Bjorn Bringert wrote:
pwrapper is not an HTTP server, though the Haddock comment can make you think so. pwrapper allows you to talk *CGI* over a TCP port, but I have no idea why anyone would like to do that.
Here is a scenerio. I want a basic web application: someone makes a request, and my program computes a response. * For one reason or another, I settle with CGI. * The program is huge and slow to load. (Let's say it statically links in the whole GHC API and therefore is basically GHC itself. :) ) It would suck to re-load this program at every request. * Or, the program performs work that requires more file-system privilege than the admin of the web server grants. You know, a good admin sets up a web server and all CGI scripts to run with nobody's privilege. * Or, nevermind performance or privilege. I am a cheapo, and I use a cheapo hosting provider, which only provides me with 3MB of storage. My program weighs 17MB (recall that it links in the whole GHC :) ). Here is a solution. The program runs as a daemon and never quits; it can run somewhere with sufficient privilege and storage. It talks CGI over TCP. At the web server, which is super-slow, super-paranoid, and super-cheapo, the CGI script is a lightweight C program that redirects everything over TCP to my daemon. (Here is a counter-solution. The program still runs as a daemon somewhere, but it talks my own protocol over TCP. The CGI script is a lightweight C program that parses CGI into my own protocol. Besides having to design my own protocol carefully, here is a problem: C is a great language for writing parsers that are incomplete, inconsistent, and insecure. :) )