Обсуждение: create table explicitly mention that unique|primary key constraint will create an

Hi. minor doc issue.
create table s1(a int, constraint s2 PRIMARY key (a));
create table s2(a int);
ERROR:  relation "s2" already exists

https://www.postgresql.org/docs/current/sql-createtable.html#SQL-CREATETABLE-PARMS-UNIQUE
maybe for the following 2 sentence
"Adding a unique constraint will automatically create a unique btree
index on the column or group of columns used in the constraint."
"Adding a PRIMARY KEY constraint will automatically create a unique
btree index on the column or group of columns used in the constraint."

maybe we can mention that: the unique btree index name will be the
constraint name.
also is "a unique" or "an unique"?
I personally thought this part is obscure.

On Mon, 2023-11-27 at 08:00 +0800, jian he wrote:
> Hi. minor doc issue.
> create table s1(a int, constraint s2 PRIMARY key (a));
> create table s2(a int);
> ERROR:  relation "s2" already exists
>
> https://www.postgresql.org/docs/current/sql-createtable.html#SQL-CREATETABLE-PARMS-UNIQUE
> maybe for the following 2 sentence
> "Adding a unique constraint will automatically create a unique btree
> index on the column or group of columns used in the constraint."
> "Adding a PRIMARY KEY constraint will automatically create a unique
> btree index on the column or group of columns used in the constraint."
>
> maybe we can mention that: the unique btree index name will be the
> constraint name.
> also is "a unique" or "an unique"?

It would be "a unique", because "unique" is pronounced "juneek", which
does not start with a vowel.

> I personally thought this part is obscure.

True; I don't find it documented that all objects in pg_class share a
namespace and that constraints are implemented by indexes of the same
name.  But I think that the first part is a property of schemas and had
better be documented there.

What do you think of the attached patch?

Yours,
Laurenz Albe

On Mon, Nov 27, 2023 at 10:30 AM Laurenz Albe <laurenz.albe@cybertec.at> wrote:
>
> What do you think of the attached patch?
>
> Yours,
> Laurenz Albe

looks good to me.

On 27.11.23 03:30, Laurenz Albe wrote:
> True; I don't find it documented that all objects in pg_class share a
> namespace and that constraints are implemented by indexes of the same
> name.  But I think that the first part is a property of schemas and had
> better be documented there.

It is documented prominently on the CREATE INDEX reference page.  We 
could document it in more places, of course.  I find the specific change 
proposal for ddl.sgml a bit weird, though, because this is a very 
introductory section, and you are referring people to pg_class (what is 
that?!?) for details.  If we want to put something there, it should 
respect the order in which that chapter introduces concepts.

The changes on create_table.sgml seem ok.  Although I had actually 
expected that the system applies the find-a-unique-name routine rather 
than taking the constraint name for the index name unaltered.

Perhaps taking the create_table.sgml changes and combination with the 
existing text on CREATE INDEX is sufficient.

On Thu, 2024-01-18 at 15:54 +0100, Peter Eisentraut wrote:
> On 27.11.23 03:30, Laurenz Albe wrote:
> > True; I don't find it documented that all objects in pg_class share a
> > namespace and that constraints are implemented by indexes of the same
> > name.  But I think that the first part is a property of schemas and had
> > better be documented there.
>
> It is documented prominently on the CREATE INDEX reference page.  We
> could document it in more places, of course.  I find the specific change
> proposal for ddl.sgml a bit weird, though, because this is a very
> introductory section, and you are referring people to pg_class (what is
> that?!?) for details.  If we want to put something there, it should
> respect the order in which that chapter introduces concepts.
>
> The changes on create_table.sgml seem ok.  Although I had actually
> expected that the system applies the find-a-unique-name routine rather
> than taking the constraint name for the index name unaltered.
>
> Perhaps taking the create_table.sgml changes and combination with the
> existing text on CREATE INDEX is sufficient.

Ah, I didn't see the CREATE INDEX page.  (As an aside: too much
conceptual stuff is documented in our reference pages, but that's a
different issue.)

For me, the intuitive place to look for information like that is the
"Data Definition" chapter, so I think we should mention it there.
I agree that "pg_class" is too advanced for that chapter, even though
there is an earlier reference to it under "System Columns".

In the attached patch, I have copied the enumeration of relations from
the CREATE INDEX page.  I think this small redundance is alright, but I
wouldn't mind if this gets removed from CREATE INDEX.

The rest is unmodified.

Yours,
Laurenz Albe

On Fri, 19 Jan 2024 at 16:16, Laurenz Albe <laurenz.albe@cybertec.at> wrote:
>
> On Thu, 2024-01-18 at 15:54 +0100, Peter Eisentraut wrote:
> > On 27.11.23 03:30, Laurenz Albe wrote:
> > > True; I don't find it documented that all objects in pg_class share a
> > > namespace and that constraints are implemented by indexes of the same
> > > name.  But I think that the first part is a property of schemas and had
> > > better be documented there.
> >
> > It is documented prominently on the CREATE INDEX reference page.  We
> > could document it in more places, of course.  I find the specific change
> > proposal for ddl.sgml a bit weird, though, because this is a very
> > introductory section, and you are referring people to pg_class (what is
> > that?!?) for details.  If we want to put something there, it should
> > respect the order in which that chapter introduces concepts.
> >
> > The changes on create_table.sgml seem ok.  Although I had actually
> > expected that the system applies the find-a-unique-name routine rather
> > than taking the constraint name for the index name unaltered.
> >
> > Perhaps taking the create_table.sgml changes and combination with the
> > existing text on CREATE INDEX is sufficient.
>
> Ah, I didn't see the CREATE INDEX page.  (As an aside: too much
> conceptual stuff is documented in our reference pages, but that's a
> different issue.)
>
> For me, the intuitive place to look for information like that is the
> "Data Definition" chapter, so I think we should mention it there.
> I agree that "pg_class" is too advanced for that chapter, even though
> there is an earlier reference to it under "System Columns".
>
> In the attached patch, I have copied the enumeration of relations from
> the CREATE INDEX page.  I think this small redundance is alright, but I
> wouldn't mind if this gets removed from CREATE INDEX.
>
> The rest is unmodified.

CFBot shows that the patch does not apply anymore as in [1]:
=== Applying patches on top of PostgreSQL commit ID
d282e88e50521a457fa1b36e55f43bac02a3167f ===
=== applying patch ./v3-0001-Doc-All-relations-share-a-namespace.patch
...
patching file doc/src/sgml/ref/create_table.sgml
Hunk #1 FAILED at 1001.
Hunk #2 FAILED at 1054.
Hunk #3 succeeded at 1118 (offset 27 lines).
2 out of 3 hunks FAILED -- saving rejects to file
doc/src/sgml/ref/create_table.sgml.rej

Please post an updated version for the same.

[1] - http://cfbot.cputube.org/patch_46_4747.log

Regards,
Vignesh

On Fri, 2024-01-26 at 19:01 +0530, vignesh C wrote:
> CFBot shows that the patch does not apply anymore as in [1]:

There was a conflict with 46a0cd4cefb.
Updated version attached.

Yours,
Laurenz Albe

On 26.01.24 16:52, Laurenz Albe wrote:
> On Fri, 2024-01-26 at 19:01 +0530, vignesh C wrote:
>> CFBot shows that the patch does not apply anymore as in [1]:
> 
> There was a conflict with 46a0cd4cefb.
> Updated version attached.

Sprinkled in some of David's suggestions, and pushed.

I was hesitant to burden the user with the difference between "relation" 
and "table" at this point, so I took the former term out of the text.

On 18.01.24 22:21, David G. Johnston wrote:
> I started looking at this specific item and immediately got the idea to 
> actually document in user-facing (i.e., not system catalogs) what these 
> object categories are in which object types share the schema namespace. 
> The "Other Object Types" section already in the DDL chapter seems to 
> provide a near-perfect place to put this (not sure I like the word 
> "other" there being my only complaint).  The attached patch replaces 
> Laurenz's v1, leaving the create_table changes as-is but presenting an 
> alternative approach to introducing namespacing when we explain why 
> schemas exist.

I think this proposal goes a bit too far into implementation-dependent 
details.  The namespace of tables and indexes is clearly important, but 
for example, the subdivision of types into range types and multi-range 
types is really low-level and not usually practically relevant.  And you 
don't mention array types, probably because they are not mentioned in 
typtype, but they are also relevant for the namespace of types.  There 
are multiple ways to slice all this, but it's not clear why we need to 
lay this all out in the introductory documentation.

On Wed, 2024-01-31 at 12:11 +0100, Peter Eisentraut wrote:
> Sprinkled in some of David's suggestions, and pushed.

Thanks; your text is great.

Yours,
Laurenz Albe