Re: Trigger behaviour not as stated

[Bruce Momjian]

On Sun, Jan 28, 2018 at 06:17:19PM -0500, Bruce Momjian wrote:
> On Sun, Jan 28, 2018 at 06:12:01PM -0500, Tom Lane wrote:
> > Bruce Momjian  writes:
> > > Oh, I am sorry.  I was focused on the first part of the sentence and
> > > didn't notice your change to the second part.  How is this attachment?
> > 
> > Seems same as your previous version?
> 
> OK, new vesion that uses "explicitly named" in both modified doc lines.

Applied and backpatched through 10, where the text was added.

-- 
  Bruce Momjian  http://momjian.us
  EnterpriseDB http://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+  Ancient Roman grave inscription +

Re: Trigger behaviour not as stated

[Bruce Momjian]

On Mon, Jan 29, 2018 at 11:32:34AM +, Ian R. Campbell wrote:
> The second part of the confusion is that INSERT is not considered to be a row
> modification and will fire a BEFORE INSERT trigger on the parent table even
> when the data goes into a child (whereas UPDATE and DELETE will not fire a
> parent trigger).

Ian, that is not true based on my testing.  Running that attached script
that I already posted shows:

test=> INSERT INTO parent VALUES (1, 'one');
NOTICE:  Called by parent INSERT
INSERT 0 1
test=> INSERT INTO child VALUES (2, 'two');
--> NOTICE:  Called by child INSERT
INSERT 0 1

-- 
  Bruce Momjian  http://momjian.us
  EnterpriseDB http://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+  Ancient Roman grave inscription +


trigger.sql
Description: application/sql

Re: Trigger behaviour not as stated

[Tom Lane]

Bruce Momjian  writes:
> Oh, I am sorry.  I was focused on the first part of the sentence and
> didn't notice your change to the second part.  How is this attachment?

Seems same as your previous version?

regards, tom lane

Re: Trigger behaviour not as stated

[Bruce Momjian]

On Sun, Jan 28, 2018 at 03:57:07PM -0500, Tom Lane wrote:
> Bruce Momjian  writes:
> > Uh, I don't think we want to highlight the statement vs row difference
> > here but the fact that statement triggers fire on the referenced object
> > and not on the effected rows.  I have attached an updated patch which I
> > think is an improvement
> 
>  statement-level triggers for its partitions or child tables.  In 
> contrast,
> !row-level triggers are fired on the rows in effected partitions or
> !child tables, even if they are not referenced in the query.
> 
> I still think that that's not well written.  A large part of the confusion
> here is over what "referenced" means.  To my mind, child tables/partitions
> are referenced by an inherited query, just not explicitly.  So that's why
> I'd prefer wording like "directly named in the query" (or "explicitly
> named").  If you insist on using "referenced" you could write "explicitly
> referenced", but IMO that's longer and no clearer.
> 
> A lesser complaint is that this reads like the antecedent of "they" is the
> rows, not the tables containing them, making the meaning of "referenced"
> even less clear.
> 
> Maybe something like
> 
> In contrast, row-level triggers are fired for individual row change
> events, and the triggers that are fired for an event are those
> attached to the specific table containing the changed row, even if
> it is a partition or child table not directly named in the query.

Oh, I am sorry.  I was focused on the first part of the sentence and
didn't notice your change to the second part.  How is this attachment?

-- 
  Bruce Momjian  http://momjian.us
  EnterpriseDB http://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+  Ancient Roman grave inscription +
diff --git a/doc/src/sgml/ref/create_trigger.sgml b/doc/src/sgml/ref/create_trigger.sgml
new file mode 100644
index a8c0b57..6519757
*** a/doc/src/sgml/ref/create_trigger.sgml
--- b/doc/src/sgml/ref/create_trigger.sgml
*** UPDATE OF column_name1
 Modifying a partitioned table or a table with inheritance children fires
!statement-level triggers directly attached to that table, but not
 statement-level triggers for its partitions or child tables.  In contrast,
!row-level triggers are fired for all affected partitions or child tables.
 If a statement-level trigger has been defined with transition relations
 named by a REFERENCING clause, then before and after
 images of rows are visible from all affected partitions or child tables.
--- 501,510 
  

 Modifying a partitioned table or a table with inheritance children fires
!statement-level triggers directly attached to the referenced table, but not
 statement-level triggers for its partitions or child tables.  In contrast,
!row-level triggers are fired on the rows in effected partitions or
!child tables, even if they are not explicitly named in the query.
 If a statement-level trigger has been defined with transition relations
 named by a REFERENCING clause, then before and after
 images of rows are visible from all affected partitions or child tables.

Re: Trigger behaviour not as stated

[Tom Lane]

Bruce Momjian  writes:
> Uh, I don't think we want to highlight the statement vs row difference
> here but the fact that statement triggers fire on the referenced object
> and not on the effected rows.  I have attached an updated patch which I
> think is an improvement

 statement-level triggers for its partitions or child tables.  In contrast,
!row-level triggers are fired on the rows in effected partitions or
!child tables, even if they are not referenced in the query.

I still think that that's not well written.  A large part of the confusion
here is over what "referenced" means.  To my mind, child tables/partitions
are referenced by an inherited query, just not explicitly.  So that's why
I'd prefer wording like "directly named in the query" (or "explicitly
named").  If you insist on using "referenced" you could write "explicitly
referenced", but IMO that's longer and no clearer.

A lesser complaint is that this reads like the antecedent of "they" is the
rows, not the tables containing them, making the meaning of "referenced"
even less clear.

Maybe something like

In contrast, row-level triggers are fired for individual row change
events, and the triggers that are fired for an event are those
attached to the specific table containing the changed row, even if
it is a partition or child table not directly named in the query.

regards, tom lane

Re: Trigger behaviour not as stated

[Bruce Momjian]

On Sun, Jan 28, 2018 at 02:18:38PM -0500, Tom Lane wrote:
> Thomas Munro  writes:
> > On Thu, Jan 25, 2018 at 11:07 AM, Bruce Momjian  wrote:
> >> ... The attached patch updates the docs to say
> >> statement-level triggers fire on the "referenced" table, while row-level
> >> triggers fire only on the "affected" table, (vs. all affected tables)
> >> even if they are not referenced in the query.  I would backpatch this to
> >> PG 10.
> 
> > I was trying to convey that, but it does seem a little terse and
> > cryptic.  Your addition of "referenced" and "only" make it clearer.
> 
> Hm, the first part of Bruce's change seems fine, but I think this wording:
> 
>  ...  In contrast,
> !row-level triggers are fired only on affected partitions or child tables,
> !even if they are not referenced in the query.
> 
> is still confusing.  How about something like
> 
>  In contrast, row-level triggers are fired for each actual row change,
>  including changes in partitions or child tables that are not directly
>  named in the query.
> 
> Possibly "row operation" would be better than "row change".

Uh, I don't think we want to highlight the statement vs row difference
here but the fact that statement triggers fire on the referenced object
and not on the effected rows.  I have attached an updated patch which I
think is an improvement.

-- 
  Bruce Momjian  http://momjian.us
  EnterpriseDB http://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+  Ancient Roman grave inscription +
diff --git a/doc/src/sgml/ref/create_trigger.sgml b/doc/src/sgml/ref/create_trigger.sgml
new file mode 100644
index a8c0b57..42649d2
*** a/doc/src/sgml/ref/create_trigger.sgml
--- b/doc/src/sgml/ref/create_trigger.sgml
*** UPDATE OF column_name1
 Modifying a partitioned table or a table with inheritance children fires
!statement-level triggers directly attached to that table, but not
 statement-level triggers for its partitions or child tables.  In contrast,
!row-level triggers are fired for all affected partitions or child tables.
 If a statement-level trigger has been defined with transition relations
 named by a REFERENCING clause, then before and after
 images of rows are visible from all affected partitions or child tables.
--- 501,510 
  

 Modifying a partitioned table or a table with inheritance children fires
!statement-level triggers directly attached to the referenced table, but not
 statement-level triggers for its partitions or child tables.  In contrast,
!row-level triggers are fired on the rows in effected partitions or
!child tables, even if they are not referenced in the query.
 If a statement-level trigger has been defined with transition relations
 named by a REFERENCING clause, then before and after
 images of rows are visible from all affected partitions or child tables.

Re: Trigger behaviour not as stated

[Tom Lane]

Thomas Munro  writes:
> On Thu, Jan 25, 2018 at 11:07 AM, Bruce Momjian  wrote:
>> ... The attached patch updates the docs to say
>> statement-level triggers fire on the "referenced" table, while row-level
>> triggers fire only on the "affected" table, (vs. all affected tables)
>> even if they are not referenced in the query.  I would backpatch this to
>> PG 10.

> I was trying to convey that, but it does seem a little terse and
> cryptic.  Your addition of "referenced" and "only" make it clearer.

Hm, the first part of Bruce's change seems fine, but I think this wording:

 ...  In contrast,
!row-level triggers are fired only on affected partitions or child tables,
!even if they are not referenced in the query.

is still confusing.  How about something like

 In contrast, row-level triggers are fired for each actual row change,
 including changes in partitions or child tables that are not directly
 named in the query.

Possibly "row operation" would be better than "row change".

regards, tom lane

Re: Trigger behaviour not as stated

[Thomas Munro]

On Thu, Jan 25, 2018 at 11:07 AM, Bruce Momjian  wrote:
> On Wed, Jan 24, 2018 at 01:10:08PM -0500, Bruce Momjian wrote:
>> On Wed, Nov 29, 2017 at 07:39:34PM +, i...@thepathcentral.com wrote:
>> > The following documentation comment has been logged on the website:
>> >
>> > Page: https://www.postgresql.org/docs/10/static/sql-createtrigger.html
>> > Description:
>> >
>> > URL: https://www.postgresql.org/docs/current/static/sql-createtrigger.html
>> >
>> > Statement: In contrast, row-level triggers are fired for all affected
>> > partitions or child tables.
>> >
>> > Row-level triggers are not fired on child tables where the trigger ON 
>> > BEFORE
>> > UPDATE | DELETE is on the parent table. Only works on BEFORE INSERT.
>
> OK, I have some more details on this.  First there is the Stackoverflow
> report:
>
>   
> https://stackoverflow.com/questions/47557665/postgresql-on-before-delete-trigger-not-firing-on-a-parent-table-in-an-inheritan
>
> The report confirms that row-level triggers are fired _only_ on affected
> tables (meaning the table that had a row change), not on any table
> mentioned _or_ affected.  The current wording, added in this commit:
>
> commit 501ed02cf6f4f60c3357775eb07578aebc912d3a
> Author: Andrew Gierth 
> Date:   Wed Jun 28 18:55:03 2017 +0100
>
> Fix transition tables for partition/inheritance.
>
> We disallow row-level triggers with transition tables on child 
> tables.
> Transition tables for triggers on the parent table contain only 
> those
> columns present in the parent.  (We can't mix tuple formats in a
> single transition table.)
>
> Patch by Thomas Munro
>
> Discussion: 
> https://postgr.es/m/CA%2BTgmoZzTBBAsEUh4MazAN7ga%3D8SsMC-Knp-6cetts9yNZUCcg%40mail.gmail.com
>
> should be improved.  The attached patch updates the docs to say
> statement-level triggers fire on the "referenced" table, while row-level
> triggers fire only on the "affected" table, (vs. all affected tables)
> even if they are not referenced in the query.  I would backpatch this to
> PG 10.

+1

I was trying to convey that, but it does seem a little terse and
cryptic.  Your addition of "referenced" and "only" make it clearer.

-- 
Thomas Munro
http://www.enterprisedb.com