Assert that we're using IEEE floats and two's complement integers.

README.md

@@ -182,9 +182,10 @@ The length of the array is guaranteed to be equal to `row_count`.
 ## Data encoding
 
 Integer data (`"integer"`) are encoded as a DEFLATE-compressed array of 32-bit signed integers in the specified `byte_order`.
+It is expected that integers are using a two's complement representation.
 Missing values are represented as -2147483648.
 
-Double-precision data (`"double"`) are encoded as a DEFLATE-compressed array of 64-bit IEEE double-precision floats in the specified `byte_order`.
+Double-precision data (`"double"`) are encoded as a DEFLATE-compressed array of 64-bit IEEE754 double-precision floats in the specified `byte_order`.
 This may contain IEEE special values like NaN and infinity.
 Missing values are encoded as NaN with a payload of 1954, inherited from R
 (see discussion [here](https://stackoverflow.com/questions/70471859/difference-between-na-real-and-nan/70472081#70472081)).

src/get_byte_order.cpp

@@ -1,11 +1,22 @@
 #include "Rcpp.h"
 
 #include <cstdint>
+#include <limits>
 
 //[[Rcpp::export(rng=false)]]
 Rcpp::CharacterVector get_byte_order() {
     uint32_t val = 1;
     auto ptr = reinterpret_cast<unsigned char*>(&val);
+
+    // must use IEEE754 doubles for the binary representation to be correct, see:
+    // https://stackoverflow.com/questions/19351843/why-is-ieee-754-floating-point-not-exchangable-between-platforms
+    static_assert(std::numeric_limits<double>::is_iec559);
+
+    // check that we're using 2's complement, see:
+    // https://stackoverflow.com/questions/64842669/how-to-test-if-a-target-has-twos-complement-integers-with-the-c-preprocessor
+    // https://stackoverflow.com/questions/12276957/are-there-any-non-twos-complement-implementations-of-c
+    static_assert((static_cast<int32_t>(-1) & static_cast<int32_t>(3)) == static_cast<int32_t>(3));
+
     if (ptr[0] == 1) {
         return Rcpp::CharacterVector::create("little_endian");
     } else {