Replace substring with binary strings - string

Question

I want to perform a substring replace operation on binary strings. There is a function available that does this exact thing for strings of type text (c.f.):
replace(string text, from text, to text)
But unfortunately none for binary strings of type bytea (c.f.).
Now I wonder, do I need to reimplement this operation for binary strings or can I use the corresponding basic string function for this task? Are there edge cases that could break my application:
select replace('\000\015Hello World\000\015Hello World'::bytea::text,
'World',
'Jenny')::bytea
I couldn't find a specific note in the documentation so far. Can someone help me on that?

Answer 1

According to the suggestion by #DanielVérité I have implemented a plpgsql function that does a string replace with binary strings of type bytea.
In the implementation I only used functions from the binary strings section, so I think it should be safe to use.
Here's my code:
CREATE OR REPLACE FUNCTION
replace_binary(input_str bytea, pattern bytea, replacement bytea)
RETURNS bytea
AS $$
DECLARE
buf bytea;
pos integer;
BEGIN
buf := '';
-- validate input
IF coalesce(length(input_str), 0) = 0 OR coalesce(length(pattern), 0) = 0
THEN
RETURN input_str;
END IF;
replacement := coalesce(replacement, '');
LOOP
-- find position of pattern in input
pos := position(pattern in input_str);
IF pos = 0 THEN
-- not found: append remaining input to buffer and return
buf := buf || substring(input_str from 1);
RETURN buf;
ELSE
-- found: append substring before pattern to buffer
buf := buf || substring(input_str from 1 for pos - 1);
-- append replacement
buf := buf || replacement;
-- go on with substring of input
input_str := substring(input_str from pos + length(pattern));
END IF;
END LOOP;
END;
$$ LANGUAGE plpgsql
IMMUTABLE;
As for my test cases it works quite well:
with input(buf, pattern, replacement) as (values
('tt'::bytea, 't'::bytea, 'ttt'::bytea),
('test'::bytea, 't'::bytea, 'ttt'::bytea),
('abcdefg'::bytea, 't'::bytea, 'ttt'::bytea),
('\000\015Hello 0orld\000\015Hello 0orld'::bytea, '0'::bytea, '1'::bytea))
select encode(replace_binary(buf, pattern, replacement), 'escape') from input;
outputs as expected:
encode
------------------------------------
tttttt
tttesttt
abcdefg
\000\rHello 1orld\000\rHello 1orld
(4 rows)

Answer 2

The problem with casting to text and back to bytea is that it wouldn't work if the replacement strings involved quoted bytes in strings. Let's see with an example.
(I'm setting bytea_output to hex to better see the text, otherwise it's all hex numbers)
Initial query:
with input(x) as (values (('\000\015Hello World\000\015Hello World'::bytea)))
select replace(x::text, 'World', 'Jenny')::bytea from input;
The result is fine:
replace
----------------------------------------
\000\015Hello Jenny\000\015Hello Jenny
(1 row)
But if trying with a modified version that wants to replace the character 0 by 1
with input(x) as (values (('\000\015Hello 0orld\000\015Hello 0orld'::bytea)))
select replace(x::text, '0', '1')::bytea from input;
The result is:
replace
----------------------------------------
IMHello 1orldIMHello 1orld
whereas the desired result would be: \000\015Hello 1orld\000\015Hello 1orld.
This happens because the intermediate representation \000\015 gets replaced by \111\115