Failed build on FreeBSD

[forkandwait]

Hi all,

I followed the directions for building on FreeBSD, including installing libunwind, but I get an error. Any pointers would be appreciated. See below for some info.

My machine (more specs avail if important):

FreeBSD mustafa 9.0-RELEASE FreeBSD 9.0-RELEASE #0: Tue Jan  3 07:15:25 UTC 2012     root@obrian.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  i386

The most recent git commit in the log is e4ca3a9:

commit e4ca3a9d13b7a501214a3673757ac8e28b6c4667
Author: Tim Holy <tim.holy@gmail.com>
Date:   Fri May 11 14:49:46 2012 -0500

gmake output:

julia/ (j=0,r=0)$ gmake
Submodule 'deps/libuv' (https://github.com/JuliaLang/libuv.git) registered for path 'deps/libuv'
Cloning into 'deps/libuv'...
remote: Counting objects: 10053, done.
remote: Compressing objects: 100% (2610/2610), done.
remote: Total 10053 (delta 7830), reused 9534 (delta 7390)
Receiving objects: 100% (10053/10053), 2.97 MiB | 376 KiB/s, done.
Resolving deltas: 100% (7830/7830), done.
Submodule path 'deps/libuv': checked out 'd48c5bd0adcb08fc8793529aed8a19530efde9ee'
In file included from src/unix/eio/eio.c:51:
src/unix/eio/ecb.h: In function 'ecb_unreachable':
src/unix/eio/ecb.h:333: warning: 'noreturn' function does return
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  245k  100  245k    0     0   105k      0  0:00:02  0:00:02 --:--:--  126k
Hmm...  Looks like a unified diff to me...
The text leading up to this was:
--------------------------
|--- dSFMT.h    2012-01-01 11:56:06.000000000 +0530
|+++ dSFMT-julia.h  2012-01-01 11:55:19.000000000 +0530
--------------------------
Patching file dSFMT.h using Plan A...
Hunk #1 succeeded at 1 with fuzz 2.
Hunk #2 failed at 186.
Hunk #3 failed at 233.
Hunk #4 failed at 254.
Hunk #5 failed at 273.
Hunk #6 failed at 284.
Hunk #7 failed at 296.
Hunk #8 failed at 307.
Hunk #9 failed at 319.
Hunk #10 failed at 330.
Hunk #11 failed at 342.
Hunk #12 failed at 365.
Hunk #13 failed at 380.
Hunk #14 failed at 396.
Hunk #15 failed at 412.
Hunk #16 failed at 428.
Hunk #17 failed at 438.
Hunk #18 failed at 448.
Hunk #19 failed at 459.
Hunk #20 failed at 472.
Hunk #21 failed at 500.
Hunk #22 failed at 509.
Hunk #23 failed at 518.
Hunk #24 failed at 528.
Hunk #25 failed at 538.
Hunk #26 failed at 548.
Hunk #27 failed at 558.
Hunk #28 failed at 568.
Hunk #29 failed at 581.
Hunk #30 failed at 594.
Hunk #31 failed at 607.
Hunk #32 failed at 619.
31 out of 32 hunks failed--saving rejects to dSFMT.h.rej
done
gmake[2]: *** [random/jl_random.c] Error 31
gmake[1]: *** [julia-release] Error 2
gmake: *** [release] Error 2

[pao: formatting]

[Keno]

Seems like our patch for dSFMT.h is not being applied properly. Is there a functional difference between the patch command on Linux and that on FreeBSD?

[forkandwait]

There are both patch and gpatch on FreeBSD, I just discovered prompted by your question. Analogous to make and gmake, I guess. So, probably, yes.

I am not sure the best way to set this up -- I don't want to link patch to gpatch in the file system as root (it will probably mess a pile of stuff up), but I do want to make it work for the entire build of Julia.

Ideas?

[Keno]

Can you replace all calls to patch by gpatch in deps/Makefile. If that solves your issue, I'll push a fix to master that automatically uses the right tool.

[forkandwait]

Changed as recommended, get the following errors:

julia/ (j=0,r=0)$ gmake
Submodule 'deps/libuv' (https://github.com/JuliaLang/libuv.git) registered for path 'deps/libuv'
Cloning into 'deps/libuv'...
remote: Counting objects: 10053, done.
remote: Compressing objects: 100% (2610/2610), done.
remote: Total 10053 (delta 7830), reused 9534 (delta 7390)
Receiving objects: 100% (10053/10053), 2.97 MiB | 376 KiB/s, done.
Resolving deltas: 100% (7830/7830), done.
Submodule path 'deps/libuv': checked out 'd48c5bd0adcb08fc8793529aed8a19530efde9ee'
In file included from src/unix/eio/eio.c:51:
src/unix/eio/ecb.h: In function 'ecb_unreachable':
src/unix/eio/ecb.h:333: warning: 'noreturn' function does return
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 245k 100 245k 0 0 84477 0 0:00:02 0:00:02 --:--:-- 126k
(Stripping trailing CRs from patch.)
patching file dSFMT.h
/usr/home/wsprague/julia/deps/install-name-FreeBSD.sh: not found
gmake[2]: *** [random/librandom.so] Error 127
gmake[1]: *** [julia-release] Error 2
gmake: *** [release] Error 2

[Keno]

can you try cp deps/install-name-Linux.sh deps/install-name-FreeBSD.sh?

[forkandwait]

Cranking away. Summary of changes: (1) patch to gpatch. (2) copied install-name. (3) changed the latter to /usr/local/bin/bash. (4) got rid of Microsoft Windows newlines in that file.

Thanks!
I will update when it is finished building.

[forkandwait]

Whoopsie! Here is the next error.

In file included from ../common.h:326,
from setparam_NANO.c:41:
../common_x86.h: In function 'blas_lock':
../common_x86.h:57: warning: implicit declaration of function 'sched_yield'
setparam_NANO.c: In function 'get_l2_size_old':
setparam_NANO.c:507: warning: unused variable 'cpuid_level'
setparam_NANO.c: In function 'init_parameter':
setparam_NANO.c:626: warning: unused variable 'l2'
gmake[3]: wget: Command not found
gmake[3]: *** [lapack-3.4.1.tgz] Error 127
gmake[2]: *** [openblas-v0.1.1/libopenblas.so] Error 2
gmake[1]: *** [julia-release] Error 2
gmake: *** [release] Error 2

[Keno]

This is an upstream issue. Just filed it here: OpenMathLib/OpenBLAS#106. For now, can you replace wget by curl in deps/openblas-v0.1.1/Makefile:262

[forkandwait]

Installed wget, restarted gmake, and got this error w/r/t lapack.

julia/ (j=2,r=0)$ gmake
--2012-05-12 15:50:37-- http://www.netlib.org/lapack/lapack-3.4.1.tgz
Resolving www.netlib.org (www.netlib.org)... 160.36.131.121
Connecting to www.netlib.org (www.netlib.org)|160.36.131.121|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 6147915 (5.9M) [application/x-gzip]
Saving to: `lapack-3.4.1.tgz'

100%[===============================================================================================>] 6,147,915 378K/s in 16s

2012-05-12 15:50:54 (367 KB/s) - `lapack-3.4.1.tgz' saved [6147915/6147915]

md5sum: not found
test: =: unexpected operator
Cannot download lapack-3.4.1.tgz or the MD5 check sum is wrong (Please use orignal).
gmake[3]: *** [lapack-3.4.1] Error 1
gmake[2]: *** [openblas-v0.1.1/libopenblas.so] Error 2
gmake[1]: *** [julia-release] Error 2
gmake: *** [release] Error 2

[Keno]

Seems like OpenBlas is trying to verify the md5 checksum of the downloaded package. Is there anyway to compute the MD5 checksum of a file on FreeBSD?

[forkandwait]

http://lists.freebsd.org/pipermail/freebsd-questions/2004-March/038150.html

Seems like "md5" is what we want. Never used it, so I don't know how it works exactly.

[forkandwait]

grepped and changed md5sum to md5 in the makefile, restarted make, got the following:

julia/ (j=2,r=0)$ gmake
Cannot download lapack-3.4.1.tgz or the MD5 check sum is wrong (Please use orignal).
gmake[3]: *** [lapack-3.4.1] Error 1
gmake[2]: *** [openblas-v0.1.1/libopenblas.so] Error 2
gmake[1]: *** [julia-release] Error 2
gmake: *** [release] Error 2

[Keno]

What does manually downloading http://www.netlib.org/lapack/lapack-3.4.1.tgz and using md5 lapack-3.4.1.tgz give you? If it gives you 44c3869c38c8335c2b9c2a8bb276eb55, try make -C deps distclean-openblas. If not, we'll have to find another way/ (maybe disable the MD5 check temporarily).

[forkandwait]

Yes, same md5. Ran your comand with "gmake", compiling now. will report.

Fails at some point with a missing "md5sum", probably because a Makefile gets downloaded after I go through and change it to only use "md5". I just symlinked md5sum to md5, and we will see if that makes it work.

[Keno]

This has been fixed upstream. Can you download https://github.com/xianyi/OpenBLAS/zipball/develop and extract it into the deps/openblas-v0.1.1 directory to verify that it works as expected?