Re: [Tarantool-patches] [PATCH v7 1/2] fiber: set diagnostics at madvise/mprotect failure

[Cyrill Gorcunov <gorcunov@gmail.com>]

On Sat, Feb 15, 2020 at 04:41:39PM +0100, Vladislav Shpilevoy wrote:
> > 
> > It highly depends on compiler. It might be optimized and not allocated
> > on the stack at all or it might be allocated and loaded in the procedure
> > prologue (which will happen with optimization disabled).
> 
> This is what I don't like - dependency on whether a compiler will
> optimize something or not.

I mean the different thing. If variable declared as const inside
prologue it might be or might be not rolled in without load. To
prevent possible load I use static var.

> > 
> > And I'll have to modify both -- comment and code if we need
> > to change the permissions. Thanks but no. We could do better
> > and keep R|W in one place.
> 
> I can't imagine why would we need other flags here, but anyway if
> that will happen, I am sure it will change much more than just a
> comment. However, this is subjective.

Yup.

> 
> > But if you bothers this approach I could just inline R|W, no problem.
> 
> Usually for similar things, when it is wanted to have a constant
> expression as a single identifier, enum works well, and we use
> that a lot, for the same reasons. So I propose you to consider a
> enum then. Anonymous enum in fiber.c.

Sure, thanks

> > Still there is a really big question remains -- what to do with
> > the issue. I suspect Kostya has something different in mind. So
> > for now I'm giving up cooking this series (until I'll be sure
> > that I really understand the next step to walk).
> 
> Ok. Btw, I don't understand what is wrong with the current
> solution.

I though of extending code on the top of this series to drop
r|w in release builds completely, but not right now. So lets
wait for Kostya to respond.