Thread: Function array_agg(array)

Seems dangerous as heck; certainly it would have side-effects far more
wide-ranging than just making this particular function work.

A safer answer is to split array_agg into two functions,
        array_agg(anynonarray) -> anyarray
        array_agg(anyarray) -> anyarray

I rather imagine you should do that anyway, because I really doubt
that this hack is operating quite as intended.  I suspect you are
producing arrays containing arrays as elements, not true 2-D arrays.
That's not a direction we want to go in I think; certainly there are
no other operations that produce such things.

Currently, it cannot handle NULL arrays:
backend> select array_agg(a) from (values(null::int[])) a(a);
     1: array_agg    (typeid = 1007, len = -1, typmod = -1, byval = f)
    ----
ERROR:  cannot aggregate null arrays

So, is there any idea how we will handle NULL and empty array in array_agg(anyarray)? 

I propose we just reject those input because the output will make no sense:

- array_agg(NULL::int[]) --> the result will be indistinguished from array_agg of NULL ints.

- array_agg('{}'::int[]) --> how we determine the dimension of the result? is it 0? Or the result will be just an empty array {} ?

So, is there any idea how we will handle NULL and empty array in array_agg(anyarray)? 

I propose we just reject those input because the output will make no sense:

- array_agg(NULL::int[]) --> the result will be indistinguished from array_agg of NULL ints.

- array_agg('{}'::int[]) --> how we determine the dimension of the result? is it 0? Or the result will be just an empty array {} ?

This updated patch rejects NULL and {} arrays as noted above.

 

Regards,

--
Ali Akbar

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

I agree with your proposal. I have a few comments to design:

1. patch doesn't hold documentation and regress tests, please append it.

2. this functionality (multidimensional aggregation) can be interesting more times, so maybe some interface like array builder should be preferred.

3. array_agg was consistent with array(subselect), so it should be fixed too

postgres=# select array_agg(a) from test;
       array_agg      
-----------------------
 {{1,2,3,4},{1,2,3,4}}
(1 row)

postgres=# select array(select a from test);
ERROR:  could not find array type for data type integer[]

4. why you use a magic constant (64) there?

+         astate->abytes = 64 * (ndatabytes == 0 ? 1 : ndatabytes);
+         astate->aitems = 64 * nitems;

+             astate->nullbitmap = (bits8 *)
+                 repalloc(astate->nullbitmap, (astate->aitems + 7) / 8);

Thanks for the review

2014-10-22 20:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

I agree with your proposal. I have a few comments to design:

1. patch doesn't hold documentation and regress tests, please append it.

OK, i'll add the documentation and regression test 

 

2. this functionality (multidimensional aggregation) can be interesting more times, so maybe some interface like array builder should be preferred.

We already have accumArrayResult and makeArrayResult/makeMdArrayResult, haven't we?

Actually array_agg(anyarray) can be implemented by using accumArrayResult and makeMdArrayResult, but in the process we will deconstruct the array data and null bit-flags into ArrayBuildState->dvalues and dnulls. And we will reconstruct it through makeMdArrayResult. I think this way it's not efficient, so i decided to reimplement it and memcpy the array data and null flags as-is.

In other places, i think it's clearer if we just use accumArrayResult and makeArrayResult/makeMdArrayResult as API for building (multidimensional) arrays.

 

3. array_agg was consistent with array(subselect), so it should be fixed too

postgres=# select array_agg(a) from test;
       array_agg      
-----------------------
 {{1,2,3,4},{1,2,3,4}}
(1 row)

postgres=# select array(select a from test);
ERROR:  could not find array type for data type integer[]

I'm pretty new in postgresql development. Can you point out where is array(subselect) implemented?

 

4. why you use a magic constant (64) there?

+         astate->abytes = 64 * (ndatabytes == 0 ? 1 : ndatabytes);
+         astate->aitems = 64 * nitems;

+             astate->nullbitmap = (bits8 *)
+                 repalloc(astate->nullbitmap, (astate->aitems + 7) / 8);

just follow the arbitrary size choosen in accumArrayResult (arrayfuncs.c):

astate->alen = 64; /* arbitrary starting array size */

it can be any number not too small and too big. Too small, and we will realloc shortly. Too big, we will end up wasting memory.

Regards,

--
Ali Akbar

2014-10-22 16:58 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

Thanks for the review

2014-10-22 20:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

I agree with your proposal. I have a few comments to design:

1. patch doesn't hold documentation and regress tests, please append it.

OK, i'll add the documentation and regression test 

 

2. this functionality (multidimensional aggregation) can be interesting more times, so maybe some interface like array builder should be preferred.

We already have accumArrayResult and makeArrayResult/makeMdArrayResult, haven't we?

Actually array_agg(anyarray) can be implemented by using accumArrayResult and makeMdArrayResult, but in the process we will deconstruct the array data and null bit-flags into ArrayBuildState->dvalues and dnulls. And we will reconstruct it through makeMdArrayResult. I think this way it's not efficient, so i decided to reimplement it and memcpy the array data and null flags as-is.

aha, so isn't better to fix a performance for accumArrayResult ?

In other places, i think it's clearer if we just use accumArrayResult and makeArrayResult/makeMdArrayResult as API for building (multidimensional) arrays.

 

3. array_agg was consistent with array(subselect), so it should be fixed too

postgres=# select array_agg(a) from test;
       array_agg      
-----------------------
 {{1,2,3,4},{1,2,3,4}}
(1 row)

postgres=# select array(select a from test);
ERROR:  could not find array type for data type integer[]

I'm pretty new in postgresql development. Can you point out where is array(subselect) implemented?

where you can start?

postgres=# explain select array(select a from test);
ERROR:  42704: could not find array type for data type integer[]
LOCATION:  exprType, nodeFuncs.c:116

look to code ... your magic keyword is a ARRAY_SUBLINK .. search in postgresql sources this keyword

4. why you use a magic constant (64) there?

+         astate->abytes = 64 * (ndatabytes == 0 ? 1 : ndatabytes);
+         astate->aitems = 64 * nitems;

+             astate->nullbitmap = (bits8 *)
+                 repalloc(astate->nullbitmap, (astate->aitems + 7) / 8);

just follow the arbitrary size choosen in accumArrayResult (arrayfuncs.c):

astate->alen = 64; /* arbitrary starting array size */

it can be any number not too small and too big. Too small, and we will realloc shortly. Too big, we will end up wasting memory.

you can try to alloc 1KB instead as start -- it is used more times in Postgres. Then a overhead is max 1KB per agg call - what is acceptable.

You take this value from accumArrayResult, but it is targeted for shorted scalars - you should to expect so any array will be much larger.

2014-10-22 22:48 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

2014-10-22 16:58 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

Thanks for the review

2014-10-22 20:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

I agree with your proposal. I have a few comments to design:

1. patch doesn't hold documentation and regress tests, please append it.

OK, i'll add the documentation and regression test 

 

2. this functionality (multidimensional aggregation) can be interesting more times, so maybe some interface like array builder should be preferred.

We already have accumArrayResult and makeArrayResult/makeMdArrayResult, haven't we?

Actually array_agg(anyarray) can be implemented by using accumArrayResult and makeMdArrayResult, but in the process we will deconstruct the array data and null bit-flags into ArrayBuildState->dvalues and dnulls. And we will reconstruct it through makeMdArrayResult. I think this way it's not efficient, so i decided to reimplement it and memcpy the array data and null flags as-is.

aha, so isn't better to fix a performance for accumArrayResult ?

Ok, i'll go this route. While reading the code of array(subselect) as you pointed below, i think the easiest way to implement support for both array_agg(anyarray) and array(subselect) is to change accumArrayResult so it operates both with scalar datum(s) and array datums, with performance optimization for the latter.

In other places, i think it's clearer if we just use accumArrayResult and makeArrayResult/makeMdArrayResult as API for building (multidimensional) arrays.

 

3. array_agg was consistent with array(subselect), so it should be fixed too

postgres=# select array_agg(a) from test;
       array_agg      
-----------------------
 {{1,2,3,4},{1,2,3,4}}
(1 row)

postgres=# select array(select a from test);
ERROR:  could not find array type for data type integer[]

I'm pretty new in postgresql development. Can you point out where is array(subselect) implemented?

where you can start?

postgres=# explain select array(select a from test);
ERROR:  42704: could not find array type for data type integer[]
LOCATION:  exprType, nodeFuncs.c:116

look to code ... your magic keyword is a ARRAY_SUBLINK .. search in postgresql sources this keyword

Found it. Thanks.

On a side note in postgres array type for integer[] is still integer[], but in pg_type, integer[] has no array type. I propose we consider to change it so array type of anyarray is itself (not in this patch, of course, because it is a big change). Consider the following code in nodeFuncs.c:109

if (sublink->subLinkType == ARRAY_SUBLINK)

{

type = get_array_type(type);

if (!OidIsValid(type))

ereport(ERROR,

(errcode(ERRCODE_UNDEFINED_OBJECT),

errmsg("could not find array type for data type %s",

format_type_be(exprType((Node *) tent->expr)))));

}

to implement array(subselect) you pointed above, we must specially checks for array types. Quick search on get_array_type returns 10 places.

I'll think more about this later. For this patch, i'll go without changes in pg_type.h.

4. why you use a magic constant (64) there?

+         astate->abytes = 64 * (ndatabytes == 0 ? 1 : ndatabytes);
+         astate->aitems = 64 * nitems;

+             astate->nullbitmap = (bits8 *)
+                 repalloc(astate->nullbitmap, (astate->aitems + 7) / 8);

just follow the arbitrary size choosen in accumArrayResult (arrayfuncs.c):

astate->alen = 64; /* arbitrary starting array size */

it can be any number not too small and too big. Too small, and we will realloc shortly. Too big, we will end up wasting memory.

you can try to alloc 1KB instead as start -- it is used more times in Postgres. Then a overhead is max 1KB per agg call - what is acceptable.

You take this value from accumArrayResult, but it is targeted for shorted scalars - you should to expect so any array will be much larger.

Ok.

Regards,

--

Ali Akbar

2014-10-22 20:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

I agree with your proposal. I have a few comments to design:

1. patch doesn't hold documentation and regress tests, please append it.

2. this functionality (multidimensional aggregation) can be interesting more times, so maybe some interface like array builder should be preferred.

We already have accumArrayResult and makeArrayResult/makeMdArrayResult, haven't we?

Actually array_agg(anyarray) can be implemented by using accumArrayResult and makeMdArrayResult, but in the process we will deconstruct the array data and null bit-flags into ArrayBuildState->dvalues and dnulls. And we will reconstruct it through makeMdArrayResult. I think this way it's not efficient, so i decided to reimplement it and memcpy the array data and null flags as-is.

aha, so isn't better to fix a performance for accumArrayResult ?

Ok, i'll go this route. While reading the code of array(subselect) as you pointed below, i think the easiest way to implement support for both array_agg(anyarray) and array(subselect) is to change accumArrayResult so it operates both with scalar datum(s) and array datums, with performance optimization for the latter.

In other places, i think it's clearer if we just use accumArrayResult and makeArrayResult/makeMdArrayResult as API for building (multidimensional) arrays.

 

3. array_agg was consistent with array(subselect), so it should be fixed too

postgres=# select array_agg(a) from test;
       array_agg      
-----------------------
 {{1,2,3,4},{1,2,3,4}}
(1 row)

postgres=# select array(select a from test);
ERROR:  could not find array type for data type integer[]

I'm pretty new in postgresql development. Can you point out where is array(subselect) implemented?

where you can start?

postgres=# explain select array(select a from test);
ERROR:  42704: could not find array type for data type integer[]
LOCATION:  exprType, nodeFuncs.c:116

look to code ... your magic keyword is a ARRAY_SUBLINK .. search in postgresql sources this keyword

attention: probably we don't would to allow arrays everywhere. 

4. why you use a magic constant (64) there?

+         astate->abytes = 64 * (ndatabytes == 0 ? 1 : ndatabytes);
+         astate->aitems = 64 * nitems;

+             astate->nullbitmap = (bits8 *)
+                 repalloc(astate->nullbitmap, (astate->aitems + 7) / 8);

just follow the arbitrary size choosen in accumArrayResult (arrayfuncs.c):

astate->alen = 64; /* arbitrary starting array size */

it can be any number not too small and too big. Too small, and we will realloc shortly. Too big, we will end up wasting memory.

you can try to alloc 1KB instead as start -- it is used more times in Postgres. Then a overhead is max 1KB per agg call - what is acceptable.

You take this value from accumArrayResult, but it is targeted for shorted scalars - you should to expect so any array will be much larger. 

Hi

it looks well

doc: http://www.postgresql.org/docs/9.4/static/sql-expressions.html#SQL-SYNTAX-ARRAY-CONSTRUCTORS it should be fixed too

Regards

Pavel

Hi

some in last patch is wrong, I cannot to compile it:

arrayfuncs.c: In function ‘accumArrayResult’:
arrayfuncs.c:4603:9: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->alen = 64;  /* arbitrary starting array size */
         ^
arrayfuncs.c:4604:9: error: ‘ArrayBuildState’ has no member named ‘dvalues’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
         ^
arrayfuncs.c:4604:44: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
                                            ^
arrayfuncs.c:4605:9: error: ‘ArrayBuildState’ has no member named ‘dnulls’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
         ^
arrayfuncs.c:4605:42: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
                                          ^
arrayfuncs.c:4606:9: error: ‘ArrayBuildState’ has no member named ‘nelems’
   astate->nelems = 0;
         ^
arrayfuncs.c:4618:13: error: ‘ArrayBuildState’ has no member named ‘nelems’
   if (astate->nelems >= astate->alen)
             ^
arrayfuncs.c:4618:31: error: ‘ArrayBuildState’ has no member named ‘alen’
   if (astate->nelems >= astate->alen)
                               ^
arrayfuncs.c:4620:10: error: ‘ArrayBuildState’ has no member named ‘alen’
    astate->alen *= 2;

2014-10-24 16:26 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi

some in last patch is wrong, I cannot to compile it:

arrayfuncs.c: In function ‘accumArrayResult’:
arrayfuncs.c:4603:9: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->alen = 64;  /* arbitrary starting array size */
         ^
arrayfuncs.c:4604:9: error: ‘ArrayBuildState’ has no member named ‘dvalues’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
         ^
arrayfuncs.c:4604:44: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
                                            ^
arrayfuncs.c:4605:9: error: ‘ArrayBuildState’ has no member named ‘dnulls’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
         ^
arrayfuncs.c:4605:42: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
                                          ^
arrayfuncs.c:4606:9: error: ‘ArrayBuildState’ has no member named ‘nelems’
   astate->nelems = 0;
         ^
arrayfuncs.c:4618:13: error: ‘ArrayBuildState’ has no member named ‘nelems’
   if (astate->nelems >= astate->alen)
             ^
arrayfuncs.c:4618:31: error: ‘ArrayBuildState’ has no member named ‘alen’
   if (astate->nelems >= astate->alen)
                               ^
arrayfuncs.c:4620:10: error: ‘ArrayBuildState’ has no member named ‘alen’
    astate->alen *= 2;

Sorry,  correct patch attached.

This patch is in patience format (git --patience ..). In previous patches, i use context format (git --patience ... | filterdiff --format=context), but it turns out that some modification is lost.

2014-10-24 16:26 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi

some in last patch is wrong, I cannot to compile it:

arrayfuncs.c: In function ‘accumArrayResult’:
arrayfuncs.c:4603:9: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->alen = 64;  /* arbitrary starting array size */
         ^
arrayfuncs.c:4604:9: error: ‘ArrayBuildState’ has no member named ‘dvalues’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
         ^
arrayfuncs.c:4604:44: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
                                            ^
arrayfuncs.c:4605:9: error: ‘ArrayBuildState’ has no member named ‘dnulls’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
         ^
arrayfuncs.c:4605:42: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
                                          ^
arrayfuncs.c:4606:9: error: ‘ArrayBuildState’ has no member named ‘nelems’
   astate->nelems = 0;
         ^
arrayfuncs.c:4618:13: error: ‘ArrayBuildState’ has no member named ‘nelems’
   if (astate->nelems >= astate->alen)
             ^
arrayfuncs.c:4618:31: error: ‘ArrayBuildState’ has no member named ‘alen’
   if (astate->nelems >= astate->alen)
                               ^
arrayfuncs.c:4620:10: error: ‘ArrayBuildState’ has no member named ‘alen’
    astate->alen *= 2;

Sorry,  correct patch attached.

This patch is in patience format (git --patience ..). In previous patches, i use context format (git --patience ... | filterdiff --format=context), but it turns out that some modification is lost.

--
Ali Akbar

2014-10-24 11:43 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

2014-10-24 16:26 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi

some in last patch is wrong, I cannot to compile it:

arrayfuncs.c: In function ‘accumArrayResult’:
arrayfuncs.c:4603:9: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->alen = 64;  /* arbitrary starting array size */
         ^
arrayfuncs.c:4604:9: error: ‘ArrayBuildState’ has no member named ‘dvalues’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
         ^
arrayfuncs.c:4604:44: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dvalues = (Datum *) palloc(astate->alen * sizeof(Datum));
                                            ^
arrayfuncs.c:4605:9: error: ‘ArrayBuildState’ has no member named ‘dnulls’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
         ^
arrayfuncs.c:4605:42: error: ‘ArrayBuildState’ has no member named ‘alen’
   astate->dnulls = (bool *) palloc(astate->alen * sizeof(bool));
                                          ^
arrayfuncs.c:4606:9: error: ‘ArrayBuildState’ has no member named ‘nelems’
   astate->nelems = 0;
         ^
arrayfuncs.c:4618:13: error: ‘ArrayBuildState’ has no member named ‘nelems’
   if (astate->nelems >= astate->alen)
             ^
arrayfuncs.c:4618:31: error: ‘ArrayBuildState’ has no member named ‘alen’
   if (astate->nelems >= astate->alen)
                               ^
arrayfuncs.c:4620:10: error: ‘ArrayBuildState’ has no member named ‘alen’
    astate->alen *= 2;

Sorry,  correct patch attached.

This patch is in patience format (git --patience ..). In previous patches, i use context format (git --patience ... | filterdiff --format=context), but it turns out that some modification is lost.

last version is compileable, but some is still broken

postgres=# select array_agg(array[i, i+1, i-1])
            from generate_series(1,2) a(i);
ERROR:  could not find array type for data type integer[]

but array(subselect) works

postgres=# select array(select a from xx);
       array      
-------------------
 {{1,2,3},{1,2,3}}
(1 row)

Regards

Pavel

 

--
Ali Akbar

Hi Ali 

I checked a code. I am thinking so code organization is not good. accumArrayResult is too long now. makeMdArrayResult will not work, when arrays was joined (it is not consistent now). I don't like a usage of state->is_array_accum in array_userfunc.c -- it is signal of wrong wrapping.

next question: there is function array_append(anyarray, anyelement). Isn't time to define array_append(anyarray, anyarray) now? 

I fixed small issue in regress tests and I enhanced tests for varlena types and null values.

Thanks.

it is about 15% faster than original implementation.

15% faster than array_agg(scalar)? I haven't verify the performance, but because the internal array data and null bitmap is copied as-is, that will be faster.

2014-10-25 1:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi Ali 

I checked a code. I am thinking so code organization is not good. accumArrayResult is too long now. makeMdArrayResult will not work, when arrays was joined (it is not consistent now). I don't like a usage of state->is_array_accum in array_userfunc.c -- it is signal of wrong wrapping.

Yes, i was thinking the same. Attached WIP patch to reorganizate the code. makeMdArrayResult works now, with supplied arguments act as override from default values calculated from ArrayBuildStateArray.

In array_userfunc.c, because we don't want to override ndims, dims and lbs, i copied array_agg_finalfn and only change the call to makeMdArrayResult (we don't uses makeArrayResult because we want to set release to false). Another alternative is to create new makeArrayResult-like function that has parameter bool release.

next question: there is function array_append(anyarray, anyelement). Isn't time to define array_append(anyarray, anyarray) now? 

There is array_cat(anyarray, anyarray):

/*-----------------------------------------------------------------------------

 * array_cat :

 * concatenate two nD arrays to form an nD array, or

 * push an (n-1)D array onto the end of an nD array

 *----------------------------------------------------------------------------

 */

2014-10-25 10:29 GMT+07:00 Ali Akbar <the.apaan@gmail.com>:

I fixed small issue in regress tests and I enhanced tests for varlena types and null values.

Thanks.

it is about 15% faster than original implementation.

15% faster than array_agg(scalar)? I haven't verify the performance, but because the internal array data and null bitmap is copied as-is, that will be faster.

2014-10-25 1:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi Ali 

I checked a code. I am thinking so code organization is not good. accumArrayResult is too long now. makeMdArrayResult will not work, when arrays was joined (it is not consistent now). I don't like a usage of state->is_array_accum in array_userfunc.c -- it is signal of wrong wrapping.

Yes, i was thinking the same. Attached WIP patch to reorganizate the code. makeMdArrayResult works now, with supplied arguments act as override from default values calculated from ArrayBuildStateArray.

In array_userfunc.c, because we don't want to override ndims, dims and lbs, i copied array_agg_finalfn and only change the call to makeMdArrayResult (we don't uses makeArrayResult because we want to set release to false). Another alternative is to create new makeArrayResult-like function that has parameter bool release.

adding makeArrayResult1 (do you have better name for this?), that accepts bool release parameter. array_agg_finalfn becomes more clean, and no duplicate code in array_agg_anyarray_finalfn.

 

next question: there is function array_append(anyarray, anyelement). Isn't time to define array_append(anyarray, anyarray) now? 

There is array_cat(anyarray, anyarray):

/*-----------------------------------------------------------------------------

 * array_cat :

 * concatenate two nD arrays to form an nD array, or

 * push an (n-1)D array onto the end of an nD array

 *----------------------------------------------------------------------------

 */

Regards, 

--
Ali Akbar

2014-10-25 8:19 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

2014-10-25 10:29 GMT+07:00 Ali Akbar <the.apaan@gmail.com>:

I fixed small issue in regress tests and I enhanced tests for varlena types and null values.

Thanks.

it is about 15% faster than original implementation.

15% faster than array_agg(scalar)? I haven't verify the performance, but because the internal array data and null bitmap is copied as-is, that will be faster.

2014-10-25 1:51 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi Ali 

I checked a code. I am thinking so code organization is not good. accumArrayResult is too long now. makeMdArrayResult will not work, when arrays was joined (it is not consistent now). I don't like a usage of state->is_array_accum in array_userfunc.c -- it is signal of wrong wrapping.

Yes, i was thinking the same. Attached WIP patch to reorganizate the code. makeMdArrayResult works now, with supplied arguments act as override from default values calculated from ArrayBuildStateArray.

In array_userfunc.c, because we don't want to override ndims, dims and lbs, i copied array_agg_finalfn and only change the call to makeMdArrayResult (we don't uses makeArrayResult because we want to set release to false). Another alternative is to create new makeArrayResult-like function that has parameter bool release.

adding makeArrayResult1 (do you have better name for this?), that accepts bool release parameter. array_agg_finalfn becomes more clean, and no duplicate code in array_agg_anyarray_finalfn.

 makeArrayResult1 - I have no better name now

I found one next minor detail.

you reuse a array_agg_transfn function. Inside is a message "array_agg_transfn called in non-aggregate context". It is not correct for array_agg_anyarray_transfn

probably specification dim and lbs in array_agg_finalfn is useless, when you expect a natural result. A automatic similar to array_agg_anyarray_finalfn should work there too.

makeMdArrayResultArray and appendArrayDatum  should be marked as "static"

 

next question: there is function array_append(anyarray, anyelement). Isn't time to define array_append(anyarray, anyarray) now? 

There is array_cat(anyarray, anyarray):

/*-----------------------------------------------------------------------------

 * array_cat :

 * concatenate two nD arrays to form an nD array, or

 * push an (n-1)D array onto the end of an nD array

 *----------------------------------------------------------------------------

 */

Regards, 

--
Ali Akbar

 makeArrayResult1 - I have no better name now

I found one next minor detail.

you reuse a array_agg_transfn function. Inside is a message "array_agg_transfn called in non-aggregate context". It is not correct for array_agg_anyarray_transfn

probably specification dim and lbs in array_agg_finalfn is useless, when you expect a natural result. A automatic similar to array_agg_anyarray_finalfn should work there too.

makeMdArrayResultArray and appendArrayDatum  should be marked as "static"

I am thinking so change of makeArrayResult is wrong. It is developed for 1D array. When we expect somewhere ND result now, then we should to use makeMdArrayResult there. So makeArrayResult should to return 1D array in all cases. Else a difference between makeArrayResult and makeMdArrayResult hasn't big change and we have a problem with naming. 

 makeArrayResult1 - I have no better name now

I found one next minor detail.

you reuse a array_agg_transfn function. Inside is a message "array_agg_transfn called in non-aggregate context". It is not correct for array_agg_anyarray_transfn

Fixed.

 

probably specification dim and lbs in array_agg_finalfn is useless, when you expect a natural result. A automatic similar to array_agg_anyarray_finalfn should work there too.

makeMdArrayResultArray and appendArrayDatum  should be marked as "static"

ok, marked all of that as static.

 

I am thinking so change of makeArrayResult is wrong. It is developed for 1D array. When we expect somewhere ND result now, then we should to use makeMdArrayResult there. So makeArrayResult should to return 1D array in all cases. Else a difference between makeArrayResult and makeMdArrayResult hasn't big change and we have a problem with naming. 

I'm thinking it like this:

- if we want to accumulate array normally, use accumArrayResult and makeArrayResult. If we accumulate scalar the result will be 1D array. if we accumulate array, the resulting dimension is incremented by 1.

- if we want, somehow to affect the normal behavior, and change the dimensions, use makeMdArrayResult.

Searching through the postgres' code, other than internal use in arrayfuncs.c, makeMdArrayResult is used only in src/pl/plperl/plperl.c:plperl_array_to_datum, while converting perl array to postgres array.

So if somehow we will accumulate array other than in array_agg, i think the most natural way is using accumArrayResult and then makeArrayResult.

CMIIW

Regards,

--

Ali Akbar

2014-10-25 10:16 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

 makeArrayResult1 - I have no better name now

I found one next minor detail.

you reuse a array_agg_transfn function. Inside is a message "array_agg_transfn called in non-aggregate context". It is not correct for array_agg_anyarray_transfn

Fixed.

 

probably specification dim and lbs in array_agg_finalfn is useless, when you expect a natural result. A automatic similar to array_agg_anyarray_finalfn should work there too.

makeMdArrayResultArray and appendArrayDatum  should be marked as "static"

ok, marked all of that as static.

 

I am thinking so change of makeArrayResult is wrong. It is developed for 1D array. When we expect somewhere ND result now, then we should to use makeMdArrayResult there. So makeArrayResult should to return 1D array in all cases. Else a difference between makeArrayResult and makeMdArrayResult hasn't big change and we have a problem with naming. 

I'm thinking it like this:

- if we want to accumulate array normally, use accumArrayResult and makeArrayResult. If we accumulate scalar the result will be 1D array. if we accumulate array, the resulting dimension is incremented by 1.

- if we want, somehow to affect the normal behavior, and change the dimensions, use makeMdArrayResult.

Searching through the postgres' code, other than internal use in arrayfuncs.c, makeMdArrayResult is used only in src/pl/plperl/plperl.c:plperl_array_to_datum, while converting perl array to postgres array.

So if somehow we will accumulate array other than in array_agg, i think the most natural way is using accumArrayResult and then makeArrayResult.

ok, there is more variants and I can't to decide. But I am not satisfied with this API. We do some wrong in structure. makeMdArrayResult is now ugly.

2014-10-25 15:43 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

2014-10-25 10:16 GMT+02:00 Ali Akbar <the.apaan@gmail.com>:

 makeArrayResult1 - I have no better name now

I found one next minor detail.

you reuse a array_agg_transfn function. Inside is a message "array_agg_transfn called in non-aggregate context". It is not correct for array_agg_anyarray_transfn

Fixed.

 

probably specification dim and lbs in array_agg_finalfn is useless, when you expect a natural result. A automatic similar to array_agg_anyarray_finalfn should work there too.

makeMdArrayResultArray and appendArrayDatum  should be marked as "static"

ok, marked all of that as static.

 

I am thinking so change of makeArrayResult is wrong. It is developed for 1D array. When we expect somewhere ND result now, then we should to use makeMdArrayResult there. So makeArrayResult should to return 1D array in all cases. Else a difference between makeArrayResult and makeMdArrayResult hasn't big change and we have a problem with naming. 

I'm thinking it like this:

- if we want to accumulate array normally, use accumArrayResult and makeArrayResult. If we accumulate scalar the result will be 1D array. if we accumulate array, the resulting dimension is incremented by 1.

- if we want, somehow to affect the normal behavior, and change the dimensions, use makeMdArrayResult.

Searching through the postgres' code, other than internal use in arrayfuncs.c, makeMdArrayResult is used only in src/pl/plperl/plperl.c:plperl_array_to_datum, while converting perl array to postgres array.

So if somehow we will accumulate array other than in array_agg, i think the most natural way is using accumArrayResult and then makeArrayResult.

ok, there is more variants and I can't to decide. But I am not satisfied with this API. We do some wrong in structure. makeMdArrayResult is now ugly.

One approach that i can think is we cleanly separate the structures and API. We don't touch existing ArrayBuildState, accumArrayResult, makeArrayResult and makeMdArrayResult, and we create new functions: ArrayBuildStateArray, accumArrayResultArray, makeArrayResultArray and makeArrayResultArrayCustom.

But if we do that, the array(subselect) implementation will not be as simple as current patch. We must also change all the ARRAY_SUBLINK-related accumArrayResult call.

Regarding current patch implementation, the specific typedefs are declared as:

+typedef struct ArrayBuildStateScalar

+{

+ ArrayBuildState astate;

...

+} ArrayBuildStateArray;

so it necessities rather ugly code like this:

+ result->elemtype = astate->astate.element_type;

Can we use C11 feature of unnamed struct like this? :

+typedef struct ArrayBuildStateScalar

+{

+ ArrayBuildState;

...

+} ArrayBuildStateArray;

so the code can be a little less ugly by doing it like this:

+ result->elemtype = astate->element_type;

I don't know whether all currently supported compilers implements this feature..

Can you suggest a better structure for this?

If we can't settle on a better structure, i think i'll reimplement array_agg without the performance improvement, using deconstruct_array and unchanged accumArrayResult & makeMdArrayResult.

Thanks,

--
Ali Akbar

you can check it? We can test, how performance lost we get. As second benefit we can get numbers for introduction new optimized array builder

Hi

My idea is using new ArrayBuilder optimized for building multidimensional arrays with own State type. I think so casting to ArrayBuildState is base of our problems, so I don't would to do. Code in array_agg_* is simple, little bit more complex code is in nodeSubplan.c. Some schematic changes are in attachments.

2014-10-27 1:38 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi

My idea is using new ArrayBuilder optimized for building multidimensional arrays with own State type. I think so casting to ArrayBuildState is base of our problems, so I don't would to do. Code in array_agg_* is simple, little bit more complex code is in nodeSubplan.c. Some schematic changes are in attachments.

Thanks! The structure looks clear, and thanks for the example on nodeSubplan.c. I will restructure the v10 of the patch to this structure.

Hi

I did some minor changes in code

* move tests of old or new builder style for array sublink out of main cycles

* some API simplification of new builder - we should not to create identical API, mainly it has no sense

2014-10-27 16:15 GMT+07:00 Pavel Stehule <pavel.stehule@gmail.com>:

Hi

I did some minor changes in code

* move tests of old or new builder style for array sublink out of main cycles

* some API simplification of new builder - we should not to create identical API, mainly it has no sense

minor changes in the patch:

* remove array_agg_finalfn_internal declaration in top of array_userfuncs.c

* fix comment of makeMdArray

* fix return of makeMdArray

* remove unnecesary changes to array_agg_finalfn

Regards,

--

Ali Akbar

super

I tested last version and I have not any objections.

1. We would to have this feature - it is long time number of our ToDo List

2. Proposal and design of multidimensional aggregation is clean and nobody has objection here.

3. There is zero impact on current implementation. From performance reasons it uses new array optimized aggregator - 30% faster for this purpose than current (scalar) array aggregator

4. Code is clean and respect PostgreSQL coding rules

5. There are documentation and necessary regress tests

6. Patching and compilation is clean without warnings.

This patch is ready for commit

Thank you for patch

Pavel Stehule <pavel.stehule@gmail.com> writes:
> 2014-10-27 11:20 GMT+01:00 Ali Akbar <the.apaan@gmail.com>:
>> [ array_agg_anyarray-13.patch ]

> This patch is ready for commit

I've committed this after some significant modifications.

I did not like the API chosen, specifically the business about callers
needing to deal with both accumArrayResult and accumMdArray, because that
seemed pretty messy and error-prone. 

astate->aitems = Max(astate->aitems * 2, newnitems);