1 files changed, 38 insertions, 0 deletions
diff --git a/libxcb/doc/xkb_issues b/libxcb/doc/xkb_issues
new file mode 100644
index 000000000..f10c015e5
--- /dev/null
+++ b/libxcb/doc/xkb_issues
@@ -0,0 +1,38 @@
+
+There are a number of problematic special cases in XKB. The issues 
+mentioned here are at most partly resolved.
+ 
+1. The are several XxxDoodad structures defined in xkb.xml. They are used
+   in a few lists, but in a rather special way: 
+   The struct "CommonDoodad" is supposed to be a rather generic data type, 
+   combining the most basic Doodad fields that are common in all these structures. 
+   All Doodads are encapsulated in a union type simply called "Doodad". 
+   Now this union is used in subsequent list definitions, aiming at a kind of 
+   'polymorphism': From inspection of the protocol and Xlib, the Doodads are to 
+   be discriminated based on their type field.
+   However the special meaning of the type field is not encoded in the protocol. 
+   Furthermore the TextDoodad and the LogoDoodad are variable size types due to
+   some fields of type CountedString16, thereby turning the union into a 
+   possibly variable size type as well. 
+   However, for lists with variable size elements, special sizeof functions are 
+   required. These cannot be autogenerated as it cannot be referred which 
+   Doodad type to use for the union. 
+   Therefore, the Doodad type structures are unsupported at the moment. 
+
+2. There are still some bugs in xkb.xml: Either certain fields are missing
+   that are required by the protocol, or Xlib simply has another understanding
+   of the protocol. 
+
+3. The interface for accessors should be reviewed. 
+
+4. Currently some bitcases carry 'name' attributes. These could be avoided if 
+   the data within would consist of a singe struct field only. 
+
+5. switch could get a 'fixed_size' attribute, so when rewriting valueparam to switch, 
+   an uint32_t * pointer could be used instead of void *. 
+
+6. The automatic inclusion of padding requires some complicated coding in the 
+   generator. This is errorprone and could be avoided if all padding is explicitly
+   given in the protocol definition. For variable size fields that require padding, 
+   the pad tag could get a 'fieldref' attribute. That way padding could be handled
+   a lot easier in the autogenerator. 
+\ No newline at end of file