Niibe writes:

Niibe writes:

Sorry, I was not reading the code of GNUnet well.  I overlooked how the
eddsa_d_to_a function was written and its intention.  I read it again.

Indeed, the eddsa_d_to_a function tries to handle the case where
gcry_mpi_print returns rawmpilen < 32, putting "left pad" by DIGEST.

The problem is:

	DIGEST is not cleared (although comment says so).

I think that the stack had zero-byte for some reason on your 32-bit
machine.

Here is the correction.  Clear DIGEST, as comment says.

diff --git a/src/util/crypto_ecc.c b/src/util/crypto_ecc.c
index 8d9091b23..280603234 100644
--- a/src/util/crypto_ecc.c
+++ b/src/util/crypto_ecc.c
@@ -1273,24 +1273,15 @@ eddsa_d_to_a (gcry_mpi_t d)

   b = 256 / 8; /* number of bytes in `d` */

+  memset (hvec, 0, sizeof hvec);
   /* Note that we clear DIGEST so we can use it as input to left pad
      the key with zeroes for hashing.  */
-  memset (hvec, 0, sizeof hvec);
+  memset (digest, 0, sizeof digest);
   rawmpilen = sizeof (rawmpi);
   GNUNET_assert (0 ==
                  gcry_mpi_print (GCRYMPI_FMT_USG,
 				 rawmpi, rawmpilen, &rawmpilen,
                                  d));
-  if (rawmpilen < 32)
-  {
-    memmove (rawmpi + 32 - rawmpilen,
-             rawmpi,
-             rawmpilen);
-    memset (rawmpi,
-            0,
-            32 - rawmpilen);
-    rawmpilen = 32;
-  }
   hvec[0].data = digest;
   hvec[0].off = 0;
   hvec[0].len = b > rawmpilen ? (b - rawmpilen) : 0;
--