Comment author: @protz

Are you positive that you're running the exact same executable?

Comment author: daweil

Yes.
The executable is build once with ocamlbuild in cygwin.
And executed first in cygwin and then outside in a Dos shell.

Can you reproduce the issue on one of your machine ?

Comment author: @protz

Yes, I can confirm. This is strange indeed.

Comment author: @protz

The Cygwin shell defines the TZ environment variable while the Windows shell does not. This is what's causing the difference in behavior. To reproduce the error in a Cygwin shell, just run

unset TZ

then launch your test program.

Comment author: @protz

The underlying problem has something to do with DST (Dayling Saving Time). I strongly suspect this is not a bug, in the sense that:

1. the same problem happens on Linux

protzenk@sauternes:/tmp $ TZ=Europe/Paris ./a.out 
more than one hour is 1.000000 second
one hour is 0.000000 second
Fatal error: exception Unix.Unix_error(32, "mktime", "")
protzenk@sauternes:/tmp $ TZ=USA/New_York ./a.out 
more than one hour is 3601.000000 second
one hour is 3600.000000 second
less than one hour is 3599.000000 second

and 2)
i) the documentation says in a fairly explicit manner that gmtime returns a tm struct whose tm_isdst field is always 0 (see http://msdn.microsoft.com/en-us/library/0z9czt0w.aspx)
ii) both the windows and the unix implementation of OCaml's mktime set tm_isdst to -1, thus leaving it up to the environment to determine whether DST is active in the system's TZ or not (http://msdn.microsoft.com/en-us/library/d1y53h2a.aspx).

The conclusion is, the date returned by gmtime(3599.), when interpreted with DST on or off, I have no idea which, represents one second before the epoch, thus a date that's not representable, hence the ERANGE message.

The documentation explicitly says that the tm argument of mktime is interpreted in the local time zone http://caml.inria.fr/pub/docs/manual-ocaml/libref/Unix.html

Comment author: @protz

I'm marking this as no change required, since I strongly suspect nothing's wrong with OCaml.

Comment author: daweil

OK. As the behavior is surprising, could you just add something in the documentation of the gmtime function ?