noinline attribute for functions

[PavelVozenilek]

C/C++ has nonstandard constructs like __noinline or __attribute__((noinline)). When coding I am much more certain when I do not wish something to inline, than the opposite. Reducing code bloat is easier to think about than guessing performance impact.

I propose for Zig:

• function annotation no-inline, opposite to inline. Calling @inlineCall on it would be compile error.
• builtin @noInlineCall, opposite to @inlineCall. Calling @noInlineCall on inline function would be compile error.

---

Possibly, syntax for @inlineCall and @noInlineCall could be:

fn foo() -> i32 {...}
...
var x = inline foo();
var y = no-inline foo();

[PavelVozenilek]

As with other annotations, inline/no-inline may be dependent on environment:

fn foo(comptime b : bool) 
{
  if (b) {
inline:
    return;
  }

no-inline:
   // long computation here which is better not to be inlined
   ...
}

[cztomsik]

It is already possible to specify .modifier = .never_inline for the @call

What is still missing is the counterpart for the inline fn.
Maybe what about something like inline(.always) and inline(.never)?
Or we can just close this as wontfix?

[Vexu]

noinline has existed for a long time #3178, not sure why this was left open.