I'm reviewing the strxfrm patch, and while comparing that code to the code in varstr_cmp (which uses the same UTF8/UTF16 workaround but for strcoll instead), and I noticed that in varstr_cmp we have an optimization to use a stack based buffer instead of palloc if the string is short enough. Is convert_string_datum performance-critical enough to make it worthwhile to put a similar optimization there?
//Magnus ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match