Re: PCRE: references to captures in replace string

A last note about doing this with R: I changed sub(p,r,s) into sub(p,r,s,perl=TRUE,useBytes=TRUE) And my test on 11M strings went from45s to 10s On 4 May 2016 at 10:57, Roberto Cornacchia wrote:

Another option is to use R. This is available in Jul2015 branch, and is very easy (https://www.monetdb.org/content/embedded-r-monetdb)

CREATE FUNCTION r_replace(s STRING, p STRING, r STRING) RETURNS STRING LANGUAGE R { sub(p,r,s) };

sql>select r_replace('Apple Pear', '^([a-zA-Z]+)\\s*([a-zA-Z]+)$', '\\2 \\1'); +------------------------+ | r_replace_single_value | +========================+ | Pear Apple | +------------------------+

It works out of the box on columns:

select r_replace(value, '^.*(\d).*$','\\1%\\2') from obj_string limit 100;

And luckily without loops:

X_10:bat[:oid,:str] := batrapi.eval(nil,"{\n sub(p,r,s)\n};",X_9,A0,A1);

Still not very fast - I've just tried on 11M strings, 45s - probably because strings are fully copied, but for now this seems the easiest workaround.

It would be great to have it working on pcre though. Roberto

On 4 May 2016 at 10:31, Roberto Cornacchia wrote:

...

An interesting workaround for now could be to implement a simple UDF in python, to let python perform the regex processing: https://www.monetdb.org/blog/embedded-pythonnumpy-monetdb

Unfortunately this is for now only available in a separate branch which is kept in sync with default. For me this is no option because I need to rely on a stable MonetDB version. But perhaps for other it is an option.

On 26 April 2016 at 16:57, imad hajj chahine

...

wrote:

...

I still find this to be an interested feature to add, as its very useful in string date manipulation.

On Tue, Apr 26, 2016 at 5:28 PM, Roberto Cornacchia < roberto.cornacchia@gmail.com> wrote:

...

I found this thread from October 2015 where Sjoerd concludes that back references replacement is not implemented in the MonetDB wrapper: https://www.monetdb.org/pipermail/users-list/2015-October/008492.html

Probably that is still the case?

On 20 April 2016 at 16:35, Roberto Cornacchia < roberto.cornacchia@gmail.com> wrote:

...

After declaring a wrapper for pcre.replace:

sql>create function pcre_replace(s string, pattern string, repl string, flags string) returns string external name pcre."replace"; operation successful (6.187ms)

I cannot get replacements with references to captures work (but the code in pcre.c seems to support it):

sql>select pcre_replace('Apple Pear', '^([a-zA-Z]+)\\s*([a-zA-Z]+)$', '\\1 \\2', ''); +---------------------------+ | pcre_replace_single_value | +===========================+ | \1 \2 | +---------------------------+ 1 tuple (0.309ms)

The pattern is matched, but the back references are replaced literally. Am I not getting the syntax correct? From http://www.pcre.org/original/doc/html/pcresyntax.html I see I could use any of \1, \g1, \g{1] for back references, but none seem to work.

Roberto

_______________________________________________ users-list mailing list users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list

_______________________________________________ users-list mailing list users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list