[illumos-Developer] Cluster-installed "mpi.d" interfering with ON libc plockstat.d build

[Richard Lowe]

On Wed, Mar 9, 2011 at 16:30, Dan McDonald <danmcd at nexenta.com> wrote:
> Now I'm a little surprised the dtrace invocation in an ON build takes things
> from /usr/lib/dtrace instead of $ROOT/usr/lib/dtrace, but maybe I shouldn't
> be.

Not directly related, but this shafts anyone building anything
involving USDT, I believe
not just illumos.

> Apart from removing the offending mpi.d during a build, I wonder (in order of
> what I perceive to be likely):
>
>        1.) Is mpi.d Just Plain Broken (TM), and needs a fix?  There's hints
>            of S10-specific gorp at the beginning of the file, but not on
>            line 70, and there's no indication of kthread_t shenanigans.

Yes.

>        2.) Should the dtrace(1M) invocation during libc's build consult
>            /usr/lib/dtrace at all?  Shouldn't $ROOT/usr/lib/dtrace have what
>            we need?

Perhaps, but there may be implications I'm missing.

-- Rich