Merge branch 'development' into mbedtls_private_with_python

Conflicts:
         include/mbedtls/ecp.h

Conflict resolved by using the code from development branch
and manually applying the MBEDTLS_PRIVATE wrapping.
diff --git a/include/mbedtls/ecp.h b/include/mbedtls/ecp.h
index a3b4424..05c33ff 100644
--- a/include/mbedtls/ecp.h
+++ b/include/mbedtls/ecp.h
@@ -230,7 +230,7 @@
     int (*MBEDTLS_PRIVATE(t_post))(mbedtls_ecp_point *, void *); /*!< Unused. */
     void *MBEDTLS_PRIVATE(t_data);               /*!< Unused. */
     mbedtls_ecp_point *MBEDTLS_PRIVATE(T);       /*!< Pre-computed points for ecp_mul_comb(). */
-    size_t MBEDTLS_PRIVATE(T_size);              /*!< The number of pre-computed points. */
+    size_t MBEDTLS_PRIVATE(T_size);              /*!< The number of dynamic allocated pre-computed points. */
 }
 mbedtls_ecp_group;
 
@@ -277,15 +277,16 @@
 
 #if !defined(MBEDTLS_ECP_FIXED_POINT_OPTIM)
 /*
- * Trade memory for speed on fixed-point multiplication.
+ * Trade code size for speed on fixed-point multiplication.
  *
  * This speeds up repeated multiplication of the generator (that is, the
  * multiplication in ECDSA signatures, and half of the multiplications in
  * ECDSA verification and ECDHE) by a factor roughly 3 to 4.
  *
- * The cost is increasing EC peak memory usage by a factor roughly 2.
+ * For each n-bit Short Weierstrass curve that is enabled, this adds 4n bytes
+ * of code size if n < 384 and 8n otherwise.
  *
- * Change this value to 0 to reduce peak memory usage.
+ * Change this value to 0 to reduce code size.
  */
 #define MBEDTLS_ECP_FIXED_POINT_OPTIM  1   /**< Enable fixed-point speed-up. */
 #endif /* MBEDTLS_ECP_FIXED_POINT_OPTIM */