And templates within templates?

Which is stupid. Theres context there to know what you've been parsing. You just saw a template started, why not realize that and see that its two separate >'s? Most parsers, especially generated ones, are too damn stupid to be of any real use. Just look at the horrible error messages they put out. They just don't have enough context to realize that the actual error is/was.

Hardly. GCC 4 has already replaced their DUMB bison/yacc parsers with hand rolled recursive decent parsers, which are already faster, but should allow error messages to improve (since you can place them at any point, instead of at places a generated parser finally notices an error), and allow the parser to know what its parsing. Its not that hard.

I have an oldish project that uses a lexicalanalyzer setup that allows you to push and pop modes for parsing, like for example, when parsing HTML, you can pop on a CSS analyzer, or javascript, all while the same parser pass is executing. Something like that, just done a little better would fix all the issues.

The only time complicating the parser/scanner in a compiler mattered was when you only had 2K ram Who bloody cares that theres a few extra KB of code in the compiler when it can properly parse, parses faster, and provides MUCH better error reports?

2 isn't a valid symbol name, so you obviously meant
foo< bar< (1024 >> 2) >
and there's no harm in requiring the parenthesis, IMO. Also, you're missing a >.

That implies that someone decided ">>" should be accepted as a substitute for a single ">" in that context. Surely that's going to be even more confusing than just saying "Unexpected token '>>'"?

I can see what you mean about the functions and variables now. Wow, C++ is such a mess Thanks for the example

Umm...

#include <vector>
#include <list>
int main() {
  std::vector<std::pair<int,std::list<int > > > foo1;
  std::vector<std::pair<int,std::list<int>>>>>> foo2;
  std::vector<std::pair<int,std::list<int>>>>>  foo3;
}

MSVC 7.1:

$ cl temp.cpp -nologo
temp.cpp
temp.cpp(5) : error C2947: expecting '>' to terminate template-argument-list, found '>>'
temp.cpp(5) : error C2947: expecting '>' to terminate template-argument-list, found '>>'
temp.cpp(5) : error C2143: syntax error : missing ';' before '>'
temp.cpp(5) : error C2143: syntax error : missing ';' before '>'
temp.cpp(6) : error C2947: expecting '>' to terminate template-argument-list, found '>>'
temp.cpp(6) : error C2947: expecting '>' to terminate template-argument-list, found '>>'
temp.cpp(6) : error C2143: syntax error : missing ';' before '>'
temp.cpp(6) : error C2143: syntax error : missing ';' before '>'

MSVC 8.0:

$ cl -nologo temp.cpp
temp.cpp
temp.cpp(5) : error C2143: syntax error : missing ';' before '>'
temp.cpp(5) : error C2143: syntax error : missing ';' before '>'
temp.cpp(6) : error C2143: syntax error : missing ';' before '>'
temp.cpp(6) : error C2143: syntax error : missing ';' before '>'

GCC 3.4.4:

$ gcc temp.cpp
temp.cpp: In function `int main()':
temp.cpp:5: error: `>>' should be `> >' within a nested template argument list
temp.cpp:5: error: `foo2' undeclared (first use this function)
temp.cpp:5: error: (Each undeclared identifier is reported only once for each function it appears in.)
temp.cpp:5: error: spurious `>>', use `>' to terminate a template argument list
temp.cpp:5: error: expected primary-expression before '>>' token
temp.cpp:6: error: `>>' should be `> >' within a nested template argument list
temp.cpp:6: error: spurious `>>', use `>' to terminate a template argument list
temp.cpp:6: error: expected primary-expression before '>' token
temp.cpp:6: error: `foo3' undeclared (first use this function)

The C++ standard has no mention of this in the template grammar.

Here you are

compiled without any warnings and printed out 256
gcc version 4.1.2 (Gentoo 4.1.2)

Technically, if it has access to the symboltable as it parses, and has some extra lookahead, it can see its in foo and bar, and needs a int, which 1024 >> 2 resolves to. Its not impossible, and I bet at somepoint GCC could handle it with its new parser, unless they decide to stick strictly to the specs, which it never has before (gcc has the best extensions ).

Ick. If that's C++0x, I'm sticking to C++ && Boost.

Same neat features, none of the insanity.

Edit: Trigraphs.