Use int instead of char for flag variables

The char type can be equivalent to either signed char or unsigned char.
If the latter, assigning -1 to endian in GMPy_MPZ_Method_From_Bytes
actually stores the value 255, a positive value.  This leads to
incorrect conversions, manifesting as failure of test_mpz_from_bytes.

Using char doesn't save any space, since local variables are stored in
32- or 64-bit CPU registers, or in 32-bit aligned stack slots.  Also,
using char instead of int generates less efficient code.  Modern 64-bit
processors are optimized for 64- and 32-bit quantities.  Operating on
8-bit quantities takes more clock cycles.  Finally, negative values
cannot be stored safely in char variables due to implicit conversion to
positive values of platforms where char == unsigned char.

src/gmpy2_convert.c

@@ -178,7 +178,8 @@ GMPy_RemoveIgnoredASCII(PyObject *s)
 static int
 mpz_set_PyStr(mpz_t z, PyObject *s, int base)
 {
-    char *cp, negative = 0;
+    char *cp;
+    int negative = 0;
     PyObject *ascii_str;
 
     ascii_str = GMPy_RemoveIgnoredASCII(s);

src/gmpy2_mpz_misc.c

@@ -1869,8 +1869,8 @@ GMPy_MPZ_Method_To_Bytes(PyObject *self, PyObject *const *args,
     Py_ssize_t i, nkws = 0, size, gap, length = 1;
     PyObject *bytes, *arg;
     mpz_t tmp, *px = &MPZ(self);
-    char *buffer, sign, is_signed = 0, is_negative, is_big;
-    int argidx[2] = {-1, -1};
+    char *buffer;
+    int sign, is_signed = 0, is_negative, is_big, argidx[2] = {-1, -1};
     const char *byteorder = NULL, *kwname;
 
     if (nargs > 2) {
@@ -2023,8 +2023,8 @@ GMPy_MPZ_Method_From_Bytes(PyTypeObject *type, PyObject *const *args, Py_ssize_t
 {
     Py_ssize_t i, nkws = 0, length;
     PyObject *arg, *bytes;
-    char is_signed = 0, endian, *buffer;
-    int argidx[2] = {-1, -1};
+    char *buffer;
+    int is_signed = 0, endian, argidx[2] = {-1, -1};
     const char *byteorder = NULL, *kwname;
     mpz_t tmp;
     MPZ_Object *result;