PL/SQL optimize searching a date in varchar - regex

Question

I have a table, that contains date field (let it be date s_date) and description field (varchar2(n) desc). What I need is to write a script (or a single query, if possible), that will parse the desc field and if it contains a valid oracle date, then it will cut this date and update the s_date, if it is null.
But there are one more condition - there are must be exactly one occurence of a date in the desc. If there are 0 or >1 - nothing should be updated.
By the time I came up with this pretty ugly solution using regular expressions:
----------------------------------------------
create or replace function to_date_single( p_date_str in varchar2 )
return date
is
l_date date;
pRegEx varchar(150);
pResStr varchar(150);
begin
pRegEx := '((0[1-9]|[12][0-9]|3[01])[.](0[1-9]|1[012])[.](19|20)\d\d)((.|\n|\t|\s)*((0[1-9]|[12][0-9]|3[01])[.](0[1-9]|1[012])[.](19|20)\d\d))?';
pResStr := regexp_substr(p_date_str, pRegEx);
if not (length(pResStr) = 10)
then return null;
end if;
l_date := to_date(pResStr, 'dd.mm.yyyy');
return l_date;
exception
when others then return null;
end to_date_single;
----------------------------------------------
update myTable t
set t.s_date = to_date_single(t.desc)
where t.s_date is null;
----------------------------------------------
But it's working extremely slow (more than a second for each record and i need to update about 30000 records). Is it possible to optimize the function somehow? Maybe it is the way to do the thing without regexp? Any other ideas?
Any advice is appreciated :)
EDIT:
OK, maybe it'll be useful for someone. The following regular expression performs check for valid date (DD.MM.YYYY) taking into account the number of days in a month, including the check for leap year:
(((0[1-9]|[12]\d|3[01])\.(0[13578]|1[02])\.((19|[2-9]\d)\d{2}))|((0[1-9]|[12]\d|30)\.(0[13456789]|1[012])\.((19|[2-9]\d)\d{2}))|((0[1-9]|1\d|2[0-8])\.02\.((19|[2-9]\d)\d{2}))|(29\.02\.((1[6-9]|[2-9]\d)(0[48]|[2468][048]|[13579][26])|((16|[2468][048]|[3579][26])00))))
I used it with the query, suggested by #David (see accepted answer), but I've tried select instead of update (so it's 1 regexp less per row, because we don't do regexp_substr) just for "benchmarking" purpose.
Numbers probably won't tell much here, cause it all depends on hardware, software and specific DB design, but it took about 2 minutes to select 36K records for me. Update will be slower, but I think It'll still be a reasonable time.

Answer 1

I would refactor it along the lines of a single update query.
Use two regexp_instr() calls in the where clause to find rows for which a first occurrence of the match occurs and a second occurrence does not, and regexp_substr() to pull the matching characters for the update.
update my_table
set my_date = to_date(regexp_subtr(desc,...),...)
where regexp_instr(desc,pattern,1,1) > 0 and
regexp_instr(desc,pattern,1,2) = 0
You might get even better performance with:
update my_table
set my_date = to_date(regexp_subtr(desc,...),...)
where case regexp_instr(desc,pattern,1,1)
when 0 then 'N'
else case regexp_instr(desc,pattern,1,2)
when 0 then 'Y'
else 'N'
end
end = 'Y'
... as it only evaluates the second regexp if the first is non-zero. The first query might also do that but the optimiser might choose to evaluate the second predicate first because it is an equality condition, under the assumption that it's more selective.
Or reordering the Case expression might be better -- it's a trade-off that's difficult to judge and probably very dependent on the data.

Answer 2

I think there's no way to improve this task. Actually, in order to achieve what you want it should get even slower.
Your regular expression matches text like 31.02.2013, 31.04.2013 outside the range of the month. If you put year in the game,
it gets even worse. 29.02.2012 is valid, but 29.02.2013 is not.
That's why you have to test if the result is a valid date.
Since there isn't a full regular expression for that, you would have to do it by PLSQL really.
In your to_date_single function you return null when a invalid date is found.
But that doesn't mean there won't be other valid dates forward on the text.
So you have to keep trying until you either find two valid dates or hit the end of the text:
create or replace function fn_to_date(p_date_str in varchar2) return date is
l_date date;
pRegEx varchar(150);
pResStr varchar(150);
vn_findings number;
vn_loop number;
begin
vn_findings := 0;
vn_loop := 1;
pRegEx := '((0[1-9]|[12][0-9]|3[01])[.](0[1-9]|1[012])[.](19|20)\d\d)';
loop
pResStr := regexp_substr(p_date_str, pRegEx, 1, vn_loop);
if pResStr is null then exit; end if;
begin
l_date := to_date(pResStr, 'dd.mm.yyyy');
vn_findings := vn_findings + 1;
-- your crazy requirement :)
if vn_findings = 2 then
return null;
end if;
exception when others then
null;
end;
-- you have to keep trying :)
vn_loop := vn_loop + 1;
end loop;
return l_date;
end;
Some tests:
select fn_to_date('xxxx29.02.2012xxxxx') c1 --ok
, fn_to_date('xxxx29.02.2012xxx29.02.2013xxx') c2 --ok, 2nd is invalid
, fn_to_date('xxxx29.02.2012xxx29.02.2016xxx') c2 --null, both are valid
from dual
As you are going to have to do try and error anyway one idea would be to use a simpler regular expression.
Something like \d\d[.]\d\d[.]\d\d\d\d would suffice. That would depend on your data, of course.
Using #David's idea you could filter the ammount of rows to apply your to_date_single function (because it's slow),
but regular expressions alone won't do what you want:
update my_table
set my_date = fn_to_date( )
where regexp_instr(desc,patern,1,1) > 0