On Sat, Feb 08, 2003 at 09:33:43PM +0600, Ivan Pascal wrote: > Hi, > > A short introduction: > Any key in XKB can have keysyms, xkb actions, a virtual modifier(s) and some >flags describing a key behavior such as 'repeat', 'locking' and so on. > A core protocol knows about keysyms only. If one changes a keysym binding >using the core protocol it can happen that the keysyms only will be moved but >not actions or behavior flags. > To solve this compatibility problem the XKB has a special 'compatibility map' >which sets a relation between keysyms and other attributes a key can have. >Thus when some keysym place in the keyboard map is changed the XKB searches >this keysym in the compatibility map and if the keysym is found there it also >moves an action and flags to the key. The same procedure the XKB does after >any keyboard map loading (even it made by an XKB protocol). Therefore although >an action of key can be specified in the symbols map directly it is a common >way to tie the action to some special keysym in the compatibility map >(xkb/compat/* files) and specify that keysym in the symbols map. > > Also each key has 'protection flags' which can protect the key from such >changes. These flags are >- 'forbid all changes' (include an action, a virtual modifier and other flags) >- 'forbid a repeat flag change' >- 'forbid a lock flag change' >- 'forbid a modifier change' > > All written above you can read in XKBproto or XKBlib documentation. But you >can't find there when and how those protection flags are set. > > The xkbcomp program does it using rules: >- if any action for some key is specified directly in the symbols file the >'forbid all changes' flag should be set for this key >- if a repeat flag is specified there the 'forbid a repeat flag change' bit >should be set. >- and similar rules for a locking flag and a virtual modifier. > > Now you can guess what happened with 'mouse actions'. Since new special >actions were specified directly in the symbols map the xkbcomp set the >protection flags for the corresponded keys and the 'mouse actions' which are >tied to keysyms in the compatibility map now can't be bound to those keys. > > A disappearing of autorepeat isn't so clear but caused by the same reason. >There are some details in an XKB design (I consider them as bugs) which lead >to it. All keys initialy have not an autorepeat mark unless it explicitly >specified in the symbols map (you know there are not such explicit >spcifications in existent maps). The XKB sets these marks (bits) after a new >map loading but does it for each key separately after attempt to apply >a compatibility map ( of course if the key has not any keysyms mentioned in >that map the autorepeat mark will be set unconditionaly). > Thus if the key has the protection flag the XKB interrapts compatibility >map applaying for this key and 'forgets' to set autorepeat but to the key. > > Thus you see the autorepeat losing is rather a bug and can be fixed (we need >to make all keys marked as repeating initialy or to place the procedure which >sets those marks before the protection bits checking). But disappearing >of the 'mouse actions' is a feature of the XKB and can't be avoided without >some redesign of XKB. > > Therefore a solution for the second problem is to 'make all actions equal >in their rights'. It means >- specify the mouse actions (and the repeat flag) in the same place where the > 'special actions' are (e.g. in the symbols map) >- or move the special action to the compatibility map. > In the second case we need to introduce some new keysyms (something like >xf86_Next_VMode, xf86_Prev_Mode, xf86_Ungrab, etc.). But we even don't need >to add them to the keysymdef.h file but can use the lib/X11/XKeysymDB file >instead. > > I have tested both solutions and it works in both cases. And I can make >a patch quickly. But we have to chose one. > As for me I prefer the second one becouse it is the way used for all other >xkb actions.
I think that's a good reason to go with the second method. I think that might also make it more likely that these mappings get used than is the case now. The Function keys have special action mappings too. I'm not sure if they were ever supposed to auto-repeat, but they don't now for the same reasons. Thanks for the analysis of the problem. David -- David Dawes Release Engineer/Architect The XFree86 Project www.XFree86.org/~dawes _______________________________________________ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
