT, running the compiler would be a bit of an overkill
brrt nah, we already probe for other stuff
it's a completely valid approach
mojca on mac one could theoretically set "export CXX=powerpc-apple-darwin10-g++-4.0" 20:27
I didn't check whether that name exactly is a valid compiler name 20:28
but one would get powerpc binaries by calling that compiler
and most things would work out of the box
though usually one would use export CXXFLAGS="-arch ppc" to achieve the same
brrt any binary name is valid, i think
mojca valid in the sense of "actually existing in the wild" 20:29
(shipped by Apple)
anyway, the name itself doesn't actually matter
20:29 virtualsue joined
brrt no.. you can always do make CC=powerpc-apple-darwin10-g++-4.0 20:30
mojca the problem is that 10.6 would return "i386" when asked for uname, but it would build x86_64 binaries by default
brrt hmm
that's ugly
mojca well, on ARM I don't know how to distinguish between armhf and armel; and on MIPS I don't know how to distinguish between mips and mipsel either 20:31
without actually running the compiler
the uname output is equal on both variants, but binaries are incompatible
and I'm still fighting with that problem in our distribution of some other binaries; the only luck we have is that there are practically no users of those archs 20:32
brrt should do an experiment probably with that
mojca what's the best place to ask a few simple questions about panda? #perl6? 20:34
[Coke] sdtimes.com/christmas-perl-arrives/ - that article seems confused. 20:36
ww
20:39 llfourn joined
brrt mojca, yes, that's the right place 20:51
you should ask tadzik
20:55 vendethiel joined
tadzik hm? 21:02
oh, panda, yes
mojca tadzik, I tried to create a package for panda for macports, but failed because I don't quite understand how it's supposed to work 21:04
by default it installs files to a completely weird location
and if I move it, it breaks with "/opt/local/bin/panda: line 2: syntax error near unexpected token `('" 21:05
"/opt/local/bin/panda: line 2: `sub MAIN(:$name is copy, :$auth, :$ver, *@, *%) {'"
but I should probably move the discussion to #perl6?
maybe that error is because perl6 in shebang doesn't work 21:06
tadzik oh? 21:07
timotimo src/6model/reprs/NFA.c:442:17: warning: format ‘%lld’ expects argument of type ‘long long int’, but argument 6 has type ‘MVMint64’ [-Wformat=] 21:09
fprintf(stderr,"%c with %ds target %lx offset %lld\n",cp,(int)numcur, (long)target, offset);
^- what could cause this? a "strange" platform?
tadzik overexcited compiler, imho :) 21:10
well, or a platform where long long int is not a 64bit int
timotimo well, it's only a debug statement 21:12
21:38 llfourn joined 22:24 colomon joined 22:39 llfourn joined 22:50 zakharyas joined 23:17 colomon joined 23:29 virtualsue joined 23:41 llfourn joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!