Jason Dagit writes:

On Thu, Aug 25, 2011 at 2:39 PM, Greg Fitzgerald wrote:

...

"cabal update" hangs.  "ping haskell.org" times out.  But haskell.org and hackage webpages are loading just fine.  What's going on?

Lots of servers turn off ICMP packet responses these days

Because users don't really need error messages, that's privileged information for system administrators.

so ping isn't as reliable as it once was for detecting when a server is running.

Looks like haskell.org isn't one of them. CON:~ % ping haskell.org PING haskell.org (78.46.100.180) 56(84) bytes of data. 64 bytes from lambda.haskell.org (78.46.100.180): icmp_req=1 ttl=48 time=53.8 ms 64 bytes from lambda.haskell.org (78.46.100.180): icmp_req=2 ttl=48 time=52.4 ms 64 bytes from lambda.haskell.org (78.46.100.180): icmp_req=3 ttl=48 time=52.5 ms ^C --- haskell.org ping statistics --- 3 packets transmitted, 3 received, 0% packet loss, time 2000ms rtt min/avg/max/mdev = 52.451/52.935/53.804/0.670 ms

I'm not sure why cabal update would hang.

It takes a while to complete with no visual feedback. Perhaps the network is just slow? Or perhaps the web pages are cached somewhere along the way? -k -- If I haven't seen further, it is by standing in the footprints of giants

Jason Dagit writes:

...

...

Lots of servers turn off ICMP packet responses these days

...

Because users don't really need error messages, that's privileged information for system administrators.

Besides, if someone is trying to debug http protocol issues using ICMP, they're taking an awfully indirect route.

Yes, first thing would be to check the log files. Oh, wait, users don't have read permissions for those...

I think the reason it gets disabled is typically well intentioned.

And we all know where that road leads - to frustration and low quality services, that's where. Sorry for my grumpiness - this is a constant struggle for me. :-)

...

It takes a while to complete with no visual feedback.  Perhaps the network is just slow?  Or perhaps the web pages are cached somewhere along the way?

All of those are plausible [..]

Perhaps 'cabal update' should provide some visual feedback by default? There's -v, but there's still a noticable pause (15 seconds on a fairly beefy computer and a good connection) between "Downloaded to [..]" and "Reading available packages...". The index is 3MB, it's probably not big enough that a smarter protocol (rsync/rdiff?) would improve things? Except perhaps checking that it has been updated since the last update - that'd probably save a ton of bandwidth if/when people do automatic updates. -k -- If I haven't seen further, it is by standing in the footprints of giants