default_charset and friends

Yasuo, internals, doc-friends, I'm working on the 5.6 migration guide, and I'm a little confused about what state the default encoding RFC actually ended up in after it was accepted and merged. UPGRADING says: > Changes were made to character set handling in: > - the iconv and mbstring extensions, > - and htmlentities(), htmlspecialchars(), html_entity_decode() functions > > The precedence for these is now: > > default_charset < internal/input/output_encoding < (mbstring.* || iconv.*) < function > parameter > > For example, the easiest way to use the UTF-8 encoding is to set > default_charset=UTF-8 and leave the following php.ini parameters The way this reads to me, assuming I don't have any other encoding settings set, I should be able to set default_encoding to (for example) "cp1252" and get Windows-1252 handling as the default in htmlentities(), htmlspecialchars() and html_entity_decode() if I don't specify the encoding parameter. In practice, though, that doesn't seem to be the case. I created this script: <?php ini_set('default_charset', 'cp1252'); var_dump(htmlentities("\xA3", ENT_HTML5)); var_dump(htmlentities("\xA3", ENT_HTML5, 'cp1252')); var_dump(html_entity_decode("&pound;", ENT_HTML5)); var_dump(html_entity_decode("&pound;", ENT_HTML5, 'cp1252')); ?> And ran it using the -n option, so that anything I had set in my php.ini would be ignored. My expected output would have been for the two htmlentities() and html_entity_decode() calls to return the same strings (ignore the unknown character glyphs; my terminal is UTF-8, so I expect Windows-1252 output to be broken): string(7) "&pound;" string(7) "&pound;" string(1) "�" string(1) "�" But instead, I got this output, suggesting that the calls without explicit charset parameters were treated as UTF-8: string(0) "" string(7) "&pound;" string(2) "£" string(1) "�" What am I missing here? Do htmlentities(), htmlspecialchars() and html_entity_decode() actually respect default_charset? (Also, if I set internal_encoding, input_encoding and output_encoding, which one should get used for each?) Thanks, Adam