On Wed, Feb 04, 2004 at 04:36:00PM -0800, Matt Massie wrote:
> brooks-
> 
> my main objective with ganglia is to make it as easy to install as
> possible.  the reason i put libdnet in the distribution was to prevent
> requiring people to install libdnet as a prerequisite (especially since
> i only needed a single function from the library for probing network
> interfaces).  requiring the entire library for one function call seemed
> overkill.  we no longer need libdnet.. i have code from steven's library
> which does the same thing as part of the next distribution.
>
> i know i'm coming from the redhat RPM bend here but.. tracking tons of
> external dependencies is no fun.  RPM is a poor piece of software in
> that way.  
>
> there is also another piece to this puzzle.  i don't want sysadmins to
> have to track dependencies on the thousands of machines that they
> install ganglia on.  currently, we statically link against all libraries
> to prevent that problem.  imagine requiring that a person with a 2000
> node cluster install libdnet, expat, etc on all machines they have.
> 
> so since we are statically linking anyway.. i was trying to keep ganglia
> as self-contained as possible.  it's a balancing act.  

Including the library because we're staticly linking make no sense to
me.  In fact, staticly linking is a good argument for removing libdnet
since it will only effect people who build from source.  With a
packaging system that worked, ganglia's dependency set isn't all that
large.  I hate the idea of making design decisions based on the ways
that rpm sucks when these aren't issues elsewhere.

> i guess the real problem here is why ganglia went and pulled in a
> libdnet distribution outside of itself.  
> 
> the short-term solution i recommend is that we track down why ganglia
> didn't ignore the locally installed libdnet.  i would be surprised if
> there have been major changes in libdnet for the source we have (of
> course i could be wrong).

I believe the problem is caused by the fact that I set
CFLAGS=-I/usr/local/include and LDFLAGS=-L/usr/local/lib when building
ganglia in a FreeBSD port so that it will pickup librrd.  This appears
to cause the Makefiles to be generated such that /usr/local/include is
searched before the internal ganglia directories.  This bit me once
before when I had an old version of ganglia installed and got the wrong
header files.

> the long-term solution is that we don't need libdnet source anymore.

This will be good.  Would be be reasionable to merge the changes that do
this into 2.5.x and remove libdnet?

-- Brooks

-- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4

Attachment: pgpxaHotnA8F5.pgp
Description: PGP signature

Reply via email to