On 6 August 2010 08:49, Brian Gerkey <gerkey@willowgarage.com> wrote:

On Thu, Aug 5, 2010 at 4:08 PM, Daniel Stonier <d.stonier@gmail.com> wrote:
> I had to fix a couple of linkages that went missing in my ros libraries when
> a dependency didn't get linked in (namely rt and pthreads) but haven't had
> any other problems. Mind you, haven't been testing this for very long! I'm
> enabling it as a default in our packages for the moment for testing.
> Fedora and now gentoo (which is how I found out about it) are setting this
> as a default in their builds now (some others as well). Mind you, just
> because someone else is doing it is no justification, but it does mean its
> seeing a bit of testing out in the wild and isn't one of those really
> obscure settings with rather unknown effects.

Happy to hear that it works in general (would be pretty bad if it
didn't...). I'm more concerned about whether we're currently
specifying everything completely, or unknowingly exploiting
overlinkage.

I guess the only way to find out is to build and test everything with
that option included.

brian.

Quick update re testing - our mac guy came back to work this morning and immediately hit a problem - the mac linker does not have --as-needed as an option. This may be a problem across other platforms too?

Mac does have another option which seems to be similar and does appear to work after a few compiles:

-mark_dead_strippable_dylib

Specifies that the dylib being built can be dead strip by any client.
That is, the dylib has no initialization side effects. So if a client
links against the dylib, but never uses any symbol from it, the linker
can optimize away the use of the dylib.