- Does anyone recall why the free and cancel callback fns in GR have
the complete flag as an integer instead of a boolean (it is binary).
At first I thought it was due to interlanguage usage but now I don't
see that. I either want to change it to boolean or give a reason why
this case is different.
- We say that if you call MPI_ERROR_STRING and there is no string set
then it raises an error. First, this is inconsistent with the comm
name routine that returns blanks. Second, since you cannot check if
one is set, we have set up a situation where you can easily cause an
error in an error handler. Is there a reason not to change it to
return blank in this case?
- The current text forbids raising an error handler in an error
handler. The text around it discusses the problem of infinite
recursion. Do we really want to be this strict? Would warning about
infinite recursion while allowing raising an error handler be ok? Does
it seem possible that someone would raise an error handler in an
error handler?
- I previously proposed that we add a function to allow one to raise
a file error handler. The problem with this is that the current
routines only create errorcodes/classes for comms. One fix I have
thought of is to make the errorcodes/classes global and not associated
with comm. This seems more consistent with how they are dealt with
(we only have one set in the standard) and fixes this problem. Does
anyone have thoughts?
- The current thread section on waiting says you cannot wait for the
same request in multiple threads. It seems to me that TEST can also
cause the same problems. Is this true?
- Linda raised some good issues about what is returned by the GR
callbacks and what happens because of errors. I welcome thoughts on
this.
- The naming questions raised by Jeff have not been finalized.
Thoughts?
Thanks,
Steve