urlencode with only built-in functions

php urlencode (4)

Here is pretty short version, and it's even "pure SQL" function, not plpgsql. Multibyte chars (including 3- and 4-bytes emoji) are supported.

create or replace function urlencode(in_str text, OUT _result text) returns text as $$
        when ol>1 or ch !~ '[0-9a-za-z:/@._?#-]+' 
          then regexp_replace(upper(substring(ch::bytea::text, 3)), '(..)', E'%\\1', 'g')
        else ch
  from (
    select ch, octet_length(ch) as ol
    from regexp_split_to_table($1, '') as ch
  ) as s;
$$ language sql immutable strict;

Without using plpgsql, I'm trying to urlencode a given text within a pgsql SELECT statement.

The problem with this approach:

select regexp_replace('héllo there','([^A-Za-z0-9])','%' || encode(E'\\1','hex'),'g')

...is that the encode function is not passed the regexp parameter, unless there's another way to call functions from within the replacement expression that actually works. So I'm wondering if there's a replacement expression that, by itself, can encode matches into hex values.

There may be other combinations of functions. I thought there would be a clever regex (and that may still be the answer) out there, but I'm having trouble finding it.

Answer #1
select regexp_replace(encode('héllo there','hex'),'(..)',E'%\\1','g');

This doesn't leave the alphanumeric characters human-readable, though.

Answer #2

decodeURIComponent for postgres

You're not making yourself too clear, but maybe this will be of interest for you.

Answer #3

Here's a function I wrote that handles encoding using built in functions while preserving the readability of the URL.

Regex matches to capture pairs of (optional) safe characters and (at most one) non-safe character. Nested selects allow those pairs to be encoded and re-combined returning a fully encoded string.

I've run through a test suite with all sorts of permutations (leading/trailing/only/repeated encoded characters and thus far it seems to encode correctly.

The safe special characters are _ ~ . - and /. My inclusion of "/" on that list is probably non-standard, but fits the use case I have where the input text may be a path and I want that to remain.

CREATE OR REPLACE FUNCTION oseberg.encode_uri(input text)
  RETURNS text
  LANGUAGE plpgsql
AS $function$
  parsed text;
  safePattern text;
  safePattern = 'a-zA-Z0-9_~/\-\.';
  IF input ~ ('[^' || safePattern || ']') THEN
    SELECT STRING_AGG(fragment, '')
    INTO parsed
    FROM (
      SELECT prefix || encoded AS fragment
      FROM (
        SELECT COALESCE(match[1], '') AS prefix,
               COALESCE('%' || encode(match[2]::bytea, 'hex'), '') AS encoded
        FROM (
          SELECT regexp_matches(
            '([' || safePattern || ']*)([^' || safePattern || '])?',
            'g') AS match
        ) matches
      ) parsed
    ) fragments;
    RETURN parsed;
    RETURN input;