Use -1 % Csize_t instead of reinterpret(Csize_t, -1)

This is because reinterpret(Csize_t, -1) may give an error if -1 literal
is type Int32 instead of Int64 (as is true on 32-bit platforms), as
reinterpret requires the sizes of scalar bittypes to be the same.

Author: ScottPJones

src/iconv.jl

@@ -79,7 +79,7 @@ function iconv!(cd::Ptr{Void}, inbuf::Vector{UInt8}, outbuf::Vector{UInt8},
                 (Ptr{Void}, Ptr{Ptr{UInt8}}, Ref{Csize_t}, Ptr{Ptr{UInt8}}, Ref{Csize_t}),
                 cd, inbufptr, inbytesleft, outbufptr, outbytesleft)
 
-    if ret == reinterpret(Csize_t, -1)
+    if ret == -1 % Csize_t
         err = errno()
 
         # Should never happen unless a very small buffer is used
@@ -111,7 +111,7 @@ function iconv_reset!(s::Union{StringEncoder, StringDecoder})
                 (Ptr{Void}, Ptr{Ptr{UInt8}}, Ref{Csize_t}, Ptr{Ptr{UInt8}}, Ref{Csize_t}),
                 s.cd, C_NULL, C_NULL, s.outbufptr, s.outbytesleft)
 
-    if ret == reinterpret(Csize_t, -1)
+    if ret == -1 % Csize_t
         err = errno()
         if err == EINVAL
             error("iconv error: incomplete byte sequence at end of input")