Hey, Thanks for the info. I was trying to implement group_concat Please refer my reply to Fabian's comment.

Regards, Tapomay.

________________________________ From: Stefan Manegold Stefan.Manegold@cwi.nl To: Communication channel for developers of the MonetDB suite. developers-list@monetdb.org Sent: Wednesday, December 12, 2012 9:14 PM Subject: Re: MX file expansion

Please ignore my comment; I wrongly read "for" instead of "from". Sorry.

Fabian's comment stands:

Mx is deprecated and should (must) no longer be used (surely if you're not familiar with it). (New) source should (must) be in plain .h, .c, etc. files, not in .mx files.

Stefan

----- Original Message -----

Like any other source file: use you favorite text editor.

Stefan

----- Original Message -----

...

How do I generate the sources from a new mx file.

Thanks and Regards, Tapomay _______________________________________________ developers-list mailing list developers-list@monetdb.org http://mail.monetdb.org/mailman/listinfo/developers-list

---

developers-list mailing list developers-list@monetdb.org http://mail.monetdb.org/mailman/listinfo/developers-list

_______________________________________________ developers-list mailing list developers-list@monetdb.org http://mail.monetdb.org/mailman/listinfo/developers-list

_______________________________________________ developers-list mailing list developers-list@monetdb.org http://mail.monetdb.org/mailman/listinfo/developers-list

Hi, Thanks for the guidance. Justification: I am trying to add group_concat support. (http://www.mysqlperformanceblog.com/2006/09/04/group_concat-useful-group-by-...) Tried to follow the string reverse example using UDF. But got stuck while trying to figure out the semantics of accessing the group-related BATs returned by "group.done" MAL call(ref: explain <group by query>). This took me to the aggr module implementation in /monetdb5/modules/kernel. Did some reverse engineering to figure out that the 15K lines of code in the C sources were not hand-written but generated from the mx :).

Also the aggregation algorithms handle a lot of bat access cases (like head or tail bats being dense etc.) and some lookup cases like void/scan/merge/hash lookup etc. I have no knowledge about their internals, but I didn't want to end up with an incomplete implementation.

As I am completely new to the semantics and policies of BAT access I decided to refer aggr_bge_avg and came up with the attached mx for reference. Also I am still planning to define it as a UDF and internally make a call to the generated function in kernel module. Not sure if that's possible just by including the generated header in the UDF source.

If you still discourage this approach,  Is it fine if I ignore the above discussed cases and go for a simpler implementation(what should I refer for the same?). One solution (simpler) could be to extract the most common case from the generated C sources of average function.

Attached the mx for reference. Its a two-pass approach where in first pass we calculate the string lengths required per group and in second pass we allocate memory for the destination and strcat the strings into destination. Please note that the attached file may still be logically incomplete but the basic changes are in place. Need to test before finalizing. Thanks again and Regards, Tapomay.

________________________________ From: Fabian Groffen fabian@monetdb.org To: Tapomay Dey tapomay@yahoo.com Cc: "developers-list@monetdb.org" developers-list@monetdb.org Sent: Wednesday, December 12, 2012 8:54 PM Subject: Re: MX file expansion

On 12-12-2012 07:18:10 -0800, Tapomay Dey wrote:

How do I generate the sources from a new mx file.

We strongly discourage people adding new .mx files.  Just use normal .c, .h and .mal files.