Hi David & Florian, please find two quick examples/answers below. Best, Stefan ----- On Nov 6, 2017, at 10:11 PM, David Broneske david.broneske@ovgu.de wrote:

Dear developers of MonetDB,

I hope you are the right recipients of this mail that can help us in getting some more insights into MonetDB’s Code. Currently, me and my student (Florian) want to integrate our index structure Elf ( www.elf.ovgu.de ) into MonetDB as a proof of concept and to show its competitiveness to fast main-memory scans. However, we’ve got a hard time to understand some part of the code — we hope that you could help us with this. While the kernel (gdk) is well documented, we hardly find any documentation that may help us in answering the following questions:

* Which function takes care of filling the BATs when inserting into a table (i.e., sql_table/sql_rel)?

sql>create table t (a int, b int); operation successful (8.722ms) sql>explain insert into t values (1,2); +-------------------------------------------------------------------------------------------------+ | mal | +=================================================================================================+ | function user.s6_1():void; | | X_3:void := querylog.define("explain insert into t values (1,2);", "default_pipe", 12:int); | | X_6 := sql.mvc(); | | X_12 := sql.append(X_6, "sys", "t", "a", 1:int); | | X_17 := sql.append(X_12, "sys", "t", "b", 2:int); | | sql.affectedRows(X_17, 1:lng); | | end user.s6_1; | | # optimizer.mitosis() | | # optimizer.dataflow() | | #inline actions= 0 time=1 usec | | #remap actions= 0 time=3 usec | | #costmodel actions= 1 time=2 usec | | #coercion actions= 1 time=5 usec | | #evaluate actions= 4 time=60 usec | | #emptybind actions= 0 time=0 usec | | #pushselect actions= 0 time=7 usec | | #aliases actions= 4 time=8 usec | | #mergetable actions= 0 time=35 usec | | #deadcode actions= 0 time=7 usec | | #aliases actions= 0 time=0 usec | | #constants actions= 2 time=6 usec | | #commonTerms actions= 0 time=5 usec | | #projectionpath actions= 0 time=4 usec | | #deadcode actions= 0 time=4 usec | | #reorder actions= 1 time=26 usec | | #matpack actions= 0 time=1 usec | | #multiplex actions= 0 time=2 usec | | #profiler actions=1 time=1 usec | | #candidates actions=1 time=1 usec | | #garbagecollector actions= 1 time=64 usec | | #total actions=23 time=397 usec | +-------------------------------------------------------------------------------------------------+ 31 tuples (1.515ms)

* COPY INTO creates a MAL-procedure copy_into, which is preparing the copy, but does not finally execute the copy. Which function is executed for this?

sql>explain copy 1 records into t from stdin; +--------------------------------------------------------------------------------------------------------------------------------------------+ | mal | +============================================================================================================================================+ | function user.s10_1():void; | | X_1:void := querylog.define("explain copy 1 records into t from stdin;", "default_pipe", 20:int); | | X_4 := sql.mvc(); | | (X_24:bat[:int], X_25:bat[:int]) := sql.copy_from(nil:ptr, "|", "\\n", nil:str, "null", nil:str, 1:lng, 0:lng, 0:int, 0:int, nil:str); | | barrier X_80 := language.dataflow(); | | X_27 := sql.append(X_4, "sys", "t", "a", X_24); | | X_32 := sql.append(X_27, "sys", "t", "b", X_25); | | X_34 := aggr.count(X_25); | | exit X_80; | | sql.affectedRows(X_32, X_34); | | end user.s10_1; | | #inline actions= 0 time=1 usec | | #remap actions= 0 time=3 usec | | #costmodel actions= 1 time=2 usec | | #coercion actions=11 time=13 usec | | #evaluate actions= 0 time=4 usec | | #emptybind actions= 0 time=0 usec | | #pushselect actions= 0 time=6 usec | | #aliases actions= 0 time=0 usec | | #mergetable actions= 0 time=37 usec | | #deadcode actions= 0 time=7 usec | | #aliases actions= 0 time=0 usec | | #constants actions= 0 time=8 usec | | #commonTerms actions= 0 time=6 usec | | #projectionpath actions= 0 time=3 usec | | #deadcode actions= 0 time=6 usec | | #reorder actions= 1 time=31 usec | | #matpack actions= 0 time=1 usec | | #dataflow actions= 1 time=23 usec | | #multiplex actions= 0 time=3 usec | | #profiler actions=1 time=1 usec | | #candidates actions=1 time=1 usec | | #garbagecollector actions= 1 time=80 usec | | #total actions=25 time=404 usec | +--------------------------------------------------------------------------------------------------------------------------------------------+ 34 tuples (1.455ms) I.e., in both cases MAL command sql.append() --- respectively it's C implementation --- takes care of inserting the data into the final BATs. Does that answer your questions?

* Regarding ChangeSets: what is dset and nelm used for?

I cannot answer this one, as I don't even know what/where "ChangeSets", "dset", and "nelm" are ... Best, Stefan

Any pointer or explanation for these questions would be great!

Regards, David and Florian M.Sc. David Broneske Otto-von-Guericke-Universität Magdeburg Department of Technical and Business Information Systems Databases and Software Engineering

P.O. Box 4120, D-39016 Magdeburg, Germany Office: Building 29, Room112 Tel: +49 391 67-52994 Fax: +49 391 67-12020

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

-- | Stefan.Manegold@CWI.nl | DB Architectures (DA) | | www.CWI.nl/~manegold/ | Science Park 123 (L321) | | +31 (0)20 592-4212 | 1098 XG Amsterdam (NL) |