Protocol for thread communication

castironpi at gmail.com castironpi at gmail.com
Wed Mar 5 16:23:10 CET 2008


On Mar 4, 11:12 pm, Michael Torrie <torr... at gmail.com> wrote:
> Does anyone have any recommended ideas/ways of implementing a proper
> control and status protocol for communicating with threads?  I have a
> program that spawns a few worker threads, and I'd like a good, clean way
> of communicating the status of these threads back to the main thread.
> Each thread (wrapped in a very simple class) has only a few states, and
> progress levels in those states.  And sometimes they can error out,
> although if the main thread knew about it, it could ask the thread to
> retry (start over).  How would any of you do this?  A callback method
> that the thread can call (synchronizing one-way variables isn't a
> problem)?  A queue?  How would the main thread check these things?
> Currently the main thread is polling some simple status variables.  This
> works, and polling will likely continue to be the simplest and easiest
> way, but simple status variables are very limited.  Are there any
> pythonic patterns people have developed for this.
>
> thanks.
>
> Michael

It depends on the messages.  In the general form, 'do X', there are a
lot of specifics relevant.

- Do X at time T

- for relative time T
- for absolute time T
- for conditional time T

While the wording is vague, it illustrates a contrast.

- Do X at time T

- relative T -- T is a time offset into the future
- absolute time T -- T is an absolute time, independent of the rest of
the program
- conditional time T -- under certain conditions, such as, right after
X' finishes, right before X' starts, if Y happens during X

In the big picture, the choice of X is irrelevant.  However, in a
particular PL, in a particular data, encapsulation, or abstraction
model, some combinations of ( T, X ) may be impossible, or even long-
winded, ugly, or wordy.

For the idealists, an pure imperative model may fit best, so let X =
'set a bit', and conflict never arises.  However such one restricts
the universal turing machine-- can only write one value, which may or
may not forbid solution of some problems.

But without loss of generality, for X in { set B, reset B }, you have
a conflict when for OPS= { (T1, X1), (T1, X2) }, X1= set B, and X2=
reset B.  Then there exists an ( X, T ) such that the semantics of the
program are undefined.



More information about the Python-list mailing list