Re: [dpdk-dev] [PATCH v2] mk: fix acl library static linking

[Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>]

On 01/07/2016 11:05, Thomas Monjalon wrote:
> 2016-07-01 09:05, Sergio Gonzalez Monroy:
>> On 30/06/2016 17:22, Thomas Monjalon wrote:
>>> 2016-06-30 17:14, Sergio Gonzalez Monroy:
>>>> On 30/06/2016 17:10, Thomas Monjalon wrote:
>>>>> 2016-06-30 17:01, Sergio Gonzalez Monroy:
>>>>>> --- a/mk/rte.app.mk
>>>>>> +++ b/mk/rte.app.mk
>>>>>> @@ -76,12 +76,13 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_IP_FRAG)        += -lrte_ip_frag
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_METER)          += -lrte_meter
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_SCHED)          += -lrte_sched
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_LPM)            += -lrte_lpm
>>>>>> -_LDLIBS-$(CONFIG_RTE_LIBRTE_ACL)            += -lrte_acl
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_JOBSTATS)       += -lrte_jobstats
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_POWER)          += -lrte_power
>>>>>>     
>>>>>>     _LDLIBS-y += --whole-archive
>>>>>>     
>>>>>> +# librte_acl needs --whole-archive because of weak functions
>>>>>> +_LDLIBS-$(CONFIG_RTE_LIBRTE_ACL)            += -lrte_acl
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_TIMER)          += -lrte_timer
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_HASH)           += -lrte_hash
>>>>>>     _LDLIBS-$(CONFIG_RTE_LIBRTE_VHOST)          += -lrte_vhost
>>>>> I was suggesting to keep -lrte_acl at the same place in the group of
>>>>> algorithms libraries, in order to keep an order satisfying this comment:
>>>>> # Order is important: from higher level to lower level
>>>>>
>>>>> But I have not tested if --whole-archive -lrte_acl --no-whole-archive works.
>>>>>
>>>> Sorry, I missed that.
>>>>
>>>> Why is important being before jobstats and power?
>>> It is not.
>>> But I think we need to have some groups.
>>> And ACL is probably at the same layer level as lpm, sched, etc.
>> I guess I just don't see the groups you are mentioning :)
> I define groups as separated by blank line :)
>
>> How are timer, hash and vhost in the same group?
> It is far from perfect and subject to improvements :)
>
>> Wouldn't hash be in the same group as acl and lpm?
> It makes sense to use hash in drivers (example: enic).

You have not convinced me, but I'm not going to argue more over this.

You would just prefer to do the following, right?

+_LDLIBS-$(CONFIG_RTE_LIBRTE_ACL)            += --whole-archive
_LDLIBS-$(CONFIG_RTE_LIBRTE_ACL)            += -lrte_acl
+_LDLIBS-$(CONFIG_RTE_LIBRTE_ACL)            += --no-whole-archive