I am facing the following error:

ERROR:  functions in index expression must be marked IMMUTABLE

When trying to create index like this:

CREATE INDEX full_phone_number ON orders_clientphone (concat(area_code, phone));

On the other hand, when using alternative syntax for concatenation:

CREATE INDEX full_phone_number ON orders_clientphone ((area_code || phone));

Postgres is pretty much ok with that.
Both columns are defined as character varying(256).


The deciding factor for Postgres is that the function concat() is defined stable and not immutable in the system catalog pg_proc:

SELECT proname, provolatile, proargtypes, proargtypes[0]::regtype AS argtype, prosrc
FROM pg_proc
WHERE proname = 'concat';

proname | provolatile | proargtypes| argtype | prosrc
concat  | s           | 2276       | "any"   | text_concat

The manual on pg_proc.provolatile:

provolatile tells whether the function's result depends only on its input arguments, or is affected by outside factors. It is i for "immutable" functions, which always deliver the same result for the same inputs. It is s for "stable" functions, whose results (for fixed inputs) do not change within a scan.

I also added the argument types of the function ("any") to connect to the answers of @dezso and @jjanes, which deliver the rationale behind the decision to make this function only stable. And the name of the internal function (text_concat).

Here is a related question to illustrate why immutability of index expressions is a sine qua non condition:

As for using the operator ||:

    SELECT o.oprname, o.oprleft::regtype, o.oprright::regtype, o.oprcode, p.provolatile
    FROM   pg_operator o
    JOIN   pg_proc     p ON p.oid = o.oprcode
    WHERE  oprname = '||';

 oprname |   oprleft   |  oprright   |     oprcode     | provolatile
 ||      | anyarray    | anyelement  | array_append    | i
 ||      | anyelement  | anyarray    | array_prepend   | i
 ||      | anyarray    | anyarray    | array_cat       | i
 ||      | text        | text        | textcat         | i
 ||      | bit varying | bit varying | bitcat          | i
 ||      | bytea       | bytea       | byteacat        | i
 ||      | text        | anynonarray | textanycat      | s
 ||      | anynonarray | text        | anytextcat      | s
 ||      | tsvector    | tsvector    | tsvector_concat | i
 ||      | tsquery     | tsquery     | tsquery_or      | i
 ||      | jsonb       | jsonb       | jsonb_concat    | i

The function to use internally depends on actual data type of operands. The definition of an operator includes the operands' data types in Postgres. All of the functions are different and also different from text_concat above. || is just stable as well, when one of the operators is anynonarray. Things are not so trivial behind the curtains.

character varying(256) (like any varchar variant) is binary-coercible to text so function type resolution defaults to text.

  • 2
    So, what's the solution if one needs to index the result of text concatenation?
    – Alex R
    Nov 26 '18 at 3:00
  • 1
    @AlexR: Depends on involved data types and NULL values. For 2 non-null text columns, just use || ... Nov 26 '18 at 5:15
  • The solution, use || (as in the ops question), or wrap the call to concat into an function declared immutable. Nov 6 '20 at 12:17

I believe concat, defined in the documentation as follows:

FUNCTION                                  RETURN TYPE  DESCRIPTION
concat(str "any" [, str "any" [, ...] ])  text         Concatenate the text repre-
                                                       sentations of all the arguments. 
                                                       NULL arguments are ignored.

is not immutable because the text representation might depend on database settings, like that of a date or timestamp.


To fill out dezso's answer, here is an example of the same input yielding different outputs depending on database state:

select concat(1.0000000000003::float8,56);


set extra_float_digits TO 3;

select concat(1.0000000000003::float8,56);


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.