C++ is hard to parse. One reason for this is because it's difficult to tell what is supposed to be a variable and what is supposed to be the name of a user defined type. This means that information from later stages of parsing has to be fed back into earlier stages, making the whole thing a big tangled mess.
It's not just compilers that have problems either - the problematic constructs can also be difficult for humans to understand (especially if taken out of context).
I think C++ would be much easier to parse (for both humans and machines) if a very simple rule were added - if a name starts with a capital letter, it is the name of a type, and if it starts with a lower-case letter it is the name of a variable. Sort of a 1-bit hungarian notation (except that it has meaning to the compiler).
This is already a de facto standard in many programs (often by virtue of being a consequence of an even stricter scheme). Though of course if it were added to C++ now it would break many many programs...
A nice side effect of this is that the syntax for types could be made much more regular, whilst still retaining the "definition mirrors use" paradigm. Given that we found a captial letter at the start of an identifier, we know we are parsing a type name. So if we come across an open parenthesis ("(") we can parse that as "a function returning the type to the left of the parenthesis. So for example we can name the type "array of N pointers to functions returning pointers to functions returning pointers to characters" just by reading it backwards: Char*()*()*[N]. This is much easier than the normal C++ declaration "char*(*(*a[N])())()" which is very hard to understand and which also buries the name of the variable you are declaring in the middle of the type name rather than putting it on the right where it belongs. SPECS solves the same problem but at the cost of a less familiar syntax and losing "definition mirrors use".
[...] letter and variables and values don't. I think this is a useful convention, and some languages (including the language I am writing) enforce [...]