Note that this is currently untested with server encodings other than UTF-8. The encoding policy used is: JSON nodes and most of the JSON functions still operate in UTF-8. Strings are converted between server encoding and UTF-8 when they go in and out of varlena (text*), and a set of helper functions are implemented to make these conversions simple to apply. It is done this way because converting individual codepoints to/from whatever the server encoding may be is nontrivial (possibly requires a loaded module). The JSON code needs to encode/decode codepoints when it deals with escapes. Although a more clever and efficient solution might be to defer charset conversions to when they're necessary (e.g. round up all the escapes and encode them all at once), this is not simple, and it's probably not much more efficient, either. Conversions to/from server encoding and UTF-8 are no-ops when the server encoding is UTF-8, anyway.