On Fri, Dec 25, 2009 at 11:29:26PM +0100, jploog@... wrote:
> Jack Whitham has just announced an experimental server allowing browser
> based play of NetHack's tiles version.
Hey, I suggested something like this a while ago, and was told it's a stupid
1. This implementation has totally unplayable latency, mostly because it
builds up an entire TCP connection every time you press a key.
Possible solutions include:
* Keep-Alives/http pipelining. Too bad, the last time I looked at this
(~2005), none of the browsers I tested supported this for XMLHttpRequests.
Perhaps modern browsers of today can do it? Because Crawl playerbase tends
to have reasonable technical skills, we're in a rare position where
telling people "your browser sucks, upgrade it!" is ok.
* Unholy hacks with a persistent push connection (multipart/x-mixed-replace,
falling back to a slowly loading hidden iframe) and DNS requests. The
client can look up keypress.567.was.up-arrow.crawl.akrasiac.org and thus
deliver it to the server in a single half-trip. This does work even on
basics of sanity...
2. It's a hack on the non-tiles version rather than the tiles one. Apart
from looking butt-ugly (as in, worse than ZX Spectrum-era graphics), it
doesn't offer any of the features tiles provide.
We could do it so much better...
1KB // Microsoft corollary to Hanlon's razor:
// Never attribute to stupidity what can be
// adequately explained by malice.