Hi Paul, On 10/11/21 5:40 PM, Paul Eggert wrote:
On 10/11/21 4:12 AM, Alejandro Colomar wrote:time_t portable_timegm(struct tm *tm) { tm->tm_isdst = 0; /* * If another thread modifies the timezone during the * execution of the line below, it will produce undefined * behavior. */ return mktime(tm) - timezone; }This doesn't work for multiple reasons:
it's not thread-safe,
Actually, since timegm(3) is implemented in terms of mktime(3), as far as I could read from glibc code, the problem will be the same, I think. I don't understand why it wasn't the other way around; maybe it was more complex internally... But timegm(3) shouldn't need to depend on environment variables.
mktime might set timezone even in a single-threaded app,
Yes, I should have called tzset() before the return line.
and the subtraction might overflow.
Yup, casting to int64_t needed. BTW, I had a look at mktime source code, and it uses long, which might be 32 bits, and then there's a lot of checking for overflow. Wouldn't it be simpler to just implement mktime(3) with int64_t internally? Then, only at the return, cast it implicitly to whatever time_t means, but int64_t would simplify the code very much, I think.
Thanks, Alex -- Alejandro Colomar Linux man-pages comaintainer; https://www.kernel.org/doc/man-pages/ http://www.alejandro-colomar.es/
