On 11/06/10 15:56, Gabe Black wrote:
> On 11/05/10 18:06, nathan binkert wrote:
>>> Might be good to download a scons-local 0.98.1 and verify this works, but
>>> otherwise I'm happy.
>> I suggest that you do that. Remember the last time I had that swig
>> problem? These programs sometimes lie about what their version
>> supports.
>>
>> Nate
>> _______________________________________________
>> m5-dev mailing list
>> [email protected]
>> http://m5sim.org/mailman/listinfo/m5-dev
> I actually considered this sort of thing briefly back in the day. Maybe
> it would be a good idea to have functionality tests but also more in
> depth build tests. The functionality tests could run with m5.opt,
> getting the asserts and whatnot in play since I think we're using
> m5.fast in the regressions, and then the build tests could build all the
> variants, try extreme versions of the tools like 0.98.1, etc., but just
> make sure it builds and not that it runs. Maybe for components where
> versions might matter at runtime like the python interpreter we could
> have a simple test (hello world would be great) as a sanity check. Since
> the functionality parts take a long, long time, we could verify this
> other stuff pretty thoroughly and not affect the total runtime much.
>
> Of course, having multiple versions of tools around, selecting between
> them, etc., would be a pain in the butt I think, and I'm certainly not
> volunteering to do this myself at the moment, but it sounds nice.
>
> Gabe
> _______________________________________________
> m5-dev mailing list
> [email protected]
> http://m5sim.org/mailman/listinfo/m5-dev
My change seems to work, but in the process of wedging that old a
version of scons onto my system (no portage support, no download, had to
get it from svn, had to manually "install" it). It looks like there are
a few places where
main.Append(CCFLAGS='-pipe')
is incorrectly interpreted as adding an array of flags "-", "p", "i",
"p", "e", to CCFLAGS. This makes scons choke because it can't compile
it's test programs to configure itself and gets confused trying to find
Python.h. This is apparently not a problem with new versions of scons,
and we inconsistently deal with the problem by sometimes putting []s
around strings and sometimes not.
I don't know for sure if this is an issue with my "installation" method,
aka a big hammer and a callous disregard for how it's -supposed- to
work, but if nothing else we should be consistent with how we stick []s
in there.
Gabe
_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev
