SERVER-120255 optimized bson::getField() in SBE to use strlen() + length comparison + memcmp instead of byte-by-byte strcmp. This delivered significant improvements for realistic field name lengths (73-75% faster for 64-char names, 21-45% for 17-char names), but regressed repeated_path_traversal and coll_scan_complex_predicate benchmarks by ~17-19% (BF-41932, accepted by Product Performance).

The regression occurs because strlen() is called on every BSON field name during iteration — even on misses — to compute the field name length needed by bson::advance(). With short (1-2 char) field names and high miss rates, this is slower than the old byte-by-byte compare that bailed at the first mismatching byte.