More function inlining

[therontarigo]

It should (in some circumstances) be possible to inline a function that follows these rules:

• Is of void type, or ends in an unconditional return statement.
• Contains no other return statements.

[eldritchconundrum]

Inlining is not that simple. There are many potential issues to tackle.

• are the arguments lvalues, are they assigned in the inlined function, should they be copied to new variable declarations?
• variable shadowing (parameters, locals, and caller's)
• in/out/inout parameters
• evaluation order of expressions with side-effects, e.g. when passing arguments
• is the function called in the middle of an expression?
• does it save code size to inline functions called multiple times?

[laurentlb]

A difficulty is to handle variables correctly, even with shadowing (renaming might be required).

If the function return value is used, it can be difficult to inline it (and do side-effects in the correct order).

int f() {
  for (...) { ... }
  return ...;
} 

void main() {
  int a = g() + f();
}

To inline f() here, you might need to introduce new local variables and split the line.

[therontarigo]

I reopened #469 , as it is the only one with a chance of being easy to implement.

[therontarigo]

I was thinking that it is an unnecessary restriction in the case that the function is called in a simple assignment statement: float x = func(...); However, considering more carefully what the resulting minified code would be, there's no case for that where return-by-inout wouldn't work just as well. So that's one more reason to abandon this and focus on #469 .