interact with R event loops

[simonbyrne]

In order to get features like graphics windows working correctly, we need to be able to interact with the R event loops. Some details are covered in R-ext §8.

From what I understand, we need to implement (some of) the functionality provided by run_Rmainloop.

[randy3k]

Eventloop was implemented in RCalling.jl and Rif.jl. See for example this link. The code was written in C, however, it should be easy to port it to Julia. I can submit a PR this afternoon.

[randy3k]

Please checkout the current master branch for the eventloop. Basically, there are two functions: RCall.rgui_start() and RCall.rgui_stop().

[simonbyrne]

Once we have callbacks sorted out, we could automate this using setHook.

[randy3k]

In the eventloop, there was a dirty fix related to sigint. The problem was when a sigint is signaled from Julia (pressing Control - C), the R_interrupts_pending is set to state 1. Without a proper R eventloop, R_interrupts_pending remains state 1. And it causes segfault when evaluating R_ProcessEvents. So R_interrupts_pending was forced to be 0 in there.

For a while, I only think that it is needed for R_ProcessEvents. Until just now, I discovered this

julia> using RCall

julia> reval("gc()")
RCall.RObject{RCall.RealSxp}
         used (Mb) gc trigger (Mb) max used (Mb)
Ncells 219236 11.8     460000 24.6   350000 18.7
Vcells 322397  2.5     786432  6.0   786432  6.0


julia> unsafe_load(cglobal((:R_interrupts_pending,RCall.libR),Cint))
0

julia> ^C

julia> unsafe_load(cglobal((:R_interrupts_pending,RCall.libR),Cint))
1

julia> reval("gc()")
ERROR: RCall.jl 

 in error at error.jl:22
 in reval_p at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:11
 in reval_p at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:19
 in reval at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:29
 in reval at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:31

julia> unsafe_load(cglobal((:R_interrupts_pending,RCall.libR),Cint))
0

julia> reval("gc()")

signal (11): Segmentation fault: 11
sexp at /Users/Randy/.julia/v0.4/RCall/src/types.jl:332
jlcall_sexp_21406 at  (unknown line)
reval_p at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:18
reval at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:29
jlcall_reval_21501 at  (unknown line)
reval at /Users/Randy/.julia/v0.4/RCall/src/iface.jl:31
jl_apply at /opt/local/julia/src/interpreter.c:55
eval at /opt/local/julia/src/interpreter.c:213
jl_toplevel_eval_flex at /opt/local/julia/src/toplevel.c:527
jl_toplevel_eval_in at /opt/local/julia/src/builtins.c:579
eval_user_input at REPL.jl:62
jlcall_eval_user_input_21271 at  (unknown line)
anonymous at REPL.jl:92
jl_apply at /opt/local/julia/src/task.c:241
Selection: 

But interestingly, if reval("gc()") is replaced by rcall(:gc), it causes no issues.

It means that we might need to check the value of R_interrupts_pending constantly.

[randy3k]

One possible solution is to disable signal handling

    if !Rinited
        unsafe_store!(cglobal((:R_SignalHandlers,libR),Cint),0)
        argv = ["REmbeddedJulia","--silent","--no-save"]
        i = ccall((:Rf_initEmbeddedR,libR),Cint,(Cint,Ptr{Ptr{UInt8}}),length(argv),argv)
        i > 0 || error("initEmbeddedR failed. Try restarting and running Pkg.build(\"RCall\").")
        global const Rproc = Rinstance(i)
    end

[randy3k]

Though the eventloop implementation is still primitive, I believe this thread could be closed (at least for now).

[simonbyrne]

One thought: in the timer loop we could check if there are any open graphics devices, and if not, close the timer?

[randy3k]

If I am correct, the timer loop is also needed for some X11 applications and they are not necessary graphics.

[simonbyrne]

We could create a graphic-specific one (started by a hook), and a general (manual) one?