[Cluster-devel] [PATCH] Fix libdlm static build

[Fabio M. Di Nitto <fdinitto@redhat.com>]

On Fri, 4 Jul 2008, Christine Caulfield wrote:

> Fabio M. Di Nitto wrote:
>> On Fri, 4 Jul 2008, Christine Caulfield wrote:
>> 
>>> Lon Hohberger wrote:
>>>> On Thu, 2008-07-03 at 14:06 +0100, Christine Caulfield wrote:
>>>> 
>>>>> I don't understand the problem you are trying to fix here. Having PIC 
>>>>> objects in the dynamic library and non-PIC in the static is perfectly 
>>>>> standard practice.
>>>>> 
>>>> 
>>>> Well, you can't build a loadable module using a static library w/o
>>>> building PIC.
>>> 
>>> Good. Then build against the dynamic one like you're supposed to!
>> 
>> Generally this is right, we still want to ship a working version of the 
>> static one :)
>
> You still haven't said just what it is that is broken about the static 
> library. Given that my test programs in dlm/tests/usertest all use it, it 
> can't be totally broken surely ??

Just to summarize a bit what we discussed on IRC:

-fPIC changes the way in which the code is generated and built by gcc. So 
building the static version the same way as the shared, will keep it the 
same across. This is important on some architectures like powerpc 
(according to gcc man page).

What lon said in the other email is also a valid point.

As extra bonus:

the makefile becomes simpler and should be possible to collapse like the 
others (didn't check this in details yet since dlm is the only one 
generating more than one shared lib from the same dir.).

I don't dare to say (just kidding here!) that it takes less time to build 
since you will build 50% less objects ;)

Anyway, enough said, it's a trivial change to keep things aligned and I 
see no reason to have libdlm having this special exception.

It's up to David or you (i still don't know who of you owns what in dlm) 
if you want it or not.

Fabio

--
I'm going to make him an offer he can't refuse.