Dict->IdDict in exception types (JuliaPy#927)

* Dict->IdDict in exception types

Dicts use extensive specialization and this adds latency.
IdDicts don't and are the go-to associative container
for type-keys.

* Rely on `const` to solve inference problems

The `::Dict`, being an abstract annotation, wouldn't have been all
that much help anyway.

src/exception.jl

@@ -140,7 +140,7 @@ end
 #########################################################################
 # Mapping of Julia Exception types to Python exceptions
 
-const pyexc = Dict{DataType, PyPtr}()
+const pyexc = IdDict{DataType, PyPtr}()
 mutable struct PyIOError <: Exception end
 
 function pyexc_initialize()
@@ -213,7 +213,7 @@ end
 
 function pyraise(e, bt = nothing)
     eT = typeof(e)
-    pyeT = haskey(pyexc::Dict, eT) ? pyexc[eT] : pyexc[Exception]
+    pyeT = haskey(pyexc, eT) ? pyexc[eT] : pyexc[Exception]
     err = PyJlError(e, bt)
     ccall((@pysym :PyErr_SetObject), Cvoid, (PyPtr, PyPtr),
           pyeT, PyObject(err))

src/io.jl

@@ -11,7 +11,7 @@
 function ioraise(e, bt = nothing)
     if isa(e, MethodError) || isa(e, ErrorException)
         ccall((@pysym :PyErr_SetString), Cvoid, (PyPtr, Cstring),
-              (pyexc::Dict)[PyIOError],
+              pyexc[PyIOError],
               showerror_string(e, bt))
     else
         pyraise(e, bt)