regex - psql uppercase backreferenced string in regexp_replace -

i have string, processed initcap(), , wanto uppercase part of it.

to specific - want uppercase basic roman digits might occur.

to more specific i'd replace

jana iii sobieskiego

to

jana iii sobieskiego

i suppose use kind of upper-substring-subquery combo achieve it, trying make work in single regexp_replace, this:

select    ulica   --, regexp_matches(ulica , '((^|\s)([xxiivv]+)(\s|$))', 'g')   , regexp_replace(ulica, '((^|\s)([xxiivv]+)(\s|$))', '\2'||upper('q\3q')||'\4' , 'g')  (   select unnest(array['jana iii sobieskiego', 'xx lecia', 'xxx lecia panowania zygmunta iii wazy'])::text ulica ) src 

what happens, upper works on 'static' part of replacement string (q...q), not on backreference.

i

jana qiiiq sobieskiego

anyone has idea how this?

postgresql 9.1

short answer
unfortunately, have tried not possible regexp_replace.

long answer
intro
line

regexp_replace(ulica, '((^|\s)([xxiivv]+)(\s|$))', '\2'||upper('q\3q')||'\4' , 'g') 

is equivalent to

regexp_replace(ulica, '((^|\s)([xxiivv]+)(\s|$))', '\2q\3q\4' , 'g') 

as can see, regexp_replace won't upper case backreferences.

workaround
can create own function take ulica parameter , returns ulica basic roman digits uppercased.

• step 1
  in first step function mark (i choose $$ marker can use any.) part of ulica uppercased this:

     regexp_replace(ulica, '((^|\s)([xxiivv]+)(\s|$))', '\2$$\3$$\4' , 'g') 

• step 2
  in second step, go through resulting string , uppercase each char located between 2 markers.