X Core Protocol Scheme

Michael Titke michael.tiedtke at o2online.de
Sat Dec 12 09:31:55 PST 2015

On 12/12/2015 14:23, Ilya Anfimov wrote:
> On Sat, Dec 12, 2015 at 10:17:56AM +0100, Michael Titke wrote:
>   Good day!
>   First, quoting looks like you had posted some letters before the
> one I'm replying, but I didn't see it in my mailbox or x.org list
> archives,  so I will answer just to text in the letter written at
> 2015-12-12 10:17:56 +0100.
>>     Now the arc drawing part works: it needed a little change of the
>>     "implemented protocol" or request sequence:
>>     First map the window, second listen for events. Third draw the arc.
>>     That was the third round after authentication and putting a window on
>>     the screen. It would be nice if these sequences were  documented as
>   Generally,  the exact sequence isn't that important. If you want
> a consistent picture, and don't want to redraw 50 times a second,
> then  you  should  draw  every time you received Expose event, no
> matter what you drawed before and even no matter of  your  knowl-
> edge of the mapped state.
> (Hmm,  sorry for my english and for my descriptive abilities: ev-
> ery time doesn't mean  that after receiving 2 or 5 or  10  Expose
> events  in a single packet you should redraw the same 10 times in
> a raw. Just redraw once at least all the requested regions  imme-
> diately after some Expose events received).
>   This is described in Expose event paragraph in section 11 of the
> protocol specification.
> btw, this automatically means  that  you  should  receive  Expose
> events  on  the  window you want to draw and therefore select Ex-
> posure in the event mask at a window creation.

Listening for expose events sounds like a good idea but the relation 
between graphics operation, their effect and expose events doesn't seem 
to be stated in the specifications. In the current experimental setup 
only keyboard events are in the event mask and it's a little bit 
surprising that the graphic operation has an effect only after a 
keyboard event has been received.

>>     part of the protocol. For now it remembers me that I need to change the
>>     low level IO routines to then buffer the graphics operations in a
>>     drawing graph and connect it the repaint events and ...
>>     (X-polifill-arc X (second v) (second g) 150 200 60 60 (* 10 64) (* 350
>>     64)) => 24  ; doesn't appear
>>     (X-map-window X (second v)) => 8
>>     (X-control X) => Window #"254 255 31 4" received a key event #"9" down.
>>     escaping-X-control
>>     VSI> (X-polifill-arc X (second v) (second g) 150 200 60 60 (* 10 64) (*
>>     350 64))  ; OK appears
>>     (X-polifill-arc X (second v) (second g) 150 200 60 60 (* 10 64) (* 350
>>     64)) => 24
>>     VSI> (X-control X)
>>     ...
>>     VSI:
>>     [1]https://code.launchpad.net/~michael-tiedtke-i/viper-system-interface
>>     /alfa
>>     On 11/12/2015 22:02, Michael Titke wrote:
>>     Hello!
>>     As part of a first incursion into the possibility to implementing
>>     native support for X starting from the wire protocol (w/o any Xlib/XCB
>>     support) I ran into a couple of situations where documentation didn't
>>     match implementation.
>>     The first surprise was the "magic" of the MIT Magic Cookie which needs
>>     that little deviation from the protocol encoding where you have to put
>>     the padding bytes at the end. Now I really made it to open a window and
>   You  have specs of MIT Magic Cookie? You lucky guy, I don't have
> this one.

It just needed some script "playing" the server to find out about the 
correct padding which usually should coincide with the terminating C 
string null byte.

Konsole output
;; SCA/X Reverse Engineering Xauth
;; xauth mcookie reverse engineering => leading padding for mcookie key
;; (bind s AF_INET INADDR_ANY 6000)
;; (listen s 5)
;; (begin
;; (newline) (newline)
;; (display "Try to open an X client on display to find out 
about the magic i
n mit-magic-cookie-1 ..."))
;; (define p (accept s))
;; (define sxau (read-bytes 1024 (port->fdes (car p))))

;; (sys-load "VSI-core/byte-structures.scm")
;; (define X-conn-client-prefix-layout ; copied from X.scm
;; (byte-structure-description
;;   (byte-order 1)
;;   (padding    1)
;;   (major-version 2)
;;   (minor-version 2)
;;   (nbytes-auth-proto  2)  ; ??? STRING8 in the standard
;;   (nbytes-auth-string 2)  ; ??? STRING8 in the standard
;;   (padding    2)))
;; (define-bs-accessors 'xconc X-conn-client-prefix-layout)
;; (xconc:byte-order sxau)
;; (xconc:major-version sxau)
;; (xconc:minor-version sxau)
;; (xconc:nbytes-auth-proto sxau)
;; (xconc:nbytes-auth-string sxau)
;; (define proto (byte-string-select sxau
;;                (bs-length X-conn-client-prefix-layout) 18))
;; (c-string->deprecated-string proto)
;; (define audat (byte-string-select sxau
;;                (+ (bs-length X-conn-client-prefix-layout) 18 1) 16))
;; audat
;; sxau

>>     receive key codes destined for it but no keysyms as the request for the
>>     keyboard mappings is silently ignored. The XKB extension as far as I
>>     understand it essentially replaced that? But there is no addendum to
>>     core protocol specifications.
>   No,  XKB  is  not  strictly  required. The next log supposes you
> fixed your issues and successfully receiving keyboard events.

The keyboard events seem to be alright but as stated they only report 
the keycode: to map keys to the labels on the caps the keysyms are 
needed. But core protocol requests to receive those mappings for the 
current maps in effect are ignored by the server.

>>     The next round was about creating a circle: somehow I found out that
>>     another map request on the window was needed to see the respective
>   The  fact  that  the newly created window is unmapped is pointed
> out in the first sentence of CreateWindow description.

Yes. The second map request was a bogus request to receive the error 

>>     errors due to simple mistakes during the preparation of the request and
>>     some misleading protocol encoding which states 3+3n for the request
>>     length.
>   Didn't  see  any  misleadings  here. Request length is specified
> consistently across the entire Requests section, and this meaning
> is the only possible if you understand paragraph "Request Format"
> in Chapter 1.
>   Moreover, it is self-evident when you try to calculate that  re-
> quest length by hand against fields byte length values.
>>     (define X (X-connection)) => #<unspecified>
>>     (define v (X-create-window X 50 50 300 400)) => #<unspecified>
>>     v => (44 #"254 255 255 3")
>>     (X-map-window X (second v)) => 8
>>     (define g (X-create-gc X (second v))) => #<unspecified>
>>     g => (16 #"253 255 255 3")
>>     (X-polifill-arc X (second v) (second g) 150 200 100 100 (* 170 64) (*
>>     180 64)) => 24
>>     (X-map-window X (second v)) => 8
>>     (X-control X) => VSI SCA/X: unhandled error: Length#"0 16 4 0  253 255
>>     255 3  0 0 71 0  0
>>     0 0 0  0 0 0 0  0 0 0 0  0 0 0 0  0 0 0 0"
>>     #<unspecified>
>>     My question is: will this continue like this? Are there any plans to
>   I hope, yes.

Well, thank you.

>>     finally deliver the protocol specifications where these kinds of
>>     interactions are layed out? Or some up to date updates on the core
>>     protocol? But as I have heard the X server doesn't even know about all
>>     registered extensions anymore - at least on Ubuntu with Unity one of
>>     the first events to be received was an impossible operation code of 192
>>     which wasn't reported by xdpyinfo to belong to any registered
>   Extension  opcodes  assigned  by server at QueryExtension reply,
> you should get that bytestream and find the extension  name  from
> there. The number 192 may mean anything.

The QueryExtension request isn't implemented in the experimental setup 
right now but as stated /xdpyinfo/ didn't report that operation code.

>>     extension. The current state of X11 is a bit puzzling: it when works
>   Yes,  the  protocol is not trivial. But it is consistent and core is
> well-documented.
>   There are some features or extensions which lacks  proper  docu-
> mentation  (MIT-MAGIC-COOKIE-1  authentication, RENDER extension,
> NV-GLX extension) but in fact all of them is crap, and you should
> probably get rid of it.

The core protocol is trivial - just not well documented. ;-) The XKB 
extension doesn't make much sense (as an integral whole) and the 
problems described in its specifications only start to show up now for 
some clients where the Render extension seems to provide access to 
modern graphic operations (in a very fine grained way? ...

Thanks for pointing out CLX. I'm definitely going to have a look at it 
but the CL model of frames and views (and the SBCL use of metaclasses) 
doesn't really fit the current development goal to create a controlling 
graphical input output path suited for Scheme. Simply transforming Lisp 
code wouldn't help much as the overhead of writing byte structure 
descriptions is small compared to the task of "updating" the Algorithm 
Notation Scheme to include Graphic (and sound) algorithms in the usual 
sleak way. At least the new memory manager doesn't need to interrupt the 
process every now and then but when we compare the implementation of the 
numerical tower with its support for four or five different number types 
in 6000 lines of C code with the expected amount of code to support half 
the graphic formats in use ... CU
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.x.org/archives/xorg/attachments/20151212/1d499de4/attachment.html>

More information about the xorg mailing list