simple parameter substitution in regexp_matches postgreSQL function - sql

Question

I have a table with a structure like this...
the_geom data
geom1 data1+3000||data2+1000||data3+222
geom2 data1+500||data2+900||data3+22232
I want to create a function that returns the records by user request.
Example: for data2, retrieve geom1,1000 and geom2, 900
Till now I created this function (see below) which works quite good but I am facing a parameter substitution problem... (you can see I am not able to substitute 'data2' for $1 in... BUT yes I can use $1 later
regexp_matches(t::text, E'(data2[\+])([0-9]+)'::text)::text)[2]::integer
MY FUNCTION
create or replace function get_counts(taxa varchar(100))
returns setof record
as $$
SELECT t2.counter,t2.the_geom
FROM (
SELECT (regexp_matches(t.data::text, E'(data2[\+])([0-9]+)'::text)::text)[2]::integer as counter,the_geom
from (select the_geom,data from simple_inpn2 where data ~ $1::text) as t
) t2
$$
language sql;
SELECT get_counts('data2') will work **but we should be able to make this substitution**:
regexp_matches(t::text, E'($1... instead of E'(data2....
I think its more a syntaxis issue, as the function execution gives no error, just interprets $1 as a string and gives no result.
thanks in advance,

Answer 1

A E'$1' is a string literal (using the escape string syntax) containing a dollar sign followed by a one. An unquoted $1 is the first parameter to your function. So this:
regexp_matches(t, E'($1[\+])([0-9]+)'))[2]::integer
as you've found, won't interpolate the $1 with the function's first parameter.
The regex is just a string, a string with an internal structure but still just a string. If you know that $1 will be a normal word then you could say:
regexp_matches(t, E'(' || $1 || E'[\+])([0-9]+)'))[2]::integer
to paste your strings together into a suitable regex. However, it is better to be a little paranoid, sooner or later someone is going to call your function with a string like 'ha ha (' so you should be prepared for it. The easiest way that I can think of to add an arbitrary string to a regex is to escape all the non-word characters:
-- Don't forget to escape the escaped escapes! Hence all the backslashes.
str := regexp_replace($1, E'(\\W)', E'\\\\\\1', 'g');
and then paste str into the regex as above:
regexp_matches(t, E'(' || str || E'[\+])([0-9]+)'))[2]::integer
or better, build the regex outside the regexp_matches to cut down on the nested parentheses:
re := E'(' || str || E'[\+])([0-9]+)';
-- ...
select regexp_matches(t, re)[2]::integer ...
PostgreSQL doesn't have Perl's \Q...\E and the (?q) metasyntax applies until the end of the regex so I can't think of any better way to paste an arbitrary string into the middle of a regex as a non-regex literal value than to escape everything and let PostgreSQL sort it out.
Using this technique, we can do things like:
=> do $$
declare
m text[];
s text;
r text;
begin
s = E'''{ha)?';
r = regexp_replace(s, E'(\\W)', E'\\\\\\1', 'g');
r = '(ha' || r || ')';
raise notice '%', r;
select regexp_matches(E'ha''{ha)?', r) into m;
raise notice '%', m[1];
end$$;
and get the expected
NOTICE: ha'{ha)?
output. But if you leave out the regexp_replace escaping step, you'll just get an
invalid regular expression: parentheses () not balanced
error.
As an aside, I don't think you need all that casting so I removed it. The regexes and escaping are noisy enough, there's no need to throw a bunch of colons into the mix. Also, I don't know what your standard_conforming_strings is set to or which version of PostgreSQL you're using so I've gone with E'' strings everywhere. You'll also want to switch your procedure to PL/pgSQL (language plpgsql) to make the escaping easier.