Failing on travis

[mlubin]

The current release/master has been failing on travis in a number of mysterious ways:

https://travis-ci.org/JuliaOpt/JuMP.jl/jobs/64000258#L873

while loading /Users/travis/build/JuliaOpt/JuMP.jl/test/model.jl, in expression starting on line 179
     - With solver SCS.SCSSolver
dyld: lazy symbol binding failed: Symbol not found: _validateLinSys
  Referenced from: /Users/travis/.julia/v0.4/Homebrew/deps/usr/lib/libscsdir64.dylib
  Expected in: flat namespace
dyld: Symbol not found: _validateLinSys
  Referenced from: /Users/travis/.julia/v0.4/Homebrew/deps/usr/lib/libscsdir64.dylib
  Expected in: flat namespace

https://travis-ci.org/JuliaOpt/SCS.jl/jobs/55384836#L268

ERROR: LoadError: LoadError: OverflowError()
 in hcat at abstractarray.jl:572
 in feasible_exponential_conic at /home/travis/.julia/v0.4/SCS/test/direct.jl:73

And on #25:
https://travis-ci.org/JuliaOpt/SCS.jl/jobs/62869361#L346

----------------------------------------------------------------------------
    SCS v1.0.7 - Splitting Conic Solver
    (c) Brendan O'Donoghue, Stanford University, 2012
----------------------------------------------------------------------------
Lin-sys: sparse-direct, nnz in A = 8545
eps = 1.00e-04, alpha = 1.80, max_iters = 20000, normalize = 1, scale = 5.00
Variables n = 173, constraints m = 520
Cones:  primal zero / dual free vars: 100
    linear vars: 150
    soc vars: 165, soc blks: 12
    sd vars: 75, sd blks: 3
    exp vars: 15, dual exp vars: 15
Setup time: 6.40e-03s
----------------------------------------------------------------------------
 Iter | pri res | dua res | rel gap | pri obj | dua obj | kap/tau | time (s)
----------------------------------------------------------------------------
     0|      inf       inf      -nan      -inf       inf       inf  6.79e-04 
   100|      inf       inf      -nan       inf       inf       inf  2.43e-02 
   160|      inf       inf      -nan       inf       inf       inf  3.79e-02 
----------------------------------------------------------------------------
Status: Infeasible
Timing: Total solve time: 3.79e-02s
    Lin-sys: nnz in L factor: 24110, avg solve time: 1.21e-04s
    Cones: avg projection time: 1.03e-04s
----------------------------------------------------------------------------
Certificate of primal infeasibility:
|A'y|_2 * |b|_2 = 4.3364e-05
dist(y, K*) = 0
b'y = -1.0000
============================================================================
ERROR: LoadError: LoadError: AssertionError: sol.ret_val == 1

[IainNZ]

Seems kinda busted on pkg.julialang.org too

[joehuchette]

I've been hitting this for a while now on OSX. Since this seems like an upstream issue, any thoughts @bodono? Should we open a ticket upstream?

[joehuchette]

(Specifically the _validateLinSys issue).

[bodono]

I can't replicate the _validateLinSys issue, is it correctly linking to the library? SCS.jl was updated recently, does that fix it? The OverflowError() is a known julia bug: JuliaLang/julia#11320

[joehuchette]

I'm wondering if it's a problem with Homebrew(.jl?), actually. Has the bottle been updated since the latest changes to the SDP format?

[mlubin]

Can't we just work around the OverflowError by reading in the data from a file?

[bodono]

@joehuchette The bottle hasn't been updated as far as I know, it should be updated for the new API.

@mlubin This PR: #28 makes the example smaller, so the OverflowError is no longer a problem, but it's failing too with a really strange error that's either my lack of understanding of julia, or perhaps another julia bug.

The nightly julia linux test is failing, but the release one is passing. The one that's failing has

ERROR: LoadError: UndefVarError: nidx not defined
 in orderconesforscs at /home/travis/.julia/v0.4/SCS/src/SCSSolverInterface.jl:253

but that line is just

252             nidx = length(idxs)
253             A = [A; -sparse(1:nidx, idxs, ones(nidx), nidx, num_vars)]

Any ideas?

[madeleineudell]

@bodono what's the right version of SCS to bottle? do you have a release or
tag for it? we'd like to get SCS working with Convex.jl in time for the
workshop we're running at JuliaCon this Saturday.

On Wed, Jun 24, 2015 at 2:47 AM, bodono notifications@github.com wrote:

@joehuchette https://github.com/joehuchette The bottle hasn't been
updated as far as I know, it should be updated for the new API.

@mlubin https://github.com/mlubin This PR: #28
#28 makes the example smaller,
so the OverflowError is no longer a problem, but it's failing too with a
really strange error that's either my lack of understanding of julia, or
perhaps another julia bug.

The nightly julia linux test is failing, but the release one is passing.
The one that's failing has

ERROR: LoadError: UndefVarError: nidx not defined
in orderconesforscs at /home/travis/.julia/v0.4/SCS/src/SCSSolverInterface.jl:253

but that line is just

252 nidx = length(idxs)
253 A = [A; -sparse(1:nidx, idxs, ones(nidx), nidx, num_vars)]

Any ideas?

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

Madeleine Udell
www.stanford.edu/~udell

[bodono]

I just released v1.1.5, that can be bottled up.

[bodono]

Once it's bottled we have to update SCS.jl to download the right version.

[madeleineudell]

@tkelman could you build SCS v1.1.5 for windows?

[tkelman]

@madeleineudell I sure can. #29 - let appveyor take a look at it before merging though

[madeleineudell]

closed by #29