Re: Collector Panes

On Wed, 8 Aug 2007 07:06:17 -0700, David McClain <dbm@refined-audiometrics.com> wrote:

> I think the point I was trying to make in the last posting was that,
> by putting responsibility for executing in the correct thread on the
> user, that implies a level of detailed system and architecture
> knowledge that would not ordinarily be known to users, or easily
> forgotten after a few years, especially on something as ubiquitous
> as Streams. (hence the question about Pane process or Editor
> process?)
>
> I had a similar situation arise when I wrote the interface to
> Aquaterm. All calls to Aquaterm absolutely had to take place on a
> special thread. But rather than force that upon the user, I made
> sure that all the primitives available to the user were instrumented
> to take care of this themselves.
>
> If I lived in a perfect world, that would be the case for Streams,
> certainly, and also for all CAPI primitives that are ordinarily user
> accessible. We are quite far removed from the design of CAPI and
> there are many details that we simply wouldn't know about. This is
> all the more critical when CAPI mutates in future versions, or when
> there are gross differences in operational characteristics among the
> various platforms, e.g., Windows, OS/X, Linux.

FWIW, I fully agree with this.  I once had significant problems trying
to port a CAPI app from Windows/Linux to OS/X because of exactly these
reasons - what should happen in which thread - and I eventually gave
up.  I hope that one of the next releases will shield us from these
issues.

Edi.