pandas merge: NaN keys match each other, unlike SQL NULL

Same result on pandas 2.2.3 and 3.1.0. merge treats missing keys as equal: a left NaN key joins a right NaN key, and an object/str column holding None joins None. In pandas 3 the str dtype turns None into NaN, and the match still happens. SQL never matches NULL to NULL, so a port of a SQL join can silently gain rows. Drop or fill null keys before merging if you want SQL semantics. Duplicate right keys multiply rows without warning (a left merge of 4 rows gave 6). Pass validate='m:1' (or '1:1') to raise MergeError on that, and indicator=True to count both/left_only/right_only. An int64 key against a string key raises ValueError rather than coercing, but int64 against float64 (a column that gained a NaN) merges silently on value.