|more useful options|
The one piece of ambiguity on this topic in the RFC is that the meta variables such as REQUEST_METHOD are described in the document as being made available to the system in a "system-defined manner". For webservers I've worked with, they're available as environment variables.
In the scary times of Windows 3.x, there was some piece of software called WinHTTPD by Robert B. Denny. It had a different implementation of CGI to work around the limitations of Windows 3.x, and to allow writing CGIs in almost all available languages. This implementation was creatively called Windows CGI. Google found a copy of the spec at http://wwwusers.di.uniroma1.it/~reti/Reti2_html/docum/CGI/WINCGI.HTM, and a newer one at http://techsupt.winbatch.com/webcgi/webbatch.exe?techsupt/nftechsupt.web+WIL~Extenders/Winsock+WebSite~Windows~CGI~1.3a~Interface.txt. According to http://www.sc.ehu.es/scrwwwsr/isapi/ch03.htm, O'Reillys HTTP server named WebSite also supported that protocol.
The main difference here is that headers are not passed via the environment, but via an INI file created for each request. Also input and output are not the stdio file handles, but instead files created by the webserver. All of this makes access very easy, all your CGI ("backend" in the spec) code needs is file I/O and optionally access to the GetPrivateProfileString() function to read the INI file. This allowed to use almost any programming language, you could even use DOS programs to implement a "Windows CGI"-conforming application.
The obvious downside is that WIndows CGI creates a lot of temp files, but then again, Windows 3.x is not your high-performance server platform anyway.
Someone even has written an interface program that runs as a classic CGI, creates the required files for the Windows CGI spec, and launches ancient an Windows CGI program.
Today I will gladly share my knowledge and experience, for there are no sweeter words than "I told you so". ;-)
In reply to Re^2: How to check the request is (GET OR POST) in CGI