What I understand from the discussion is that Mark Wai does not want to change the case-ness of words. He simply wants the method lookup to ignore case. This is not a bad idea. OTOH, what I really like about Smalltalk is the way it approaches natural language (in this case English). Capitalization is done for readability and only (IMHO) because spaces are not allowed in selectors. I.e. you would like to write 'this is how natural language looks like' but end up with 'thisIsHowNaturalLanguageLooksLike'. So, the rule I use is: 'Write it out in natural language and then replace each space and it following character with the capitalized version of that character'. This leaves no ambiguity. To repeat some examples:
unsigned integer 32 bits -> UnsignedInteger32Bits -> UInt32 as lower case -> asLowerCase as array of keywords -> asArrayOfKeywords
The abbreviation of UInt32 is a bit dirty but can be deduced using the rule (I never abbreviate selectors). In the case there are parameters you would write in natural language:
as byte array encoding x into y -> asByteArrayEncoding: x into: y
I.e. the rule is: do not capitalize after a parameter.
The first rule cannot be detected by the compiler (unless you supply a dictonairy or have it look at other selectors), the second can supported by the compiler. Furthermore, I like to mention to Mark Wai that Squeak does help a bit when you save a method and it does not recognize a selector (and lets you choose another).
Last but not least: I am not amused by using the underscore symbol ( _ ) for assignment. Try converting a piece of Visual Smalltalk code with symbol like: #UP_RIGHT, #DOWN_LEFT. You can't file this is in without editing!
Hope this helps.