Quick Internet Connection check?

I am implementing an auto update feature in a program I am writing.  It will
optionally check for a new version when the program is launched.  I would
like this to be as seamless as possible.  I am currently using
WinInetLibrary<<ping: to test for a connection.  I am also considering
wrapping and using the InternetGetConnectedState API function.  Though I
wonder if this is redundant to what ping checks, if so there is no point.
Regardless, the problem I have is that when I disconnect the network cable
from a PC (to imitate a network outage) the ping is quite slow to fail
thereby slowing the application load time.  I suppose that is due to some
sort of timeout.

Does anyone know of a robust way to reliably check for an Internet
connection passively?  I suspect that fork would not work given that it is
an API call.  I would also like to do this the simplest way possible.

Any ideas?

Chris

Hi Chris,

Christopher J. Demers wrote:
Hi Chris,

I also needed to do this, and found no easy solution.

All the reading I did suggested that if you need your application to work
across all versions of windows the easiest way is to roll your own. XP (and
probably 2000) have some new APIs that could be used, but the wininet
functions behave inconsistently across versions and are buggy on certain
versions of 98/95.

If you are targeting home users, rather than wrapping
InternetGetConnectedState I would suggest wrapping InternetAttemptConnect.
If the user connects via a network it will always answer true, but if the
user connects via dial-up on demand, it will block showing a dialog-box
asking the user to make a connection.

That does not guarantee that you have a connection (but neither does
InternetGetConnectedState and friends in the real world). I have read a
number of suggestions that after calling InternetAttemptConnect do an os
ping to a number of root servers. However what I have done is to do a number
of http HEAD requests to major servers. I set a 15 second time-out and if I
get back a response, I consider the user to be connected.

It would be easier if you are deploying only to XP and 2000, but I dont
recall the details of the new APIs.

Hope this helps,
Steve

--
Steve Waring
Email: [hidden email]
Journal: http://www.stevewaring.net/blog/home/index.html

Christopher J. Demers wrote:

> I am implementing an auto update feature in a program I am writing.
> It will optionally check for a new version when the program is
> launched.

I'd suggest a different tack.

I think it's rude for an application to make an outbound connection without the
user's consent (obtained in advance), this is especially true in cases where
the user may have a dial-up connection (which seems likely or there wouldn't be
an issue at all) since establishing a connection may cost the user money.

So, if you have an option to check for updates at user-specified times (which
is not on by default, though there's nothing wrong with a first-use prompt to
ask what the setting should be), then you can assume that the net connection is
available at the appropriate time since the *user* has told you it's OK to make
a connection then.  Hence it's OK to fail slowly if the connection is not
available after-all.

Then too, what do you mean by "available" ?  I can think of several meanings,
and it's not clear which one you should be using to determine whether to do a
check.  Physically the machine could be connected to the net*, in a position to
establish a connection to the net, or be incapable of reaching the net.  A
different axis of variation is that it could be "working off-line" (in which
case the MS wrappers for normal TCP/IP connections will fail, but a direct
connection at the sockets level will work) or fully "on-line".  A last axis of
variation is that it could be in a time or place where creating an outbound
connection will irritate the user, or where it would be perfectly welcome.

There may be security issues too but, of course, that depends on your
application and your users.

    -- chris

[*] and on which network adapter ?  My XP laptop can't do it's
get-the-time-from-the-Net trick and I think that's because it's trying to talk
NTP over the wireless link (which is, of course, not turned on) and somehow
manages not even to try routing the packets over my wire LAN (where, as it
happens, they would be blocked by my firewall, but *it* doesn't know that...)

"Chris Uppal" <[hidden email]> wrote in message
news:3f20efda$0$46013$[hidden email]...
> Christopher J. Demers wrote:
>
> > I am implementing an auto update feature in a program I am writing.
> > It will optionally check for a new version when the program is
> > launched.
>
> I'd suggest a different tack.
>
> I think it's rude for an application to make an outbound connection
without the
> user's consent (obtained in advance), this is especially true in cases
where
> the user may have a dial-up connection (which seems likely or there
wouldn't be
> an issue at all) since establishing a connection may cost the user money.

I should clarify.  I am not trying to passively check for the actual update
(unless they are already connected), I just want to passively check for the
presence of an existing Internet connection.  If they are connected the
program will check, if they are not connected I don't want it to prompt them
to connect or even try to connect.  The program will just skip the check.
All of this is an option available upon setup and within the program,
defaulting to off.  I am just trying to give them the option to check for an
update when the program is run, they can also do it manually.  I suspect
most users will have direct Internet connections, I just don't want the
program start up to be slow if it can't connect to the update site due to a
network problem.  I appreciate your concerns, which is why this is all
optional and defaulted to off.  I just want to make it work well if the user
should decide to turn it on. ;)

Chris

"Steve Waring" <[hidden email]> wrote in message
news:bfq8pl$hj7b8$[hidden email]...
> Hi Chris,
>
> Christopher J. Demers wrote:
> > Does anyone know of a robust way to reliably check for an Internet
> > connection passively?  I suspect that fork would not work given that
> > it is an API call.  I would also like to do this the simplest way
> > possible.
 ...
> That does not guarantee that you have a connection (but neither does
> InternetGetConnectedState and friends in the real world). I have read a
> number of suggestions that after calling InternetAttemptConnect do an os
> ping to a number of root servers. However what I have done is to do a
number
> of http HEAD requests to major servers. I set a 15 second time-out and if
I
> get back a response, I consider the user to be connected.

I think I may do the same.  I imagine that I could write my own socket stuff
to do a quick check, and fork it so it does not slow down the program load
due to the timeout.

Thanks,
Chris