GMP functions that take doubles choke on non-finite values, avoid that. src/ChangeLog addition: 2015-05-10 Aidan Kehoe <kehoea@parhasard.net> * floatfns.c (double_to_integer): Rename this from float_to_int to fit our newer, bignum-compatible terminology. GMP can signal SIGFPE when asked to turn NaN or infinity into a bignum, and we're not prepared to handle that signal if the OS float library routines don't do that, so check for those values explicitly. * floatfns.c (ceiling_two_float): * floatfns.c (ceiling_one_float): * floatfns.c (floor_two_float): * floatfns.c (floor_one_float): * floatfns.c (round_two_float): * floatfns.c (round_one_float): * floatfns.c (truncate_two_float): * floatfns.c (truncate_one_float): Call double_to_integer() with its new name. * number.c: Don't use the {bignum,ratio,bigfloat}_set_double functions directly here, with GMP they can choke when handed non-finite C doubles, call Ftruncate() and the new float_to_bigfloat() from floatfns.c. Maybe we should extend number-gmp.c with GMP-specific implementations that check for non-finite values. tests/ChangeLog addition: 2015-05-10 Aidan Kehoe <kehoea@parhasard.net> * automated/lisp-tests.el: Backslash a few parentheses in the first column for the sake of fontification. * automated/lisp-tests.el: Check that the rounding functions signal Lisp errors correctly when handed positive and negative infinity and NaN.