Re: os.time() and the last second of 1969

[Marc Balmer <marc@...>]

Am 12.02.2022 um 14:04 schrieb Marc Balmer <marc@msys.ch>:

mktime() is a function to calculate the number of seconds *since* the epoch. Note the *since*.

Since mktime() returns -1 to signal an error and the last second of 1969 (which is -1 seconds from the epoch) returns the same value that is otherwise used to signal an error, the program misinterprets the result.

In consequence, when mktime() returns -1, you probably should take a look at errno (in C code), as mktime() sets errno accordingly.

This is a bug in Lua, or, rather in its use of mktime().

To properly catch this cornercase, clear errno prior to the call of mktime() and check it afterwards:

--- loslib.c.orig 2022-02-12 13:57:14.000000000 +0100

+++ loslib.c 2022-02-12 14:02:57.000000000 +0100

@@ -358,10 +358,11 @@

     ts.tm_min = getfield(L, "min", 0, 0);

     ts.tm_sec = getfield(L, "sec", 0, 0);

     ts.tm_isdst = getboolfield(L, "isdst");

+    errno = 0;

     t = mktime(&ts);

     setallfields(L, &ts);  /* update fields with normalized values */

   }

-  if (t != (time_t)(l_timet)t || t == (time_t)(-1))

+  if ((t != (time_t)(l_timet)t || t == (time_t)(-1)) && errno != 0)

     return luaL_error(L,

                   "time result cannot be represented in this installation");

   l_pushtime(L, t);

(Clearing errno before mktime() is important, since it may still contain a non-zero value from a previous error).

At least on macOS, mktime() seems not to set set errno on illegal input values.  This looks like highly implementation specific and maybe nothing should be done about it :(