Julia 1.7 crashes

[ufechner7]

I have a program that crashes with Julia 1.7, but not with Julia 1.6. If you checkout this commit:
ufechner7/KiteViewer@c1d6c5e

and run ./kiteviewer.sh and then click on the button "PLAY" the program will crash on Julia 1.7, but not on 1.6.5.

This code fails:

 log = load_log(7, logfile)

This works:

  log = load_log(7, logfile)
  dummy = log.syslog[1]

My assumption: If I create a variable of type struct in Julia 1.7 the members get garbage collected unless accessed at least once in the main program.

For more details see: https://discourse.julialang.org/t/random-crashes-with-julia-1-7/73821/6

[KristofferC]

Where is load_log defined?

[ufechner7]

Line 477 in this file: https://github.com/ufechner7/KiteUtils.jl/blob/main/src/KiteUtils.jl

"""
    function load_log(P, filename::String)
Read a log file that was saved as .arrow file.  P is the number of tether
particles.
"""
function load_log(P, filename::String)
    if isnothing(findlast(isequal('.'), filename))
        fullname = joinpath(DATA_PATH[1], filename) * ".arrow"
    else
        fullname = joinpath(DATA_PATH[1], filename) 
    end
    table = Arrow.Table(fullname)
    syslog = StructArray{SysState{P}}((table.time, table.orient, table.elevation, table.azimuth, table.l_tether, table.v_reelout, table.force, table.depower, table.v_app, table.X, table.Y, table.Z))
    return SysLog{P}(basename(fullname[1:end-6]), syslog, syslog2extlog(P, syslog))
end

[garrison]

If you are determined to track this down, I would suggest using git-bisect from a known bad 1.7 commit, with 599d329 as the "good" commit (i.e., the last commit before 1.6 was branched off).

BTW, when I attempt to reproduce, I get the following error:

Importing file "data/log_8700W_8ms.csv.xz" ...
LoadError: type QuatRotation has no field w
Stacktrace:
 [1] getproperty(x::QuatRotation{Float64}, f::Symbol)
   @ Base ./Base.jl:42
 [2] df2syslog(P::Int64, df::DataFrames.DataFrame)
   @ Main.Importer ~/src/KiteViewer/src/Importer.jl:54
 [3] import_log()
   @ Main.Importer ~/src/KiteViewer/src/Importer.jl:77
 [4] top-level scope
   @ ~/src/KiteViewer/src/Importer.jl:85
 [5] include
   @ ./client.jl:451 [inlined]
 [6] macro expansion
   @ ~/src/KiteViewer/src/KiteViewer.jl:304 [inlined]
 [7] (::var"#21#34")()
   @ Main ./task.jl:423
in expression starting at /home/garrison/src/KiteViewer/src/Importer.jl:8

[ufechner7]

@garrison Thanks a lot for testing!

I created a new branch with the name "bug" for reproducing this issue. Test case (well, on Linux):

mkdir tmp
cd tmp
git clone https://github.com/ufechner7/KiteViewer.git
cd KiteViewer
git checkout bug
./kiteviewer_pure.sh 
   - after 1 min GUI appears
   - after one more minute the log file is imported
   - click on the PLAY button
     => program sometimes crashes with segfault

If you uncomment line 325 of src/KiteViewerPure.jl it doesn't crash on Julia 1.7.
It never crashes on Julia 1.6.

Problem: It does not always crash, only sometimes...

I intend to bisect the commit that caused this bug, but probably not before the weekend...

[vchuravy]

If you are on Linux you can also try to use rr to create a capture of the issue e.g. julia --bug-report=rr

[ufechner7]

I found a way to let it ALWAYS crash when using Julia 1.7.1

mkdir tmp
git clone -b bug https://github.com/ufechner7/KiteViewer.git
cd KiteViewer
./create_sys_image.sh
    - wait 8 minutes
./kiteviewer_pure.sh 
   - after 12s min GUI appears
   - after one more minute the log file is imported
   - click on the PLAY button
     => program crashes with segfault

Stacktrace: https://github.com/ufechner7/KiteViewer/blob/bug/doc/crash_report.txt

But that it always crashes happens only when using a custom system image, created with the script create_sys_image.sh .

[ViralBShah]

Please reopen if still occuring with 1.9.x