Re: [PATCH] compiler: enable CONFIG_OPTIMIZE_INLINING forcibly

[Masahiro Yamada]

On Fri, Oct 4, 2019 at 2:02 AM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Oct 2, 2019 at 7:11 PM Masahiro Yamada
> <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> >
> > Macrofying the 'inline' is a horrid mistake that makes incorrect code work.
> > It would eternally prevent people from writing portable, correct code.
> > Please do not encourage to hide problems.
>
> Honestly, if the alternative to hiding problems is "use a macro", then
> I'd rather hide the problems and just make "inline" means "inline".
>
> If "inline" means "it's just a hint, use macros", then inline is useless.

For clarification,
I am not saying "use macros" at all.


I just want to annotate __always_inline for the case
"2. code that if not inlined is somehow not correct."



> If "inline" means "using this means that there are known compiler
> bugs, but we don't know where they trigger", then inline is _worse_
> than useless.
>
> I do not see the big advantage of letting the compiler say "yeah, I'm
> not going to do that, Dave".
>
> And I see a *huge* disadvantage when people are ignoring compiler
> bugs, and are saying "use a macro". Seriously.


Again, not saying "use a macro".



>
> Right now we see the obvious compiler bugs that cause build breakages.
> How many non-obvious compiler bugs do we have? And how sure are you
> that our code is "correct" after fixing a couple of obvious cases?
>
> As to "portable", nobody cares. We're a kernel. We aren't portable,
> and never were.
>
> If this is purely about the fact that x86 is different from other
> architectures, then let's remove the "compiler can do stupid things"
> option on x86 too. It was never clear that it was a huge advantage.
>
>                Linus



-- 
Best Regards
Masahiro Yamada