Le samedi 12 mars 2016 à 12:30 -0800, Douglas Bates a écrit :
> I have been adding more tests to the MixedModels package and
> encountered a peculiar situation where, AFAICS, a test should produce
> one type of failure and it doesn't. It fails at a later point but I
> can't see how it gets there.
>
> To reproduce, check out a copy of the MixedModels master branch.
>
> Pkg.add("MixedModels")
> Pkg.checkout("MixedModels")
> MixedModels.inject!(speye(4), sparse(ones(4, 4)))
>
> The unexported function inject!(A, B) zeros A, then copies the
> contents B into A. For sparse matrices, the nonzero positions of B
> must be a subset of those in A. What I expect to happen in this case
> is that the code will enter the block in lines 52 - 54 of https://git
> hub.com/dmbates/MixedModels.jl/blob/master/src/inject.jl. In fact,
> if you uncomment line 51 in that file the output from the @show macro
> shows that indeed
>
> kk > length(dnzrv)
>
> the second time through the "for k in snzr" loop. But when I run
> that example, the error thrown is a BoundsError at line 55. What I
> don't understand is how it gets to line 55 the second time through
> the loop.
>
> I have been looking at this for a long time trying to see if I am
> doing something stupid but I can't. Does the BoundsError get thrown
> for others? My versioninfo is
I was puzzled for a few minutes too, but the explanation is actually
quite simple: you need to call throw() on the error, i.e.
throw(ArgumentError("cannot inject sparse s into sparse d"))
This is indeed an easy to make and hard to spot mistake. Maybe a code
checking tool like Lint.jl would/could have detected that the
ArgumentError wasn't used at all.
Regards
> julia> versioninfo()
> Julia Version 0.4.2
> Commit bb73f34 (2015-12-06 21:47 UTC)
> Platform Info:
> System: Linux (x86_64-linux-gnu)
> CPU: Intel(R) Core(TM) i5-5200U CPU @ 2.20GHz
> WORD_SIZE: 64
> BLAS: libopenblas (NO_LAPACKE DYNAMIC_ARCH NO_AFFINITY Prescott)
> LAPACK: liblapack.so.3
> LIBM: libopenlibm
> LLVM: libLLVM-3.3
>
> I haven't upgraded to 0.4.3 because the Ubuntu package is not yet
> available but I can test on 0.4.3 if that would help.
