GSOC:disable newlibc reentrancy

Joel Sherrill joel.sherrill at OARcorp.com
Thu May 8 17:10:14 UTC 2008


Ralf Corsepius wrote:
> On Thu, 2008-05-08 at 11:33 -0500, Joel Sherrill wrote:
>   
>> Ralf Corsepius wrote:
>>     
>>> On Thu, 2008-05-08 at 10:05 -0500, Joel Sherrill wrote:
>>>
>>>       
>>>> Ralf Corsepius wrote:
>>>>         
>
>   
>>> Then remove this silly CONFIGURE*REENTRANCY crap you added to RTEMS.
>>>
>>> THIS makes builds using it a SPECIAL build and renders shipping binaries
>>> of cpukit IMPOSSIBLE.
>>>
>>>       
>> Ralf.. it doesn't.  I don't know what it is going to take to
>> make you see that configures in confdefs.h have no impact
>> on how cpukit is built.
>>     
>
> The OP asked to use it inside of cpukit!
>   
Either bad English or bad implementation choice.  It didn't
have to be implemented that way.

A patch adding this as a cpukit conditional compile would
have been rejected with the suggestion that it could be
a pluggable thing confdefs.h triggers.
> BTW: The origin of what you call  reentrancy bloat is in newlib and
> newlib's support routines inside of RTEMS.
>
> As I already wrote elsewhere: You are in error to believe RTEMS will
> ever be able to compete with highly optimized bare-metal scheduling
> libraries. This is what makes this whole GSOC project a waste of time.
>
>   
We disagree here.  ARM/Thumb fully linked executables
can be 24k.  A while back, Ray produced a ticker with
some hacks that was around 18K.   Three tasks, full startup,
shutdown, etc.  So we are already approaching the right
range

It isn't about beating a small low functionality scheduler
in size -- it is getting into the same ballpark so we can
run on the same hardware target.  We will win users because
RTEMS is a full system with the capability to scale from that
small machine up to much larger systems.  The
smaller schedulers are the toys that don't scale.

I haven't examined the minimum executable on the ARM/Thumb
in the past two months but the last time I looked, it could be
reduced significantly with no recompile of cpukit.  It just needed
some way to break dependencies and eliminate code it
didn't need.  No cpukit recompile -- all at link time.
> Ralf
>
>
>   


-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel.sherrill at OARcorp.com        On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985





More information about the users mailing list