Re: [gpsd-dev] SHM broken

[Gary E. Miller]

Yo Hal!

On Tue, 24 Mar 2015 01:06:17 -0700
Hal Murray <address@hidden> wrote:

> 
> > Well, except RedHat gcc 4.9.2 is not the same a Gnu gcc 4.9.2
> > RedHat always plays games with version numerbers and upstream
> > version numbers.
> 
> > Last I checked RedHat gcc 4.9.2 was not C11 by default, thus no
> > memory barriers.  GNU gcc 4.9.2 is C11 by default, thus memory
> > barriers. 
> 
> Thanks.  I've mostly been sheltered from that layer of "interesting"
> quirks.
> 
> It's not just RedHat.  I get similar problems on Debian, Ubuntu,
> NetBSD, and FreeBSD.

Well, more a lack of a new feature than a problem. gpsd lived without
memory barriers for a long time.  ntpd still lacks them.

I see Gentoo is not C11 by default either.  They musta just changed that
back to non-C11 recently.  I know I tested that case.

> The only systems I have handy access to that default to supporting
> c11 are Alpha versions of the next Fedora release.  Their man page
> for gcc says it defaults to gnu11.
> cc (GCC) 5.0.0 20150319 (Red Hat 5.0.0-0.21)
> 
> I tried CC="cc -std=c11".  That barfed on the asm stuff.

Well, the point of C11 was to avoid the asm stuff, so I guess back to 
the lab on C11.

> CC="cc -std=gnu11" worked on some machines dies on others with:
>   unrecognized command line option

Which is why scons checks for what options a compiler takes before
using them.  If that works on some systems we can try it, but full C11
is the goal.

> Removing the C11 test works on a few machines.

Do you actually see working memory barriers on non-C11 compiles?

Any way to detect that so it can be tested for in SConstruct?

> I haven't found a good solution.  (or even a crappy one)

Finding the problem is the hard part, thanks.
 
> Sigh.  I thought this stuff was under control.

Life on the bleeding edge...

>  I guess I'll go bang
> my head against some other wall.

I want to beat my head on the TIOMCIWAIT wall, but other things keep
popping up.  I'll be looking at this some more.

I'm interested in what you find about memory barriers in your non-C11
systems.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97701
        address@hidden  Tel:+1(541)382-8588