corrupt data for Array{Union{Missing, Float32},1}

[anj00]

I've boiled down to this basic code

using JLD2, FileIO, Printf

data = Union{Missing, Float32}[80.34f0, 42.17f0, 2.27f0, 20.91f0, 11.11f0, 161.08f0, 18.49f0, 1.55f0, 143.47f0, 2.4f0, 16.36f0, 30.09f0, 11.74f0, 63.28f0, 84.75f0, 14.71f0, 1.3752f0, 8.93f0, 37.2f0, 297.57f0, 69.5f0, 15.04f0, 162.98f0, 21.13f0, missing, 19.24f0, 2.82f0, 4.11f0, 35.42f0, 7.93f0, 2.95f0, 196.25f0, 4.06f0, 3.505f0, 6.27f0, 6.81f0, 4.4f0, 30.42f0, 60.94f0, 20.59f0, 18.09f0, 36.34f0, 171.44f0, 12.82f0, 5.38f0, 1.49f0]

save("data.jld2", Dict("data" => data))
data_jld = load("data.jld2")["data"]
@printf("data size  %5d\n", size(data,1))
@printf("jld2 size  %5d\n", size(data_jld,1))
@printf("data valid %5d\n", size(    data[.!ismissing.(data    ), :], 1))
@printf("jld2 valid %5d\n", size(data_jld[.!ismissing.(data_jld), :], 1))

the data has one missing value and that seems to trigger some bug. The output I am getting is the following: the saved data has the same size as original, but bunch of values are a set to missing

data size     46
jld2 size     46
data valid    45
jld2 valid    37

Note, it only happens with Union{Missing, Float32} without Missing or with Float64 it works as expected.

I run Win10, Julia 1.2.0 and relevant packages versions

  [5789e2e9] FileIO v1.0.7        
  [033835bb] JLD2 v0.1.3

Also in my real code (where similar arrays are columns in a DataFrame) it tends to crash after several times I try to save such types of data in a loop (but I couldn't so far reproduce it a simple example like above). I'll attach the crash in the file, in case it is relevant here
jld2_crash.txt

[anj00]

The same issue happens with Int64

using JLD2, FileIO, Printf

data = Union{Missing, Int64}[0,0,missing,missing,0,0,0,0,0,0,missing,missing,missing,0,0,0,0]

save("data.jld2", Dict("data" => data))
data_jld = load("data.jld2")["data"]
@printf("data size  %5d\n", size(data,1))
@printf("jld2 size  %5d\n", size(data_jld,1))
@printf("data valid %5d\n", size(    data[.!ismissing.(data    ), :], 1))
@printf("jld2 valid %5d\n", size(data_jld[.!ismissing.(data_jld), :], 1))

in this case all the data is wiped out

data size     17
jld2 size     17
data valid    12
jld2 valid     0

[anj00]

above code is working correctly with

Win10, Julia 1.4.0 and
FileIO v1.2.4
JLD2 v0.1.13

wonder if the bug as a whole is fixed or this is just a coincidence. (have code which deals with this bug for my specific data, would be great to remove it if the bug is gone)

[JonasIsensee]

I also cannot reproduce the error on linux with
julia 1.4.1
FileIO v1.3.0
JLD2 v0.1.13

[JonasIsensee]

Tests were added on master