<div dir="ltr"><div>Hello,<br></div><div>I am trying to find out what should happen, according to the <span tabindex="-1" id=":1os.1" style="" class="">MPI</span> 3.0* standard, in the following scenario:<br></div><div>1) process 0 initiates a <span tabindex="-1" id=":1os.2" style="" class="">nonblocking</span> send operation with <span tabindex="-1" id=":1os.3" style="" class="">MPI</span>_<span tabindex="-1" id=":1os.4" style="" class="">Issend</span><br></div><div>2) process 1 matches the message with <span tabindex="-1" id=":1os.5" style="" class="">MPI</span>_<span tabindex="-1" id=":1os.6" style="" class="">Mprobe</span>, but doesn't yet call <span tabindex="-1" id=":1os.7" style="" class="">MPI</span>_<span tabindex="-1" id=":1os.8" style="" class="">Mrecv</span>/<span tabindex="-1" id=":1os.9" style="" class="">MPI</span>_<span tabindex="-1" id=":1os.10" style="" class="">Imrecv</span><br></div><div>3) process 0 calls <span tabindex="-1" id=":1os.11" style="" class="">MPI</span>_Cancel to cancel the send.<br><br></div><div>I
 cannot think of anything the cancel operation could do that wouldn't 
contradict either the specification of <span tabindex="-1" id=":1os.12" style="" class="">MPI</span>_Cancel, or of the matched 
receive functions. How can this situation be handled in compliance with the standard?<br><br></div><div>Thank you in advance,<br></div><div><span tabindex="-1" id=":1os.13" style="" class="">Marek</span> <span tabindex="-1" id=":1os.14" style="" class="">Tomáštík</span><br></div><br><br><br>*or 3.1 -- I believe there weren't any changes in this area?</div>