Migrate to Julia 0.5-dev

[jeff-regier]

That is, add "nightly" to .travis.yml without causing tests to fail.

[jeff-regier]

@kpamnany How did you get Celeste running on Julia 0.5? I get a segmentation fault when I try running benchmark_elbo_with_hessian.jl.

Calculating ELBO.

signal (11): Segmentation fault
while loading /home/jeff/git/Celeste.jl/bin/benchmark_elbo_with_hessian.jl, in expression starting on line 155
subtract_kl_value_wrapper at /home/jeff/.julia/v0.5/Celeste/src/ElboKL.jl:81
_calc_hessian at /home/jeff/.julia/v0.5/ForwardDiff/src/api/hessian.jl:98
unknown function (ip: 0x7f47661d158b)
[inline] at /home/jeff/Downloads/julia/src/julia_internal.h:69
jl_call_method_internal at /home/jeff/Downloads/julia/src/gf.c:1915
#hessian#62 at /home/jeff/.julia/v0.5/ForwardDiff/src/api/hessian.jl:69
unknown function (ip: 0x7f476620029e)
[inline] at /home/jeff/Downloads/julia/src/julia_internal.h:69
jl_call_method_internal at /home/jeff/Downloads/julia/src/gf.c:1915
unknown function (ip: 0x7f476620407c)
unknown function (ip: 0x7f47662040ae)
[inline] at /home/jeff/Downloads/julia/src/julia_internal.h:69
jl_call_method_internal at /home/jeff/Downloads/julia/src/gf.c:1915
subtract_kl! at /home/jeff/.julia/v0.5/Celeste/src/ElboKL.jl:106
unknown function (ip: 0x7f476620c011)
unknown function (ip: 0x7f476620c24e)
[inline] at /home/jeff/Downloads/julia/src/julia_internal.h:69
jl_call_method_internal at /home/jeff/Downloads/julia/src/gf.c:1915
elbo at /home/jeff/.julia/v0.5/Celeste/src/ElboDeriv.jl:1119
unknown function (ip: 0x7f4766283d8f)
unknown function (ip: 0x7f476628400e)
[inline] at /home/jeff/Downloads/julia/src/julia_internal.h:69
jl_call_method_internal at /home/jeff/Downloads/julia/src/gf.c:1915
do_call at /home/jeff/Downloads/julia/src/interpreter.c:66
eval at /home/jeff/Downloads/julia/src/interpreter.c:185
eval at /home/jeff/Downloads/julia/src/interpreter.c:189
eval_body at /home/jeff/Downloads/julia/src/interpreter.c:552
jl_interpret_toplevel_thunk_with at /home/jeff/Downloads/julia/src/interpreter.c:582
jl_toplevel_eval_flex at /home/jeff/Downloads/julia/src/toplevel.c:552
[inline] at /home/jeff/Downloads/julia/src/toplevel.c:421
jl_toplevel_eval_flex at /home/jeff/Downloads/julia/src/toplevel.c:487
jl_parse_eval_all at /home/jeff/Downloads/julia/src/ast.c:784
jl_load at /home/jeff/Downloads/julia/src/toplevel.c:577
include at ./boot.jl:264
jl_apply_generic at /home/jeff/Downloads/julia/src/gf.c:1918
include_from_node1 at ./loading.jl:417
unknown function (ip: 0x7f4970519f0c)
jl_apply_generic at /home/jeff/Downloads/julia/src/gf.c:1918
process_options at ./client.jl:248
_start at ./client.jl:300
unknown function (ip: 0x7f49705150d9)
jl_apply_generic at /home/jeff/Downloads/julia/src/gf.c:1918
unknown function (ip: 0x401b13)
unknown function (ip: 0x4016df)
__libc_start_main at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x401725)
Allocations: 77101224 (Pool: 77043587; Big: 57637); GC: 773
Segmentation fault
``

[kpamnany]

Not sure why.

$ ./julia -E versioninfo\(\)
Julia Version 0.5.0-dev+2313
Commit e22c20c* (2016-01-26 17:47 UTC)
Platform Info:
  System: Linux (x86_64-redhat-linux)
  CPU: Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1
nothing
$ echo $JULIA_NUM_THREADS
18
$ ./julia ~/.julia/v0.5/Celeste/bin/benchmark_elbo_with_hessian.jl 
WARNING: New definition 
    write(GZip.GZipStream, Array{#T<:Any, N<:Any}) at /home/kpamnany/.julia/v0.5/GZip/src/GZip.jl:456
is ambiguous with: 
    write(Base.IO, Array{UInt8, N<:Any}) at io.jl:154.
To fix, define 
    write(GZip.GZipStream, Array{UInt8, N<:Any})
before the new definition.
WARNING: Method definition cfitsio_typecode(Type{Int64}) in module Libcfitsio at /home/kpamnany/.julia/v0.5/FITSIO/src/libcfitsio.jl:162 overwritten at /home/kpamnany/.julia/v0.5/FITSIO/src/libcfitsio.jl:162.
Loading data.
Intializing ModelParams.
Getting VP from sources.
Processing the bands.
  Initializing band 1 patches.  Initializing band 1 tiled image sources.
  Initializing band 2 patches.  Initializing band 2 tiled image sources.
  Initializing band 3 patches.  Initializing band 3 tiled image sources.
  Initializing band 4 patches.  Initializing band 4 tiled image sources.
  Initializing band 5 patches.  Initializing band 5 tiled image sources.

Calculating ELBO.
  9.713185 seconds (19.57 M allocations: 10.120 GB, 11.15% gc time)
  4.873643 seconds (14.80 M allocations: 9.191 GB, 23.20% gc time)

[kpamnany]

You can try to set Threaded=false in ElboDeriv.jl, or export JULIA_NUM_THREADS=1 before running to rule out threads as the problem. Beyond that... hmm.

[jeff-regier]

It still fails with Threaded=false :-( The errors look like they have
something to do with ForwardDiff.jl, but the unit tests for ForwardDiff.jl
all pass.

$ julia5 -E versioninfo\(\)
Julia Version 0.5.0-dev+2399
Commit 1f7c72e (2016-01-29 12:28 UTC)
Platform Info:
  System: Linux (x86_64-unknown-linux-gnu)
  CPU: Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Core2)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1
nothing

On Thu, Feb 11, 2016 at 4:00 PM kpamnany notifications@github.com wrote:

You can try to set Threaded=false in ElboDeriv.jl, or export
JULIA_NUM_THREADS=1 before running to rule out threads as the problem.
Beyond that... hmm.

—
Reply to this email directly or view it on GitHub
#118 (comment)
.

[kpamnany]

Any idea why it says line 155 below? It's a 29 line file...

signal (11): Segmentation fault
while loading /home/jeff/git/Celeste.jl/bin/benchmark_elbo_with_hessian.jl, in expression starting on line 155

[jeff-regier]

No, that's kind of weird. It's a 29 line file on my machine too.

$ wc -l benchmark_elbo_with_hessian.jl 
29 benchmark_elbo_with_hessian.jl

I've seen this kind of thing before though, where Julia reports the line number from a different file. It may be unrelated to the seg fault, well, except in that it makes it the seg fault harder to locate.

[kpamnany]

Can try to get a backtrace through gdb?

[jeff-regier]

I just checked out the version of Julia you're using (Julia Version 0.5.0-dev+2313; commit e22c20c). That "fixes" the problem. The problem reappears when I switch back to the newer version of Julia. I suggest we don't upgrade our versions of Julia 0.5 for awhile....

[kpamnany]

lol... sounds good to me.