On Tue, 2011-03-22 at 13:27 +1100, Nathan Scott wrote:
> ----- Original Message -----
> > Making libpcp threadsafe is a (stretch) goal for PCP 4.0.
>
> Oh, related note, if we are going to do this 4.0 seems like the
> right time, so we should make it a concrete (non-stretchy) goal.
Agreed ... but I am not sure I have the expertise nor the time to tackle
all of the items on Greg's list, plus there may be other issues that
have been introduced through ignorance since his analysis was done.
And we don't have any testing framework in place to check any of this
out in a multi-threaded environment (not to mention the fact that my
main development platform has a single cpu driven off a coal-fired steam
boiler).
And this is just libpcp, not any of the other libraries.
And finally there are not build tools in place to choke on the build if
we accidently reintroduce something that is blatently thread unsafe.
With this caveats, I think "stretch" is about as far as I can commit at
the moment!
> I think for 4.0 we should also plan to transition away from the
> old libpcp_trace library and come up with a different plan, be
> that either telling people to use something else (like dtrace,
> stap, LTT/UST instrumentation) etc. I'll look into this, but if
> you could add that to the 4.0 list, that'd be great.
I'll add this to the list.
Ta.
|