comments on chapter

Rajeev Thakur (thakur@mcs.anl.gov)
Thu, 24 Oct 1996 18:33:12 -0500

The chapter looks quite good. Here are my comments...

General Comment
---------------

I think a new reader would find it hard to understand data
partitioning just from the definitions in 10.1.1. It was better
explained in the few paragraphs in Section 10.1.1 of the old document.
For example, the definition of displacement in the current 10.1.1
doesn't clearly indicate what displacement is actually used for.
In general, I prefer the old 10.1.1.

Important Changes
-----------------

1. 190/22. Should be MPI_File *fh

2. 191/10. Should say something like "sets new values for hints"

3. 191/42. Don't think the sentence is necessary, since MPI_info is
an opaque object. User will define it as "MPI_Info info".
I think it is an MPI convention that memory for OUT arguments
is allocated by the user, not by MPI. The only time we need
to specify that an object needs to be freed is when a
special function like Type_free or Comm_free has to be used
for freeing.

4. 195/27. MPI-1 requires that derived datatypes must be committed
before they can be used for anything except for
constructing other datatypes. I would like that
to hold for I/O as well. Therefore, etype and filetype
passed to MPI_Set_view should be committed.

5. 195/34. The word "re-opened" may be a bit misleading to the new
reader, though I understand what you mean.

6. 196/25-30. Same as (4). I would prefer the original etype and
filetype to be committed; therefore, the ones returned
by Get_view will also be committed.

7. 197/37. Add "in the current view" after "next data item".

8. 197/40. There should be a mention somewhere that new_file_position
and old_position are both in units of etype.

9. 198/25-27. At the last meeting there was a discussion about whether
to allow read access to a buffer used in an MPI_Iwrite.
Does the chapter now disallow this?
(I'm not necessarily complaining.)

10. 218/13-24. These lines should be moved to the previous page.

11. 218/36. The second part of the sentence, "or all updates ...", is
not clear. Surely one can write and then read back the same
data without doing a sync.

Small Changes
-------------

1. 187/43-48. Add "See Section 10.5"

2. 189/3. Delete the word "as"

3. 191/13. Delete "the".

4. 186/41-44. If definitions are retained, change all "which" to "that"
in these lines.