Convert hex in text representation to decimal number

pg-bignum

Internally, pg-bignum uses the SSL library for big numbers. This method has none of the drawbacks mentioned in the other answers with numeric. Nor is it slowed down by plpgsql. It's fast and it works with a number of any size. Test case taken from Erwin's answer for comparison,

CREATE EXTENSION bignum;

SELECT hex, bn_in_hex(hex::cstring) 
FROM   (
   VALUES ('ff'::text)
        , ('7fffffff')
        , ('80000000')
        , ('deadbeef')
        , ('7fffffffffffffff')
        , ('8000000000000000')
        , ('ffffffffffffffff')
        , ('ffffffffffffffff123')
   ) t(hex);

         hex         |        bn_in_hex        
---------------------+-------------------------
 ff                  | 255
 7fffffff            | 2147483647
 80000000            | 2147483648
 deadbeef            | 3735928559
 7fffffffffffffff    | 9223372036854775807
 8000000000000000    | 9223372036854775808
 ffffffffffffffff    | 18446744073709551615
 ffffffffffffffff123 | 75557863725914323415331
(8 rows)

You can get the type to numeric using bn_in_hex('deadbeef')::text::numeric.


If anybody else is stuck with PG8.2, here is another way to do it.

bigint version:

create or replace function hex_to_bigint(hexval text) returns bigint as $$
select
  (get_byte(x,0)::int8<<(7*8)) |
  (get_byte(x,1)::int8<<(6*8)) |
  (get_byte(x,2)::int8<<(5*8)) |
  (get_byte(x,3)::int8<<(4*8)) |
  (get_byte(x,4)::int8<<(3*8)) |
  (get_byte(x,5)::int8<<(2*8)) |
  (get_byte(x,6)::int8<<(1*8)) |
  (get_byte(x,7)::int8)
from (
  select decode(lpad($1, 16, '0'), 'hex') as x
) as a;
$$
language sql strict immutable;

int version:

create or replace function hex_to_int(hexval text) returns int as $$
select
  (get_byte(x,0)::int<<(3*8)) |
  (get_byte(x,1)::int<<(2*8)) |
  (get_byte(x,2)::int<<(1*8)) |
  (get_byte(x,3)::int)
from (
  select decode(lpad($1, 8, '0'), 'hex') as x
) as a;
$$
language sql strict immutable;

You have two immediate problems:

  1. to_number doesn't understand hexadecimal.
  2. X doesn't have any meaning in a to_number format string and anything without a meaning apparently means "skip a character".

I don't have an authoritative justification for (2), just empirical evidence:

=> SELECT to_number('123', 'X999');
 to_number 
-----------
        23
(1 row)

=> SELECT to_number('123', 'XX999');
 to_number 
-----------
         3

The documentation mentions how double quoted patterns are supposed to behave:

In to_date, to_number, and to_timestamp, double-quoted strings skip the number of input characters contained in the string, e.g. "XX" skips two input characters.

but the behavior of non-quoted characters that are not formatting characters appears to be unspecified.

In any case, to_number isn't the right tool for converting hex to numbers, you want to say something like this:

select x'deadbeef'::int;

so perhaps this function will work better for you:

CREATE OR REPLACE FUNCTION hex_to_int(hexval varchar) RETURNS integer AS $$
DECLARE
    result  int;
BEGIN
    EXECUTE 'SELECT x' || quote_literal(hexval) || '::int' INTO result;
    RETURN result;
END;
$$ LANGUAGE plpgsql IMMUTABLE STRICT;

Then:

=> select hex_to_int('DEADBEEF');
 hex_to_int 
------------
 -559038737 **
(1 row)

** To avoid negative numbers like this from integer overflow error, use bigint instead of int to accommodate larger hex numbers (like IP addresses).


Ways without dynamic SQL

There is no cast from hex numbers in text representation to a numeric type, but we can use bit(n) as waypoint. There are undocumented casts from bit strings (bit(n)) to integer types (int2, int4, int8) - the internal representation is binary compatible. Quoting Tom Lane:

This is relying on some undocumented behavior of the bit-type input converter, but I see no reason to expect that would break. A possibly bigger issue is that it requires PG >= 8.3 since there wasn't a text to bit cast before that.

integer for max. 8 hex digits

Up to 8 hex digits can be converted to bit(32) and then coerced to integer (standard 4-byte integer):

SELECT ('x' || lpad(hex, 8, '0'))::bit(32)::int AS int_val
FROM  (
   VALUES
      ('1'::text)
    , ('f')
    , ('100')
    , ('7fffffff')
    , ('80000000')     -- overflow into negative number
    , ('deadbeef')
    , ('ffffffff')
    , ('ffffffff123')  -- too long
   ) AS t(hex);
   int_val
------------
          1
         15
        256
 2147483647
-2147483648
 -559038737
         -1

Postgres uses a signed integer type, so hex numbers above '7fffffff' overflow into negative integer numbers. This is still a valid, unique representation but the meaning is different. If that matters, switch to bigint; see below.

For more than 8 hex digits the least significant characters (excess to the right) get truncated.

4 bits in a bit string encode 1 hex digit. Hex numbers of known length can be cast to the respective bit(n) directly. Alternatively, pad hex numbers of unknown length with leading zeros (0) as demonstrated and cast to bit(32). Example with 7 hex digits and int or 8 digits and bigint:

SELECT ('x'|| 'deafbee')::bit(28)::int
     , ('x'|| 'deadbeef')::bit(32)::bigint;
  int4     | int8
-----------+------------
 233503726 | 3735928559

bigint for max. 16 hex digits

Up to 16 hex digits can be converted to bit(64) and then coerced to bigint (int8, 8-byte integer) - overflowing into negative numbers in the upper half again:

SELECT ('x' || lpad(hex, 16, '0'))::bit(64)::bigint AS int8_val
FROM  (
   VALUES
      ('ff'::text)
    , ('7fffffff')
    , ('80000000')
    , ('deadbeef')
    , ('7fffffffffffffff')
    , ('8000000000000000')     -- overflow into negative number
    , ('ffffffffffffffff')
    , ('ffffffffffffffff123')  -- too long
   ) t(hex);
       int8_val
---------------------
                 255
          2147483647
          2147483648
          3735928559
 9223372036854775807
-9223372036854775808
                  -1
                  -1

uuid for max. 32 hex digits

The Postgres uuid data type is not a numeric type. But it's the most efficient type in standard Postgres to store up to 32 hex digits, only occupying 16 bytes of storage. There is a direct cast from text to uuid (no need for bit(n) as waypoint), but exactly 32 hex digits are required.

SELECT lpad(hex, 32, '0')::uuid AS uuid_val
FROM  (
   VALUES ('ff'::text)
        , ('deadbeef')
        , ('ffffffffffffffff')
        , ('ffffffffffffffffffffffffffffffff')
        , ('ffffffffffffffffffffffffffffffff123') -- too long
   ) t(hex);
              uuid_val
--------------------------------------
 00000000-0000-0000-0000-0000000000ff
 00000000-0000-0000-0000-0000deadbeef
 00000000-0000-0000-ffff-ffffffffffff
 ffffffff-ffff-ffff-ffff-ffffffffffff
 ffffffff-ffff-ffff-ffff-ffffffffffff

As you can see, standard output is a string of hex digits with typical separators for UUID.

md5 hash

This is particularly useful to store md5 hashes:

SELECT md5('Store hash for long string, maybe for index?')::uuid AS md5_hash;
           md5_hash
--------------------------------------
 02e10e94-e895-616e-8e23-bb7f8025da42

See:

  • What is the optimal data type for an MD5 field?