Re: A sleep function for NSE

[David Fifield <david () bamsoftware com>]

On Wed, Feb 18, 2009 at 10:07:15PM +0000, jah wrote:
> On 18/02/2009 20:24, David Fifield wrote:
> > Does someone else have a use for a sleep function? Is the design
> > suitable for you?
>
> sleep() looks good.  I wonder whether t should take milliseconds as
> other operations do - socket.set_timeout for example.  I guess it
> largely depends on how long scripts generally want to sleep a coroutine
> for - just a thought.

Thanks for bringing that up. In high-level languages I prefer times to
be always in seconds, but consistency with other parts of the library is
more important than that. I found nmap.clock_ms and nmap.set_timeout
that take times in milliseconds. msrpctypes.lua converts times to
seconds even when the protocol uses some other representation. grepping
for "seconds" didn't find anything else relevant.

http://nmap.org/nsedoc/modules/nmap.html#clock_ms
http://nmap.org/nsedoc/modules/nmap.html#set_timeout
http://nmap.org/nsedoc/modules/msrpctypes.html#marshall_NTTIME

A function called "sleep" normally takes an argument in seconds, even if
it only provides integer precision. So if the function changes to take
milliseconds its name has to change too. In the name of consistency, I
think there are only two choices: sleep taking a floating-point number
of seconds, or sleep_ms taking an integer number of milliseconds.

My preference would be to have sleep take floating-point seconds, and to
change nmap.clock_ms and nmap.set_timeout to deal in seconds also. The
settimeout in Python does it in seconds:

http://docs.python.org/library/socket.html#socket.socket.settimeout

Furthermore, clock_ms should be renamed "time" because it is more like
the time function in C or os.time in Lua (seconds since 1970) than it is
the C clock function (time since some arbitrary point in arbitrary
units). clock_ms is doubly frustrating because it uses gettimeofday,
which provides microsecond precision, then purposely degrades the result
to return milliseconds.

It may be too late for nmap.set_timeout because that would be an
incompatible change. nmap.clock_ms is easier because it is currently
hardly used and its name should change anyway. Any thoughts?

David Fifield

_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://SecLists.Org