[Trilinos-Users] Some Trilinos compilation warnings

[Devine, Karen D]

Andrey:  Thanks for sharing this output.  See comments on Zoltan issues
below. --Karen


On 8/25/12 12:11 AM, "Andrey Prokopenko" <prok at math.uh.edu> wrote:

>Hi everyone,
>
>Late at night, nitpicking at Trilinos, i.e. some compilation warnings
>that could be potential bugs. Maybe, it is a good idea to include
>"-Wall -Wextra" to CXX_FLAGS.
>
>
>(1) trilinos/packages/zoltan/src/hier/hier.c:1780:43: warning: array
>subscript is above array bounds [-Warray-bounds]
>
>seems like an obvious misprint
>
>-    for (i=1; i <= ZOLTAN_HIER_LAST_PLATFORM; i++){
>+   for (i=0; i < ZOLTAN_HIER_LAST_PLATFORM; i++) {

Yes, this is a bug.  I will push a fixed to the development branch.


>(2) trilinos/packages/zoltan/src/rcb/shared.c:2121:7: warning:
>Œcurrent_candidate¹ may be used uninitialized in this function
>[-Wuninitialized]
>
>the code in question does have potentially uninitalized pointer which
>is then dereferenced
>
>ZOLTAN_ID_PTR current candidate;
>[...]
>if (dindx) {
>    int prevpart = -1;
>    /* there is more than one part on this proc. */
>    /* use dindx to access dots grouped by part */
>    for (i = 0; i < dotnum; i++) {
>      if (dotpt->Part[dindx[i]] != prevpart) {
>        /* Assuming dots are grouped according to part number. */
>        /* Then candidate is first dot in new part */
>        current_candidate = &(gidpt[dindx[i]*num_gid_entries]);
>        prevpart = dotpt->Part[dindx[i]];
>      }
>      ZOLTAN_SET_GID(zz, &(dot_candidates[dindx[i]*num_gid_entries]),
>current_candidate);
>  }
>}

Reviewing the code and knowing the data values, I know that
current_candidate will not be used uninitialized.  The compiler can't know
that, though, and we dislike warnings, so I will fix this issue as well.