Re: HEADS UP! KSE needs more attention

Doug Rabson wrote:
> On Mon, 2004-06-07 at 16:33, Scott Long wrote:
> 
>>Doug Rabson wrote:
>>
>>>On Sunday 06 June 2004 20:55, Daniel Eischen wrote:
>>>
>>>
>>>>On Sun, 6 Jun 2004, Scott Long wrote:
>>>>
>>>>
>>>>>All,
>>>>>
>>>>>We are about 4-6 weeks away from starting the 5.3 release cycle. 
>>>>>As it stands, KSE still only works reliably on i386.  There are
>>>>>reports of significant instability on amd64, and it doesn't work at
>>>>>all on alpha and sparc64.  I'm willing to drop the alpha
>>>>>requirement and maybe even the sparc64 requirement, but there
>>>>>absolutely will not be a 5.3 until amd64 is solid.  Please contact
>>>>>myself, Dan Eischen, and David Xu if you are interested in helping
>>>>>out.
>>>>
>>>>amd64 looks to be a problem in readline which doesn't seem
>>>>to redispatch signal handlers with SA_SIGINFO arguments.
>>>>
>>>>David also has patches for debugging support at:
>>>>
>>>> http://people.freebsd.org/~davidxu/kse/dbg/
>>>>
>>>>Doug Rabson also has basic TLS support working in perforce.
>>>>It'd be nice to get TLS and debugging in before 5.3-release.
>>>
>>>
>>>I'll probably try to commit some kind of TLS support into current in the 
>>>next couple of weeks. Its likely to only support i386 and will be 
>>>stubbed out for other platforms. Right now, I'm just waiting for some 
>>>kind of feedback from an nvidia developer whos testing it.
>>
>>I'm happy to see that it's gotten this far along, but it needs to be
>>available on all tier-1 platforms when it goes in.  For the sake of
>>argument, we'll call that amd64 and possibly sparc64.  My understanding
>>is that amd64 should be very similar to i386, yes?  Please let me know
>>what kind of help you need with getting this done.
> 
> 
> Well I'm not so sure that it does need to be supported on all tier 1
> platforms. If a given tier 1 platform has a TLS-capable toolchain then
> sure - its only about a hundred lines of code to add support once the
> toolchain support is there. Currently our toolchain only supports TLS on
> i386 and ia64. I think that a binutils upgrade might fix amd64 but the
> rest need a new compiler.
> 
> Right now, there is only one real application which absolutely needs TLS
> and that is the NVidia OpenGL driver on i386. It looks like the DRM
> people will start using TLS at some point but they aren't using it now.
> If we reach 5.3 without any new toolchain support for TLS, I think we
> should ship with TLS support on only i386 and ia64. It seems silly to
> not support TLS on i386 (which people are begging for) because we can't
> do it on sparc64 (which no-one needs).
> 
> 

I agree completely.  I forgot to stress in the previous mail that there
is a strong push at the moment to get in gcc34 and binutils 2.15(?),
which should make TLS infrastructure available to amd64 and sparc64.
I don't think that sparc64 should be a hard requirement, but it should
at least be looked at and documented so that someone else can come along
and do the work.

Scott