Clang: “warning: variable 'b' is uninitialized when used”

[chris0e3]

This is based on the v1.0.8 sources. I see the following warning when compiling libde265-1.0.8:

libde265/encoder/algo/pb-mv.cc:228:24: warning: variable 'b' is uninitialized when used here
      [-Wuninitialized]
    else { b=abs_value(b+2); }
                       ^
libde265/util.h:115:25: note: expanded from macro 'abs_value'
#define abs_value(a) (((a)<0) ? -(a) : (a))
                        ^
libde265/encoder/algo/pb-mv.cc:224:10: note: initialize the variable 'b' to silence this warning
    int b;
         ^
          = 0
libde265/encoder/algo/pb-mv.cc:217:24: warning: variable 'b' is uninitialized when used here
      [-Wuninitialized]
    else { b=abs_value(b+2); }
                       ^
libde265/util.h:115:25: note: expanded from macro 'abs_value'
#define abs_value(a) (((a)<0) ? -(a) : (a))
                        ^
libde265/encoder/algo/pb-mv.cc:213:10: note: initialize the variable 'b' to silence this warning
    int b;
         ^
          = 0

In the function Algo_PB_MV_Search::analyze in file encoder/algo/pb-mv.cc, lines 217 & 228 (else { b=abs_value(b+2); }) are incorrect as b is uninitialized.

If b is assumed to be 0 then this should be else { b = 2; }. But, in the context of the function, that doesn’t look right. Also, it would be good to replace all usage of the abomination that is abs_value with std::abs :-).

Additionally at line 206 in the same function you have:

  double lambda = 10.0;

  double *bits_h = new double[2*hrange+1];
  double *bits_v = new double[2*vrange+1];

The use of double is unnecessary. This could be replaced with:

  constexpr int lambda = 10;
  auto const bits_h = new int[2 * (hrange + vrange) + 2],
             bits_v = bits_h + 2 * hrange + 1;

and line 315 (delete[] bits_v;) removed.

This would save memory, time & many redundant int → double → int conversions.

[chris0e3]

Ping!

[farindk]

This is in the encoder source. The encoder is basically dead code. Nobody is using it.