Re: Future development

David Abrahams wrote:
> on Sun Oct 07 2007, Larry Evans
<cppljevans-AT-cox-internet.com> wrote:
> 
>> On 10/06/07 19:08, Douglas Gregor wrote:
>> [snip]
>>> Okay. One could certainly make the executables
for some tests  
>>> dependent on the executables of other tests, so
if the type_traits  
>>> tests fail to build then one doesn't build the
tests for other  
>>> libraries that depend on type_traits.
>>>
>>> I don't think we could make chain tests based
on whether a test  
>>> *executes* correctly.
>>>
>> redflag!
> 
> I'm not alarmed.  It means some tests may attempt to
run needlessly,
> but we'll still get all the results we need.  
> 
>> I think this is important to have.  I actually had
such a need
>> before:
>>
>> http://article.gmane.org/gmane.comp.lib.boost.build/6301

>>
>> I'd think it would be useful because, in general,
the higher
>> level classes or objects in a library depend on the
correct
>> functioning of lower level parts of the library or
other
>> libraries.  Obviously it would be useless to run
tests
>> for the higher level parts if the lower level
parts
>> didn't work correctly. ( Hopefully the adjectives
"higher level"
>> and "lower level" are clear, but if not,
I could try and
>> define them or give examples. )  I'm surprised the
topic
>> hasn't come up more often.
> 
> I'm not aware of any Boost library that actually
creates such
> dependencies among its tests, are you?

For the longest time we had the problem that the
Serialization library 
would get intermittent failures. The problem was that many
of the tests 
relied on a particular ordering. The "load" tests
needed to run after 
the "save" tests. Then at one point I made some
rather ridiculous 
kludges to BBv1 to order the tests correctly. And now with
BBv2 all 
Robert has to do is add a <dependency>save_test on the
load tests and it 
works swimmingly. So it's not a matter of running tests
needlessly. It's 
a matter reliable testing... again.


-- 
-- Grafik - Don't Assume Anything
-- Redshift Software, Inc. - http://redshift-software
.com
-- rrivera/acm.org - grafik/redshift-software.com
-- 102708583/icq - grafikrobot/aim - grafikrobot/yahoo
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost-build

Rene Rivera wrote:
> David Abrahams wrote:
>> on Sun Oct 07 2007, Larry Evans
<cppljevans-AT-cox-internet.com> wrote:
>>
>>> On 10/06/07 19:08, Douglas Gregor wrote:
>>> [snip]
>>>> Okay. One could certainly make the
executables for some tests  
>>>> dependent on the executables of other
tests, so if the type_traits  
>>>> tests fail to build then one doesn't build
the tests for other  
>>>> libraries that depend on type_traits.
>>>>
>>>> I don't think we could make chain tests
based on whether a test  
>>>> *executes* correctly.
>>>>
>>> redflag!
[snip]
> For the longest time we had the problem that the
Serialization library 
> would get intermittent failures. The problem was that
many of the tests 
> relied on a particular ordering. The "load"
tests needed to run after 
> the "save" tests. Then at one point I made
some rather ridiculous 
> kludges to BBv1 to order the tests correctly. And now
with BBv2 all 
> Robert has to do is add a <dependency>save_test
on the load tests and it 
> works swimmingly. So it's not a matter of running tests
needlessly. It's 
> a matter reliable testing... again.

FYI, Kitware has funding this year to add parallel test
execution
support to CMake/CTest.  This will require
inter-dependencies among
tests to be explicitly specified to make sure they execute
in the
correct order.  We should be able to use these dependencies
also to make
sure dependers do not run if their dependees fail.  Perhaps
the
dependers could simply be reported as not run because of
missing
dependencies.

-Brad
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost-build