Re: io_date

["D Burgess" <db@...>]

What is your platform? and what exception, SIGFPE?
I had a problem with OS400 throwing an exception on
floating point to int assignment. 
e.g. int n = (double)d;
I think your change is wrong because there is no
guarantee that you will have an integer value on
the LHS of (n == -1).

+++++++++++++++++++++++++++++++++
>The other day I discovered that the C code for
>io_date in liolib.c (Lua 5.0 and previous)
>was causing floating point exceptions on my platform.
>For some mysterious reason coercion of (double) -1 
>to a time_t makes my compiler produce code that raises
>an error. The code for io_date starts:
>
>static int io_date (lua_State *L) {
>  const char *s = luaL_optstring(L, 1, "c");
>  time_t t = (time_t)(luaL_optnumber(L, 2, -1));
>  struct tm *stm;
>  if (t == (time_t)(-1))  /* no time given? */
>    t = time(NULL);  /* use current time */
>  .........
>  
>You will note that argument 2 is pulled from the
>Lua stack, immediately coerced to a time_t, then
>compared with a (time_t) -1. This seems to me a bit
>like travelling from Manhattan to Newark via
>Montreal. So I altered the code to
>
>static int io_date (lua_State *L) {
>  const char *s = luaL_optstring(L, 1, "c");
>  time_t t;
>  struct tm *stm;  
>  const lua_Number n = luaL_optnumber(L, 2, -1);
>  if (n == -1)
>    t = time(NULL);
>  else
>    t = (time_t) n;   
>  ........
> 
>which seems to have cured the problem. Anybody see
>any snags I may have overlooked?
>     
>-- 
>Gavin Wraith (gavin@wraith.u-net.com)
>Home page: http://www.wraith.u-net.com/