word_stem: Presto vs Velox value mismatch

[peterenescu]

Description

word_stem produces mismatched values between Presto/Velox queries.

Reproduction

(Velox followed by Presto)

presto:di> select word_stem(normalize(c0)) from (values (varchar '0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n')) t(c0);
                                        _col0                                        
-------------------------------------------------------------------------------------
 0q7,xcii4$[;jz<vjp11ndu<k`?\nd`qg<|yj8hf4hjtfk9o+dvnz7y()=9"s9o{9i{:{/}jsv-<anwj`^n 
(1 row)
presto:di> select word_stem(normalize(varchar '0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n'));
                                        _col0                                        
-------------------------------------------------------------------------------------
 0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n 
(1 row)
presto:di> 

Relevant logs

normalize produces identical results: (Velox followed by Presto)

presto:di> select normalize(c0) from (values (varchar '0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n')) t(c0);
                                        _col0                                        
-------------------------------------------------------------------------------------
 0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n 
(1 row)
presto:di> select normalize(varchar '0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n');
                                        _col0                                        
-------------------------------------------------------------------------------------
 0Q7,XcIi4$[;Jz<vJp11ndu<k`?\Nd`qG<|YJ8hf4HJtfk9o+DvNz7y()=9"S9o{9I{:{/}jsV-<AnWJ`^n 
(1 row)

[kagamiori]

@peterenescu, is this mismatch in word_stem or in normalize? Can we replace normalize(c0) with its result as a constant literal to make it clearer?

[peterenescu]

normalize is not the issue, added some additional queries in the Relevant logs section to demonstrate so.