On Wed, 2009-09-30 at 17:17 +1000, Nathan Scott wrote:
> ...
> I've pulled the dynamic PMNS changes out into a separate patch for
> reviewing and given it an initial read-through. So far, the only
> thing that I've noticed is the (leftover?) commented out TODO in
> src/libpcp_pmda/src/mainloop.c where a manual loadnamespace looks
> to have been inserted (+ later removed) at some point.
Yep that was a a residual piece from when I was confused ... the code
passed QA without this loadnamespace call, and indeed by analysis it is
not needed ... it is gone in my local tree.
> Otherwise it looks pretty good. I wonder if we should be bumping
> the libpcp_pmda shared library soname for this change? ...
I don't think this is necessary ... these changes are both forwards and
backwards compatible in the sense that there are no loader symbol
changes, none of the ABI syntax is changes, and the semantic changes are
all guarded by PMDA_INTERFACE_? that will work in the old and the new
code.
Changing to a 9-bit domain number might be an argument for moving the
dso version numbers but that would be for all libraries (I suspect) and
not really necessary either as no existing code will break with these
library changes, unless a domain number > 255 is assigned.
> ...
> Also it'd
> be good to convert a "real" PMDA like MMV (which it sounds like
> you've started from that other mail) - maybe a helper routine or
> two might fall out of that that could live in libpcp_pmda, and if
> so, good to get those into the library for the dso vesion change.
That's probably a good idea ... I can help, but I don't profess any
expertise in the mmv pmda, so probably requires some collaborative
effort.
> The patch is at http://oss.sgi.com/~nathans/dynamic_pmns.patch if
> others would like to review it standalone too (please do!).
Thanks.
> cheers.
>
|