change log for rtems (2010-05-25)

[Chris Johns]

On 26/05/10 12:30 AM, Ralf Corsepius wrote:
> On 05/25/2010 04:10 PM, Thomas Doerfler wrote:
>> Ralf,
>>
>> this define has been added to various BSPs in the past.
> May-be this design flaw already has infected other BSPs - I haven't
> checked :)
>
> The problem is polluting BSPs with an application specific (here:
> testsuites) defines or (worse) files (e.g. my very friend: The tm27.h
> stuff).
>

I am responsible for the "infection". Maybe there are better ways of 
doing this but this has been a simple "hack" so we keep moving while a 
better solution is considered.

>> Is there a
>> problem when this is added during configure time?
> Well, there essentially is nothing wrong in adding more "--enable-<*>"
> options to enable users to skip something, except that such options
> should be avoided whenever possibile.
>

I agree an enable option is not the correct way to handle this problem 
and I avoid them where ever possible.

I see specifically adding a "I have little memory" or something like 
this as no different to adding the CPU clock speed or saying this BSP 
uses a device family's parallel port interface with only 2 ports not 4. 
It is a static setting that does not change. Therefore disabling a small 
memory option building RTEMS does not make the memory appear. It is 
static and so defined in the BSP.

If you have a target that can have a varying amount of memory from big 
to small then the BSP has to absorb this complexity. If all tests could 
run then the linker command file should be able to handle it.

> However there are better approaches ... unfortunately the current time
> is not right to add them.

Agreed. I do not agree with an enable option.

I am open to discuss a better way we could handle localised 
configuration aspects derived from a BSP. Specifically if we know 
statically a test cannot be run for say memory reasons it would be good 
if the test was not build. The current hack is not great.

Chris