error during bootstrap on OS X #1089
Comments
|
I woke up a bit further, and realized that I know how to type |
|
Try again with the latest master. I fixed a small llvm type error that could maybe cause something like this. If that doesn't work, you might have to do a more thorough rebuild due to changed system libraries. |
|
No dice. Same error. And I just rebuilt everything on the system because On Fri, Jul 27, 2012 at 3:36 PM, Jeff Bezanson <
|
|
Is this using apple's LLVM or the one we download and build? |
|
I'm the maintainer of the Homebrew install process, and we download/compile As far as the C compiler goes, I believe this is using Clang 4.0, provided On Fri, Jul 27, 2012 at 5:19 PM, Jeff Bezanson <
|
|
Aha --- we are now primarily targeting LLVM 3.1. We made some effort to continue to support 3.0, but I don't test it regularly so I suspect that could be the problem. |
|
It's possible that's the issue...... but this compiled just fine on OSX 10.7 not three days ago, so unless there have been significant changes in the last two days, I'm thinking it's more likely something else is freaking out. I'll try to get LLVM 3.1 installed and see if that fixes it. UPDATE: |
|
#1092 reports the same problem on Ubuntu 12.04. |
|
No problem on my Lubuntu 11.10 VM. I've taken a snapshot and will do the upgrade to 12.04. |
|
I have the same issue on my mac and I still have OS 10.7 It started on Thursday. After that I updated from xcode 4.1 or 4.2 to 4.4 and it still gave the same "error during bootstrap". I have tried it multiple times with clean installs with the same result. |
|
I have encountered such a problem on an Ubuntu 12.04 amd_64 system. I compiled Julia with clang 3.1 (installed from an Ubuntu ppa) using the system llvm 3.1 (also from the ppa) and the system libreadline. When I enabled the definitions starting with if false
# simple print definitions for debugging. enable these if something
# goes wrong during bootstrap before printing code is available.
length(a::Array) = arraylen(a)in JULIA usr/lib/julia/sys.ji
Warning: New definition writeSegmentation fault (core dumped)
make[1]: *** [/home/bates/build/julia/usr/lib/julia/sys.ji] Error 139 |
|
And reproduced after upgrading the VM to Lubuntu 12.04. Maybe gcc or glibc? |
|
Can't be a problem with gcc, it still exists if you build with clang. |
|
Yep, right, good point. I'll be taking a shot at bisecting this. Wish me luck. UPDATE: Infuriatingly, now I cannot reproduce. I was doing builds with |
|
make clean && make did not work for me, same error on a mac with os 10.7 |
|
Damn, I also get this on wheezy - even using the llvm 3.1 - I'm only new to Julia today, so can't help more than this at present... (gdb) r -b sysimg.jl The program being debugged has been started already. Start it from the beginning? (y or n) y Starting program: /julia/usr/bin/julia-debug-basic -b sysimg.jl [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". Warning: New definition write Program received signal SIGSEGV, Segmentation fault. 0x00007ffff6bc92cf in jl_method_table_assoc_exact (mt=0x64e9f8, args=0x7fffffffd590, n=2) at gf.c:199 199 jl_value_t *ty = (jl_value_t*)jl_typeof(a0); (gdb) bt #0 0x00007ffff6bc92cf in jl_method_table_assoc_exact (mt=0x64e9f8, args=0x7fffffffd590, n=2) at gf.c:199 #1 0x00007ffff6bcc244 in jl_apply_generic (F=0x7d0420, args=0x7fffffffd590, nargs=2) at gf.c:1211 #2 0x00007ffff6bcfac3 in jl_apply (f=0x7d0420, args=0x7fffffffd590, nargs=2) at julia.h:875 #3 0x00007ffff6bd1d13 in jl_show (stream=0x0, v=0x101c070) at builtins.c:555 #4 0x00007ffff6bcb576 in check_ambiguous (ml=0x64eaa0, type=0x101c070, sig=0x7d0440, fname=0x7bd5e0) at gf.c:942 #5 0x00007ffff6bcb8dd in jl_method_list_insert (pml=0x64ea78, type=0x101c070, method=0x101c050, tvars=0x13edab8, check_amb=1) at gf.c:997 #6 0x00007ffff6bcbc01 in jl_method_table_insert (mt=0x64ea68, type=0x101c070, method=0x101c050, tvars=0x13edab8) at gf.c:1068 #7 0x00007ffff6bcccb1 in jl_add_method (gf=0x7d0480, types=0x101c070, meth=0x101c050, tvars=0x11a9898) at gf.c:1405 #8 0x00007ffff6c1ab62 in jl_method_def (name=0x7bd5e0, bp=0x7d8120, bnd=0x7d8118, argtypes=0x101c070, f=0x101c050, t=0x11a9898) at toplevel.c:394 #9 0x00007ffff56dd110 in ?? () #10 0x00007fffffffec70 in ?? () #11 0x00007fffffffd7f8 in ?? () #12 0x0000000000000003 in ?? () #13 0x00007fff00000000 in ?? () #14 0x00007fffffffd950 in ?? () #15 0x000000000101c070 in ?? () #16 0x000000000101c050 in ?? () #17 0x00000000011a9898 in ?? () #18 0x00007fffffffd900 in ?? () #19 0x00000000004018a0 in jl_show@plt () #20 0x0000000000000000 in ?? () (gdb) Thanks if you can help out. |
|
I'm always building from scratch, Homebrew does a @11Kilobytes; Clang was just upgraded to 4.0, let's not rule out compiler choice just yet. |
|
I meant the other way around--left-over cruft from deps could be causing builds to succeed. I'm going to try another bisect tonight with an added |
|
I pushed something to get past the segfault under jl_show. |
|
I ran into this problem too. I'm running Ubuntu 11.10 x64. This is the first time I've tried julia, so I didn't have an existing install. I was doing everything from scratch. I reverted back to fadd70d and still had the same problem. I then reverted back to 6f0b890 from Thursday and it worked. I noticed something different when I was compiling the different versions. The versions that didn't work ended with: The version that worked had this in its place: I didn't see the reference to sys0.ji anywhere in the faulty builds. |
|
There have been some major changes and |
|
With your recipe I get a failure to open the file "build_h.jl" when running base/sysimg.jl Breakpoint 1, jl_raise (e=0x7ac8980) at task.c:525 Breakpoint 1, jl_raise (e=0x798f050) at task.c:525 make clean && make base/build_h.jl && make testall succeeds |
|
"make clean && make base/build_h.jl && make testall" worked for me. It did say "Warning: replacing module Base" |
|
I got the same thing as @dmbates when using aa029e3 |
|
Aha! That's it! One of my commits broke building of |
|
And it was so broken that |
|
That fixed it for me, thanks! |
|
Ah, nice. So it looks like af6e780 addresses this issue? It works for me in any case. @staticfloat, @dmbates, @timema, @pao, @11Kilobytes — can you guys chime in yea or nay on this fixing things? |
|
This (or some other change from today's commits) also solved the problem for me as well. |
|
I did a clean install with the latest and it worked for me. |
|
It works sometimes, but still hangs for me every now and then. On Mon, Jul 30, 2012 at 1:42 PM, timema <
|
|
After af6e780 the build works fine for me. Good to be able to build again. |
|
No dice, I deleted the julia folder cloned into it again and then when I run make the following error happens: JULIA usr/lib/julia/sys.ji |
|
Hmmm, actually, I am getting the same error as @11Kilobytes now. When simply running When running When running in |
|
@JeffBezanson - Is fixed for me now, thanks! |
|
Right now, I installed readline, LAPACK and BLAS using apt-get since I couldn't make them from source. Also I used Clang as the compiler. I am running Ubuntu 12.04. This obviously isn't the only way to build Julia, that's why I am going to make a fresh install of Ubuntu 12.04 on a VM and see what it takes to build Julia. |
|
That sounds very helpful 11K, thanks. A HOWTO based on that will be useful to a lot of people. |
|
Pinging you on this, as I still have the problem described above. |
|
Looks pretty nasty. One thing to try is uncommenting the call to "verifyFunction" at codegen.cpp:177, which might give better llvm diagnostics. |
|
Same errors, even with it uncommented. (I ran a I'm going to try building this outside of Homebrew, letting Julia download On Thu, Aug 2, 2012 at 11:43 AM, Jeff Bezanson <
|
|
I'm seeing an error on bootstrap but in a different location: repl-readline.c: In function âinit_historyâ: make[1]: *** [/n/home00/pedmon/test/julia/usr/lib/julia/sys.ji] Error 1 This is building on RedHat 5.1, but I also get it on CentOS 6.2 |
|
@pedmon, this is likely a different issue alltogether, so I suggest you start a new thread, as this one is focused on the issues involved in building on OSX. |
|
Will do. |
|
Folks, I am using clang 4.0 with Xcode 4.4.1 on OS X 10.7.4, with no issues. I believe the toolchain is the same as that on OS X 10.8. @staticfloat any updates? Did it work for you outside the homebrew environment? Since the bootstrap issue is fixed, and you have random hangs, perhaps we can have a new issue to track that, with some reproducibility? There was a clang issue with random hangs, which was fixed (#1013), but it appears to be earlier to your bug filing. |
|
Hey Viral, Yes, it did work outside of the homebrew environment. I've narrowed it down to likely be an issue with Homebrew-built FFTW; if I have Julia build her own FFTW, it works, if I use the Homebrew-supplied FFTW, it doesn't. I'm trying to get another mac homebrew user that is interested in Julia (samueljohn) to verify, but he's been pretty busy as of late. If you know anyone else that would be interested in testing this out for me, send 'em my way. As of yet I don't know that enabling the compilation of FFTW within julia works for anyone else. |
|
Actually, the error has disappeared, but been replaced with a different one! This one seems to be independent of |
|
Does this error occur outside of the homebrew environment as well? Is this is on Mountain Lion as you had said? I haven't yet updated to Mountain Lion, but soon will. @JeffBezanson can you take a look at the gist posted in the comment above? |
|
By golly, that was useful. This was an llvm type error, which is not necessarily caught unless assertions are enabled in LLVM. Fixed. |
|
Nice. Once everyone confirms that this is fixed for them too, we can close this. I actually think this issue has been causing sporadic flakiness, at least on OS X, for some time now. |
|
Alright, we're back to where we were before, with homebrew-provided EDIT: To be clear, JeffBezanson's fix got rid of the "new" error, but the "old" error persists. |
|
it has been fine for me for the last 3 weeks |
|
Progress! I've discovered that copying Homebrew's fftw libraries into Julia's |
|
Alright, I'm closing this issue as I've managed to solve it, finally! I have to symlink in the |
|
Ah interesting, @staticfloat, that "homebrew not in /usr/local" was the difference, why julia works on my Mac and not on yours. Ok good to know it's solved now. I somehow wished julia would respect LDFLAGS and CFLAGS, so that all the libs homebrew provides (btw. currently llvm 3.1 with the option to have the |
|
It's not an LDFLAGS problem, it's actually a DYLD_LIBRARY_PATH problem, and
|
Hey there, updated to 10.8 and I'm doublechecking that OSX 10.8 can still handle Julia's awesomeness. Everything looked good up until the
JULIA usr/lib/julia/sys.jibuildstep. At that point, stuff just dies witherror during bootstrap.I ran
make debug, and got a slightly more informative error message:Next debugging steps?
The text was updated successfully, but these errors were encountered: