OK, I guess MPI standard wants to have this kind of code.

int valuelen;
MPI_Info_get_valuelen(..., &valuelen,...);
int buflen = valuelen + 1;
char* value = (char *)malloc(buflen);
MPI_Info_get(..., valuelen, value, ...);

Though, from a user's point of view, I fell better if it was MPI_Info_get(..., buflen, value, ...), that is, using buflen to specify the length of buffer I pass to MPI.

--Junchao Zhang

On Tue, Feb 3, 2015 at 4:26 PM, William Gropp <wgropp@illinois.edu> wrote:

Thanks. Yes, I believe this is a bug. The implementation of MPI_Info_get in MPICH passes valuelen to the internal copy routine, yet it is clear that the definition in the standard for MPI_Info_get, it should pass valuelen+1 (it should also check for the error return code from the internal routine and return a value MPI error code if the buffer is too small.

Bill

On Feb 1, 2015, at 12:13 AM, Hajime Fujita <hfujita@uchicago.edu> wrote:

Hello,

I came across some question about a behavior of MPI_Info_get in MPICH.

If I knew the size of the value, what should I pass to MPI_Info_get as `valuelen`?
Suppose I have 3-byte value in the info object, and I allocate 4-byte buffer including a space for null-termination. Then I think I should specify 3 as valuelen to retrieve the whole value, as the MPI-3 specifications says
"In C, valuelen should be one less than the amount of allocated space to allow for the null terminator.”

However, when I run the attached program in MPICH 3.1.1, I get:
$ ./test
key found: vaule=va # The last character ‘l’ is missing

If I specify `sizeof(buff)` as valuelen, it shows “val” correctly.

So apparently MPICH thinks valuelen includes null terminator.
Am I missing anything, or any misunderstanding here?

Thank you in advance,
Hajime

--
Hajime Fujita
Postdoctoral Scholar, Large-Scale Systems Group
Department of Computer Science, The University of Chicago
http://www.cs.uchicago.edu/people/hfujita

_______________________________________________
discuss mailing list discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss

<test.c>

_______________________________________________
discuss mailing list discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss