reproducible segfault inside cfunction(...)

[luchr]

I'm trying to understand some ideas (which I was shown in JuliaMath/Cubature.jl#23) concerning elimination of dynamic dispatch. With the following code (which may be weird and silly; sorry I'm still learning) I always get a segfault simply by calling cfunction (even before trying to use the output pointer):

versioninfo()
type callinfos{FUNC_FT<:Function}
  f :: FUNC_FT
end

evalf_callback{FUNC_FT}(ci::callinfos{FUNC_FT}) = ci.f(0.5)::Float64

evalf_callback_c{FUNC_FT}(ci::callinfos{FUNC_FT}) = cfunction(
    evalf_callback, Float64, (callinfos{FUNC_FT},))

evalf_callback_c( callinfos(sin) )

Here is the output:

Julia Version 0.5.1-pre+31
Commit 6a1e339 (2016-11-17 17:50 UTC)
Platform Info:
  System: Linux (x86_64-redhat-linux)
  CPU: Intel(R) Core(TM) i7-4700MQ CPU @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, haswell)

signal (11): Segmentation fault
while loading <Path>/Desktop/mytest2.jl, in expression starting on line 12
unknown function (ip: 0x7f792313aa9f)
jl_function_ptr at <Path>/julia/julia_v0.5/src/codegen.cpp:982
evalf_callback_c at <Path>/Desktop/mytest2.jl:9
unknown function (ip: 0x7f76fc3df985)
jl_call_method_internal at <Path>/julia/julia_v0.5/src/julia_internal.h:189 [inlined]
jl_apply_generic at <Path>/julia/julia_v0.5/src/gf.c:1945
do_call at <Path>/julia/julia_v0.5/src/interpreter.c:66
eval at <Path>/julia/julia_v0.5/src/interpreter.c:190
jl_toplevel_eval_flex at <Path>/julia/julia_v0.5/src/toplevel.c:558
jl_parse_eval_all at <Path>/julia/julia_v0.5/src/ast.c:717
jl_load at <Path>/julia/julia_v0.5/src/toplevel.c:596
jl_load_ at <Path>/julia/julia_v0.5/src/toplevel.c:605
include_from_node1 at ./loading.jl:488
unknown function (ip: 0x7f791741c56b)
jl_call_method_internal at <Path>/julia/julia_v0.5/src/julia_internal.h:189 [inlined]
jl_apply_generic at <Path>/julia/julia_v0.5/src/gf.c:1945
process_options at ./client.jl:262
_start at ./client.jl:318
unknown function (ip: 0x7f7917441c78)
jl_call_method_internal at <Path>/julia/julia_v0.5/src/julia_internal.h:189 [inlined]
jl_apply_generic at <Path>/julia/julia_v0.5/src/gf.c:1945
jl_apply at <Path>/julia/julia_v0.5/ui/../src/julia.h:1392 [inlined]
true_main at <Path>/julia/julia_v0.5/ui/repl.c:112
main at <Path>/julia/julia_v0.5/ui/repl.c:232
__libc_start_main at /lib64/libc.so.6 (unknown line)
unknown function (ip: 0x4012c8)
Allocations: 2358732 (Pool: 2357723; Big: 1009); GC: 1
Segmentation fault (core dumped) 

[vtjnash]

Looks like something isn't restoring the builder context correctly

[KristofferC]

Now:

julia> type callinfos{FUNC_FT<:Function}
         f :: FUNC_FT
       end

julia> evalf_callback{FUNC_FT}(ci::callinfos{FUNC_FT}) = ci.f(0.5)::Float64
evalf_callback (generic function with 1 method)

julia> evalf_callback_c{FUNC_FT}(ci::callinfos{FUNC_FT}) = cfunction(
           evalf_callback, Float64, (callinfos{FUNC_FT},))
evalf_callback_c (generic function with 1 method)

julia> evalf_callback_c( callinfos(sin) )
ERROR: ccall: the type of argument 1 doesn't correspond to a C type
Stacktrace:
 [1] evalf_callback_c(::callinfos{Base.#sin}) at ./REPL[6]:1

[luchr]

"Now" means Julia v0.6 and v0.7 (= trunk). So for this versions the above message is the right error message to show. So fixed since v0.6.