fix for julia v1.5; jl_function_ptr was removed

[JeffBezanson]

This seems to work well enough for now as a workaround; ImageView.jl works at least. Would be good to try some more Gtk apps to confirm whether it works. I wanted to to just un-break 1.5, and we can work on a redesign later.

[vtjnash]

Suggested change

	ret = closure.callback.f(p, eventp, closure.closure)
	ret = closure.callback(p, eventp, closure.closure)

[vtjnash]

This would be preferable to avoid increasing yet more the dependence on internal design details.

[JeffBezanson]

I agree. I don't think we have a call overload for CFunction though.

[vtjnash]

No, but let's just not allocate it then

[JeffBezanson]

Oh, wait a minute, is this only called from julia? Why was it using cfunction in the first place? @tknopp any reason?

[timholy]

With ProfileView I get a long inference stacktrace from @JeffBezanson's version, and with @vtjnash's I get

julia> FATAL ERROR: Gtk state corrupted by error thrown in a callback:
ERROR: MethodError: objects of type Base.CFunction are not callable
Stacktrace:
 [1] notify_motion(::Ptr{GObject}, ::Ptr{Gtk.GdkEventMotion}, ::Gtk.Gtk_signal_motion{GtkReactive.MouseHandler{GtkReactive.UserUnit}}) at /home/tim/.julia/dev/Gtk/src/events.jl:46
 [2] (::Gtk.var"#237#238")() at /home/tim/.julia/dev/Gtk/src/events.jl:2
 [3] g_sigatom(::Any) at /home/tim/.julia/dev/Gtk/src/GLib/signals.jl:176
 [4] gtk_main() at /home/tim/.julia/dev/Gtk/src/events.jl:1

[JeffBezanson]

Looking into it.

[JeffBezanson]

Works on 1.5 --- the inference issue is due to a recent change on master.

[tknopp]

Thanks a lot for having a look into this. Especially nice is that we do not need to change the public interface. Would be interesting how "bad" you would consider this workaround. As far a I can see you are using a generated function to move the call to @function into an earlier stage of the compiler. I know that generated functions should not be used too much, but in my opinion we currently only have a hand full calls to the native signal_connect function.

[JeffBezanson]

I don't think it's that bad. Certainly better than calling an internal C function, so for purposes of keeping the current API it's fine.

[timholy]

Do you want this merged as-is or do you want #512 (comment) and/or other changes?

[giordano]

Note that arm64 is failing on Drone because tests that were broken aren't anymore:

SetCoordinates: Error During Test at /drone/src/test/gui.jl:548
 Unexpected Pass
 Expression: mtrx.xx == 300
 Got correct result, please change to @test if no longer broken.

SetCoordinates: Error During Test at /drone/src/test/gui.jl:549
 Unexpected Pass
 Expression: mtrx.yy == 280
 Got correct result, please change to @test if no longer broken.

armv7l is failing at init time instead:

Stacktrace:
 [1] pipeline_error at ./process.jl:525 [inlined]
 [2] read(::Cmd) at ./process.jl:412
 [3] (::Gtk.var"#301#307"{String})() at /drone/src/src/Gtk.jl:94
 [4] withenv(::Gtk.var"#301#307"{String}, ::Pair{String,String}) at ./env.jl:161
 [5] (::Gtk.var"#300#306")(::String) at /drone/src/src/Gtk.jl:93
 [6] (::gdk_pixbuf_jll.var"#8#9"{Gtk.var"#300#306"})() at /root/.julia/packages/gdk_pixbuf_jll/kzXJa/src/wrappers/arm-linux-gnueabihf.jl:50
 [7] withenv(::gdk_pixbuf_jll.var"#8#9"{Gtk.var"#300#306"}, ::Pair{String,String}, ::Vararg{Pair{String,String},N} where N) at ./env.jl:161
 [8] #gdk_pixbuf_query_loaders#7(::Bool, ::Bool, ::typeof(gdk_pixbuf_jll.gdk_pixbuf_query_loaders), ::Gtk.var"#300#306") at /root/.julia/packages/gdk_pixbuf_jll/kzXJa/src/wrappers/arm-linux-gnueabihf.jl:49
 [9] gdk_pixbuf_query_loaders at /root/.julia/packages/gdk_pixbuf_jll/kzXJa/src/wrappers/arm-linux-gnueabihf.jl:33 [inlined]
 [10] __init__() at /drone/src/src/Gtk.jl:92
 [11] _include_from_serialized(::String, ::Array{Any,1}) at ./loading.jl:692
 [12] _require_from_serialized(::String) at ./loading.jl:743
 [13] _require(::Base.PkgId) at ./loading.jl:1034
 [14] require(::Base.PkgId) at ./loading.jl:922
 [15] require(::Module, ::Symbol) at ./loading.jl:917
 [16] include at ./boot.jl:328 [inlined]
 [17] include_relative(::Module, ::String) at ./loading.jl:1105
 [18] include(::Module, ::String) at ./Base.jl:31
 [19] include(::String) at ./client.jl:424
 [20] top-level scope at none:6
during initialization of module Gtk
in expression starting at /drone/src/test/runtests.jl:2

[vtjnash]

Yeah, looks good. Eventually, this package should probably consider ceasing to use the cfunction_ hack entirely, but it's not immediately necessary to edit the other code.

[timholy]

@tknopp, with apologies to you I presumed to release 1.1.4, so it doesn't hold back progress on the Julia 1.5 release.

[tknopp]

That is just fine. Wanted to have a look at this right now but am happy that things are already done.