Inference crash due to BoundsError in meta_elim_pass!

[timholy]

Using the master branch of ImagesFiltering, julia master gives this:

julia> using ImagesFiltering, Colors

julia> img = fill(Gray(0), 5, 7); img[3,4] = 1
1

julia> kernel = centered([0.1 0.2; 0.4 0.5])
OffsetArrays.OffsetArray{Float64,2,Array{Float64,2}} with indices 0:1×0:1:
 0.1  0.2
 0.4  0.5

julia> imfilter(img, kernel)
WARNING: An error occurred during inference. Type inference is now partially disabled.
BoundsError(a=Array{Bool, 1}[], i=(0,))
rec_backtrace at /home/tim/src/julia-0.5/src/stackwalk.c:84
record_backtrace at /home/tim/src/julia-0.5/src/task.c:238
jl_throw at /home/tim/src/julia-0.5/src/task.c:560
jl_bounds_error_ints at /home/tim/src/julia-0.5/src/builtins.c:183
meta_elim_pass! at ./inference.jl:3382
finish at ./inference.jl:1926
typeinf_frame at ./inference.jl:1834
typeinf_loop at ./inference.jl:1641
typeinf_edge at ./inference.jl:1576
unknown function (ip: 0x7f40f64f722a)
jl_call_method_internal at /home/tim/src/julia-0.5/src/julia_internal.h:185 [inlined]
jl_apply_generic at /home/tim/src/julia-0.5/src/gf.c:1931
typeinf_edge at ./inference.jl:1582
unknown function (ip: 0x7f40f64f59fa)
jl_call_method_internal at /home/tim/src/julia-0.5/src/julia_internal.h:185 [inlined]
jl_apply_generic at /home/tim/src/julia-0.5/src/gf.c:1931
abstract_call_gf_by_type at ./inference.jl:866
unknown function (ip: 0x7f40f64f53d6)
...
abstract_eval at ./inference.jl:1125
unknown function (ip: 0x7f40f64efef6)
jl_call_method_internal at /home/tim/src/julia-0.5/src/julia_internal.h:185 [inlined]5×7 Array{ColorTypes.Gray{Float64},2}:
 Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)
 Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.5)  Gray{Float64}(0.4)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)
 Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.2)  Gray{Float64}(0.1)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)
 Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)
 Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)  Gray{Float64}(0.0)

julia> 

(with many similar lines omitted in the ...).

This does not happen on release-0.5.

[yuyichao]

Simpler repro (i.e. test)

julia> Base.@propagate_inbounds function g(a)
           @inbounds b = a[1]
           return b
       end
g (generic function with 1 method)

julia> g([1])

Fix coming.

[tkelman]

wrong milestone given

This does not happen on release-0.5.

[StefanKarpinski]

Then don't just delete the milestone – apply the correct one.

[yuyichao]

So are we going to apply 0.6.0 milestone to every issue?

[StefanKarpinski]

Every issue that we expect/hope to get done in the next release, yes. If we aren't sure it will be in the next release but expect it in 1.0, then apply the 1.0 milestone. If we'd like it some day, but it seems unlikely to happen by 1.0, then apply the 2.0 milestone. If it's unclear if we want it at all or if it's a real issue, then leave it with no milestone.

[tkelman]

if it's not release blocking then we shouldn't put it on a release milestone.