Add: two template-heads is a cursed parse (as per Sean)

[atomgalaxy]

@seanbax writes:

---

Yesterday I was favoring separating the template parameters into separate lists for the class and function, but I started doing some work this morning and realized that would be a cursed parse. Very hard to distinguish that usage from an extraneous template-header scenario:

template<typename T>
extern void func<T>(T x, T* p);   // uses one template-head

template<typename T1>
template<typename T2>
virtual void cpo<T1>(T2 x);  // uses two template-heads, yet looks exactly the same except for the virtual keyword.

Supporting that syntax would require matching the virtual keyword, knowing that we're in a namespace, and advancing the header index when parsing the temp-id cpo<T1>, with the anticipation that the next token is a (. This gets into modifications of this paragraph:
https://eel.is/c++draft/temp.expl.spec#8

Not a show-stopper by any means, but I'm against adding novel rules to the already super difficult act of balancing template heads and argument lists.

---

Add this to the discussion

[lewissbaker]

Would this parse be even more difficult if we replaced the virtual keyword with a trailling context-sensitive keyword?
You wouldn't know it was a customisable function declaration then until the very end of the declaration line.