I hadn't realized till just now that ts_debug()'s output is not
compatible with the way the function was defined in 8.2 contrib.
But since apparently backwards-compatibility is not a controlling
factor here, I have a couple suggestions:
* It seems like a bad idea to merge the controlling-dictionary and
resulting-lexemes values into a single text column. This may be
readable enough, but it's pretty horrid if you want to do any
postprocessing on the result. I suggest splitting this into a
regdictionary column and a text[] column, both of which yield NULL
for an unrecognized token. As far as I can see at the moment this
will require two evaluations of the pg_ts_config_map sub-select,
which is a tad annoying, but we shouldn't be foreclosing easy
postprocessing of the result.
* Personally I find the forced mixed-case names of the output columns
to be pretty darn inconvenient when I want to do anything with the
output. Since the previous incarnation of the function didn't use
mixed-case names, it's obvious that there's no field experience
suppporting this decision. May I suggest dropping the mixed-case names?
It's not too late to reconsider this stuff before beta2 ...
regards, tom lane