Re: new filter proposal

Rajeev Thakur (thakur@mcs.anl.gov)
Wed, 16 Apr 1997 10:54:42 -0500

Some comments on the new filter proposal sent yesterday:

1/36: I'd like to rename etype_file_extent_fn to dtype_file_extent_fn,
since it is not just for etypes.

2/13: Any restrictions on what datatypes can be passed to
dtype_file_extent_fn should be mentioned here. I think it need only be
passed the basic datatypes used to construct the current etype,
including MPI_LB and MPI_UB.
(In other words, dtype_file_extent_fn need only work for those basic
datatypes and not all 50 others defined in MPI.)

2/27: "at the location"

2/31-32: The I/O chapter no longer uses the word "buftype", so it
should be replaced by "datatype" here.

2/32: The term "derived from" has a different meaning in MPI than what
I think you want over here. I think you want it to be anything
"datatype" was constructed out of. (Even the wording in my sentence is
not precise.) Is it necessary to allow a derived datatype constructed
from the datatype passed in the read/write function? That can be taken
care of using count.

3/9: "storing the" instead of "the storing"

3/21: "The function must copy count data items of type etype from..."

3/28-30: same as in 2/31-32.

3/32: I'd like a space or a subheading before this sentence to
indicate that what follows applies to both read and write filter
functions.

3/40-41: This will need to be changed according to what is in 2/31-32
and 3/28-30. Actually, the sentence may not be needed at all, since it
is already mentioned.

Rajeev