* depth of the tree but potentially wastes more memory. That is, this is a
* direct space versus time tradeoff.
*
- * The large memory model uses twelve bits which means that the are 4096
- * pointers in each tree node. This is more than sufficient to hold the
- * largest defined NID (as of Feb 2019). This means that using a NID to
- * index a sparse array becomes a constant time single array look up.
- *
- * The small memory model uses four bits which means the tree nodes contain
- * sixteen pointers. This reduces the amount of unused space significantly
- * at a cost in time.
+ * The default is to use four bits which means that the are 16
+ * pointers in each tree node.
*
* The library builder is also permitted to define other sizes in the closed
- * interval [2, sizeof(ossl_uintmax_t) * 8].
+ * interval [2, sizeof(ossl_uintmax_t) * 8]. Space use generally scales
+ * exponentially with the block size, although the implementation only
+ * creates enough blocks to support the largest used index. The depth is:
+ * ceil(log_2(largest index) / 2^{block size})
+ * E.g. with a block size of 4, and a largest index of 1000, the depth
+ * will be three.
*/
#ifndef OPENSSL_SA_BLOCK_BITS
-# ifdef OPENSSL_SMALL_FOOTPRINT
-# define OPENSSL_SA_BLOCK_BITS 4
-# else
-# define OPENSSL_SA_BLOCK_BITS 12
-# endif
+# define OPENSSL_SA_BLOCK_BITS 4
#elif OPENSSL_SA_BLOCK_BITS < 2 || OPENSSL_SA_BLOCK_BITS > (BN_BITS2 - 1)
# error OPENSSL_SA_BLOCK_BITS is out of range
#endif