The Hand of FuManChu

The correct answer is: "nobody knows". But here are some ideas I've been kicking around the ol' cranium lately...

[09:32] *** now talking in #cherrypy
[10:22] <Lawouach> where to start
[10:22] <Lawouach> what's your basic idea toward 3.0?
[10:22] <@fumanchu> oh, I have so many ;)
[10:22] <Lawouach> lol
[10:22] <Lawouach> say big general ones :)
[10:22] <Lawouach> not details per se
[10:23] <@fumanchu> 1) make CP have a kick-butt,
    non-CP-specific toolkit (lib/httptools), that is SO
    good that Quixote, Django, et al can't *help* but
    decide to use it instead of their own server processes
[10:24] <@fumanchu> even if they don't like the way CP
    maps handlers to URL's, for example
[10:24] <@fumanchu> they should be able to build a
    server with the behavior they like out of lib/httptools
[10:25] <Lawouach> we want to be lib that rule them all :)
[10:25] <@fumanchu> yup
[10:26] <Lawouach> i agree as long as we don't become a
    framework on our own, but i already know it's not what
    you intend :)
[10:26] <@fumanchu> right
[10:26] <@fumanchu> it's an anti-framework approach
[10:26] <@fumanchu> we make writing-a-web-framework
    into a weekend's work
[10:27] <@fumanchu> take some from column A; try all of column B
[10:27] <Lawouach> do you want to stay very low-level
    (aka HTTP wrapper level) or make it a bit higher level
    and provide functions such as the bast_match() we were
    talking about last week?
[10:27] <@fumanchu> best_match would be fine as long
    as it doesn't depend upon cherrypy
[10:28] <Lawouach> right, this was a bad example
[10:28] <Lawouach> but basically where httptools should stop?
[10:28] <@fumanchu> I think that can be open-ended
[10:28] <Lawouach> i think we should keep the level
    you've been doing till now
[10:29] <@fumanchu> 2) then, by pulling a ton of code
    out of _cphttptools (putting it in lib/httptools instead),
    I want to see if we can get the Request and Response
    objects down to a tiny size
[10:34] <@fumanchu> the trunk version of _cphttptools
    is already 60% of its 2.1 size
[10:35] <Lawouach> right. hmmm
[10:37] <@fumanchu> and a *lot* of what's left is very OO
[10:38] <@fumanchu> so, one idea I'm toying with: allow
    developers to use their own subclasses of Request
    and Response
[10:40] <@fumanchu> if we make it super-easy to use custom
    Request subclasses, then they will want to start
    overriding Request.run
[10:40] <@fumanchu> take out the filter logic, and
    Request.run becomes:

def _run(self, requestLine, headers, rfile):

    self.headers = list(headers)
    self.headerMap = httptools.HeaderMap()
    self.simpleCookie = Cookie.SimpleCookie()
    self.rfile = rfile
    self.processRequestLine(requestLine)

    try:
        self.processHeaders()
        self.processBody()
        self.main()
        cherrypy.response.finalize()
    except cherrypy.RequestHandled:
        pass
    except (cherrypy.HTTPRedirect, cherrypy.HTTPError), inst:
        inst.set_response()
        cherrypy.response.finalize()

[10:40] <Lawouach> regarding the subclassing of request
    and response, i'm know that it could interest very
    much the guys behind itools
[10:40] <@fumanchu> yes
[10:40] <@fumanchu> and Ben Bangert (routes)
[10:41] <@fumanchu> anyway, if Request.run is *that* simple,
    then who needs filters?
[10:41] <@fumanchu> just code them procedurally into your
    Request.run method
[10:43] <@fumanchu> looking over the filters that are built in...
[10:44] <@fumanchu> I think that half could be done just as
    easily as lib/httptools functions
[10:44] <@fumanchu> and half could be "always on"
[10:44] <@fumanchu> (if we continue to improve them, like
    encodingfilter, to meet the HTP spec)
[10:44] <@fumanchu> HTTP
[10:44] <Lawouach> that's my white cheap :) (i don't think this
    expression exists so i make it up!)
[10:45] <Lawouach> i really want CP to be HTTP conditionnaly compliant
    at least :)
[10:45] <Lawouach> and maybe in CP 4.0 to be unconditionnaly compliant!
[10:45] <Lawouach> :p
[10:45] <@fumanchu> I completely agree
[10:46] <@fumanchu> anyway, I want to stress that I'm still playing
    with these ideas
[10:46] <@fumanchu> nothing's set in stone
[10:47] <Lawouach> since you've be proposing them a while back,
    i've been a great fan of them
[10:47] <@fumanchu> and trying to implement them will turn up
    lots of problems, I'm sure
[10:47] <@fumanchu> oh, well thanks
[10:47] <Lawouach> that's why i don't have so many different
    things to bring for cp 3.0
[10:51] <@fumanchu> one of the nice things about these ideas
    for 3.0 is that the bulk of the work can be done within
    the 2.x branch