Bill Fischofer(Bill-Fischofer-Linaro) replied on github web page:

.travis.yml
line 7
@@ -53,9 +53,6 @@ env:
         - CONF=""
         - CONF="--disable-abi-compat"
         - CONF="--enable-deprecated"
-        - CONF="--enable-schedule-sp"
-        - CONF="--enable-schedule-iquery"
-        - CONF="--enable-schedule-scalable"
         - CONF="--enable-dpdk-zero-copy"


Comment:
@psavol Is @lumag's latest suggestion OK with you? 

> Dmitry Eremin-Solenikov(lumag) wrote:
> @psavol this would also require seprate compile testing of `--with-scheduler` 
> option. I'm thinking about `--with-scheduler-default` option with the 
> following behaviour:
>  - If environment option is given, it selects scheduler/queue implementation
>  - If no environment option is given, compile-time option selects the 
> scheduler/queue
>  - If no compile-time option was given, we use some sensible default (basic 
> for now, maybe scalable in future).


>> Petri Savolainen(psavol) wrote:
>> There can be just one config option: --with-scheduler=foo. When that's 
>> defined then "foo" would be used (fixed at build time). If --with-scheduler 
>> is not found, the value of ODP_SCHEDULER=foo environment variable is used 
>> (run time selection is the default behavior). If the variable is not found 
>> then the default scheduler is used.
>> 


>>> Dmitry Eremin-Solenikov(lumag) wrote:
>>> This would mean that we have now 4 options to test, instead of just one. 
>>> What about renaming 'default' scheduler to some sensible name and adding 
>>> `--with-scheduler-default=FOO`? Then we will have just one combination to 
>>> test.


>>>> Bill Fischofer(Bill-Fischofer-Linaro) wrote:
>>>> Perhaps `--enable-schedule-dynamic` (default) to indicate run time 
>>>> selection that can be overridden by a specific static selection at 
>>>> `./configure` time via the other `--enable-schedule-xxx` options?


>>>>> Petri Savolainen(psavol) wrote:
>>>>> Run time selection is good addition and can be even the default behavior. 
>>>>> Anyway, I'd like to keep the build time selection also, since usually one 
>>>>> application uses the same scheduler always. Fixing the scheduler at build 
>>>>> time avoids accidentally usage of different schedulers in different 
>>>>> servers (or VMs or test environments) and thus avoid problems/debugging 
>>>>> caused by that.
>>>>> 


https://github.com/Linaro/odp/pull/467#discussion_r168653245
updated_at 2018-02-16 00:57:20

Reply via email to