Ruby 3.3.5p100 (2024-09-03 revision ef084cc8f4958c1b6e4ead99136631bef6d8ddba)
noexcept.h File Reference

Defines RBIMPL_ATTR_NOEXCEPT. More...

Go to the source code of this file.

Macros

#define RBIMPL_ATTR_NOEXCEPT(_)
 Wraps (or simulates) C++11 noexcept
 

Detailed Description

Defines RBIMPL_ATTR_NOEXCEPT.

Author
Ruby developers ruby-.nosp@m.core.nosp@m.@ruby.nosp@m.-lan.nosp@m.g.org
Warning
Symbols prefixed with either RBIMPL or rbimpl are implementation details. Don't take them as canon. They could rapidly appear then vanish. The name (path) of this header file is also an implementation detail. Do not expect it to persist at the place it is now. Developers are free to move it anywhere anytime at will.
Note
To ruby-core: remember that this header can be possibly recursively included from extension libraries written in C++. Do not expect for instance __VA_ARGS__ is always available. We assume C99 for ruby itself but we don't assume languages of extension libraries. They could be written in C++98.

This isn't actually an attribute in C++ but who cares...

Mainly due to aesthetic reasons, this one is rarely used in the project. But can be handy on occasions, especially when a function's noexcept-ness depends on its calling functions.

Q&A

  • Q: Can a function that raises Ruby exceptions be attributed noexcept?
  • A: Yes. noexcept is about C++ exceptions, not Ruby's. They don't interface each other. You can safely attribute a function that raises Ruby exceptions as noexcept.
  • Q: How, then, can I assert that a function I wrote doesn't raise any Ruby exceptions?
  • A: __attribute__((__leaf__)) is for that purpose. A function attributed as leaf can still throw C++ exceptions, but not Ruby's. Note however, that it's extremely difficult – if not impossible – to assert that a function doesn't raise any Ruby exceptions at all. Use of that attribute is not recommended; mere mortals can't properly use that by hand.
  • Q: Does it make sense to attribute an inline function noexcept?
  • A: I thought so before. But no, I don't think they are useful any longer.
    • When an inline function attributed noexcept actually doesn't throw any exceptions at all: these days I don't see any difference in generated assembly by adding/removing this attribute. C++ compilers get smarter and smarter. Today they can infer if it actually throws or not without any annotations by humans (correct me if I'm wrong).
    • When an inline function attributed noexcept actually does throw an exception: they have to call std::terminate then (C++ standard mandates so). This means exception handling routines are actually enforced, not omitted. This doesn't impact runtime performance (The Itanium C++ ABI has zero-cost exception handling), but does impact on generated binary size. This is bad.

Definition in file noexcept.h.

Macro Definition Documentation

◆ RBIMPL_ATTR_NOEXCEPT

#define RBIMPL_ATTR_NOEXCEPT ( _)
Value:
noexcept(noexcept(_))
#define _(args)
This was a transition path from K&R to ANSI.
Definition stdarg.h:35

Wraps (or simulates) C++11 noexcept

Definition at line 85 of file noexcept.h.