[ofa-general] Re: [PATCH 0/3] ibnetdiscover library "libibnetdisc"

[Ira Weiny]

On Mon, 24 Nov 2008 22:01:51 +0200
Sasha Khapyorsky <sashak at voltaire.com> wrote:

> On 11:30 Mon 24 Nov     , Ira Weiny wrote:
> > On Mon, 24 Nov 2008 21:10:50 +0200
> > Sasha Khapyorsky <sashak at voltaire.com> wrote:
> > 
> > > On 09:42 Mon 24 Nov     , Ira Weiny wrote:
> > > > > 
> > > > > Do not you think this library should be rather part of infiniband-diags,
> > > > > rather than separate package/management sub-project? Personally I would
> > > > > prefer to have this as part of infiniband-diags.
> > > > 
> > > > No, I would like to see it be a stand alone library.  Currently
> > > > infiniband-diags does not provide any library functionality and simply depends
> > > > on the libraries provided by the rest of the management tree.  Don't you think
> > > > this is a good model to follow?
> > > 
> > > Why it must be so - infiniband-diags will be useless without this library.
> > > 
> > > And I would really hate to handle one more package (actually not just one
> > > - libibnetdisc, libibnetdisc-devel, libibnetdisc-static, etc.). I wanted
> > > to remove libibcommon...
> > > 
> > 
> > I think the argument against ibcommon is that it does not provide enough
> > additional functionality to warrant an entire new library.
> 
> It is probably the same case with libibnetdisc (at least now).
> 
> > On the other hand
> > infiniband-diags depends on many libraries:
> > 
> >    AC_CHECK_LIB(ibcommon, ...  <== delete this...
> >    
> > And you still have the following...
> > 
> >    AC_CHECK_LIB(ibumad, ...
> >    AC_CHECK_LIB(ibmad, ...
> >    AC_CHECK_LIB(osmcomp, ...
> >    AC_CHECK_LIB(osmvendor, ...
> >    AC_CHECK_LIB(opensm, ...
> > 
> > I don't think it is in appropriate to have utilities which are dependent on
> > libraries, it is done all the time.
> 
> OTOH it doesn't mean that any new shared code must be done as separate
> subproject.
> 
> The stuff is new. I think it is better to integrate it in smaller
> iterations, to start with the code and functionality and to not bother
> with packaging, dependencies, etc.. If there will be a reason to make
> separate library we can do it, but then we will have a stable code
> already.

As long as the library exists any dependant package can of course use the
library from whatever package we chose (libibnetdisc or infiniband-diags).  We
have some code which is prototyped against ibnetdiscover but we plan on using
this library instead.  This would be separate from infiniband-diags.  But we
can just as easily put a dependancy on infiniband-diags as on libibnetdisc.

The fact is that it was actually easier to put this in a new package rather
than try and integrate with infiniband-diags.  I thought it made for a very
clean conversion by putting the library in as a new patch and then we could
convert the diags as appropriate.

Anyway, I will integrate it as you say and resubmit the patch.

Ira