Better stacktraces for ccall libraries that do not exist

[vchuravy]

After:

julia> f() = ccall((:time, "doesnotexit"), Int, ())
f (generic function with 1 method)

julia> h() = f()
h (generic function with 1 method)

julia> h()
ERROR: ccall: could not load library
 in library doesnotexit
Stacktrace:
 [1] f at ./REPL[1]:1 [inlined]
 [2] h() at ./REPL[2]:1
 [3] top-level scope at REPL[3]:1

Before:

julia> f() = ccall((:time, "doesnotexit"), Int, ())
f (generic function with 1 method)

julia> h() = f()
h (generic function with 1 method)

julia> h()
ERROR: error compiling h: could not load library "doesnotexit"
doesnotexit.so: cannot open shared object file: No such file or directory
Stacktrace:
 [1] top-level scope at REPL[3]:1
caused by [exception 1]
could not load library "doesnotexit"
doesnotexit.so: cannot open shared object file: No such file or directory
Stacktrace:
 [1] top-level scope at REPL[3]:1

julia> g() = h()
g (generic function with 1 method)

julia> g()
ERROR: error compiling g: could not load library "doesnotexit"
doesnotexit.so: cannot open shared object file: No such file or directory
Stacktrace:
 [1] top-level scope at REPL[5]:1
caused by [exception 1]
could not load library "doesnotexit"
doesnotexit.so: cannot open shared object file: No such file or directory
Stacktrace:
 [1] top-level scope at REPL[5]:1

Currently we lose the ability to report why the library couldn't be opened.

cc: @maleadt @simonbyrne

[vtjnash]

SGTM, can you fix the tests and we can get this into the release?

[vchuravy]

Any idea on how to get a better error?

[vtjnash]

Aren't dynamic semantics great? You don't have to codegen the errors—just fallback on the existing runtime.