Ftoa shift math

[stefano-zanotti-88]

These calculations:

nanoprintf/nanoprintf.h

Lines 657 to 670 in f0687bc

	// This if-else statement can be completely optimized at compile time.
	if (NPF_DOUBLE_BIN_BITS > NPF_FTOA_MAN_BITS) {
	man_f = (npf_ftoa_man_t)(bin_f >> ((unsigned)(NPF_DOUBLE_BIN_BITS -
	NPF_FTOA_MAN_BITS) %
	NPF_DOUBLE_BIN_BITS));
	carry = (uint_fast8_t)((bin_f >> ((unsigned)(NPF_DOUBLE_BIN_BITS -
	NPF_FTOA_MAN_BITS - 1) %
	NPF_DOUBLE_BIN_BITS)) & 0x1);
	} else {
	man_f = (npf_ftoa_man_t)((npf_ftoa_man_t)bin_f
	<< ((unsigned)(NPF_FTOA_MAN_BITS -
	NPF_DOUBLE_BIN_BITS) % NPF_FTOA_MAN_BITS));
	carry = 0;
	}

Why are those there, instead of the following ones?

if (NPF_DOUBLE_BIN_BITS > NPF_FTOA_MAN_BITS) {
  man_f = (npf_ftoa_man_t)(bin_f >> (NPF_DOUBLE_BIN_BITS - NPF_FTOA_MAN_BITS));
  carry = (uint_fast8_t)((bin_f >> (NPF_DOUBLE_BIN_BITS - NPF_FTOA_MAN_BITS - 1)) & 0x1);
} else {
  man_f = (npf_ftoa_man_t)((npf_ftoa_man_t)bin_f << (NPF_FTOA_MAN_BITS - NPF_DOUBLE_BIN_BITS));
  carry = 0;
}

The current ones seem always ok, as long as both sizes are powers of 2, but also needlessly cryptic.
The simplified ones are perfectly fine, as far as I can tell.

Edit: I see that my code causes a "right shift by negative amount" in the first carry assignment, iff NPF_DOUBLE_BIN_BITS == NPF_FTOA_MAN_BITS. This is not a problem in itself, since that branch is not taken in this case; but maybe this can trigger UB and screw the whole function.
The man_f assignments seem always ok, though.