On 17-Jul-11 5:50 PM, Jeff Squyres wrote:
> Are you saying that you don't link against symbols from that header file -- you only use struct definitions / #defines ? If so, then I would agree -- it's not necessary.
I'm using macros, data types and inline functions that work on
these data types - basically, getters and setters for struct fields.
> Can you confirm?
> On Jul 17, 2011, at 9:19 AM, Yevgeny Kliteynik wrote:
>> Hi Jeff,
>> On 14-Jul-11 5:26 PM, Jeff Squyres wrote:
>>> On Jul 14, 2011, at 10:16 AM, Yevgeny Kliteynik wrote:
>>>> Also, besides checking if compilation works (assuming that I get
>>>> this file), I have no way checking if linkage also works...
>>> You can AC_LINK_IFELSE to see if the symbols you need are in the libraries.
>>> Er... I'm sorry, I should have noticed this earlier. :-( We usually check for types in headers *and* use AC_LINK_IFELSE to ensure that we can actually link against those symbols as well. You should add this in the ompi_check_openib.m4 testing.
>> I'm not sure it's necessary here.
>> I was talking about Solaris w.r.t. linkage, because I don't know how
>> IB Management stuff looks there. For Linux, I don't actually do any
>> linkage - all I need in the header file. If it's there, then I'm OK.
>> -- YK