Thread: [DOCS] Doc BUG and/or Change Request: Include Postgres's system columns inreserved words in docs, Chnage to previx sytem colums with 2 underscores?

Vitals:

Postgres 9.6 on Windows 7

by way of NpgSql C# library (v4.0.30319)



DOC BUG:  
I'm curious if PostgreSQL's own special columns should be included on the "Reserved Words" list.  This might be obvious to those familiar with the inner working of Postgres, for us that are more application level developers this may not be as obvious..  Easy to know the core sql tokens are reserved such as "select, min, like, from, and others..." are all reserved.  The issue was encountered when trying to create a table with cmax as a column name.   Following the instructions on filing bugs I searched the main site and eventually found it is documented but not clearly.  Found that cmax is a system column reserved by Postgres.  At first when i got the error specifically this one:

Npgsql.PostgresException (0x80004005): 42701: column name "cmax" conflicts with a system column name

My first step was to google "Postgres Reserved Words"   will eventually get you to Appendix C in the doc but not mention about any system columns, url for help page:

https://www.postgresql.org/docs/9.6/static/sql-keywords-appendix.html

title on page shows (This page in other versions: 9.2 / 9.3 / 9.4 / 9.5 / current (9.6) )

It wasn't until i looked at the process to open a bug, did it get to the main Postgres site and searched for cmax.  Searching on the Postgres site, bingo, I found cmax instead is a system column.  So it seems at first to be a documentation bug that Postgres's own reserved column names are not put in the "Reserved Words" list. 


Change Request:

Got me thinking what the convention use to be in C starting out 30 years ago, should work good in Postgres as well?  Most all C compilers would prefix variables or items within system  struct that were considered system/reserved with an underscores or two.  This worked great before C++ which brought classes and namespaces to assist in scoping.  This practice worked well to prevent name collisions between users code the compiler itself should work nicely in Postgres too.  Finally if this convention is adapted there could be one generic note in the docs saying "User columns should never being with an underscore."  This convention worked well for C before C++ exists, should work good for Postgres too?

Thanks for your consideration.

dan d:)

2017-10-02 22:30 GMT-03:00 dan d <dano2k3@hotmail.com>:
> Npgsql.PostgresException (0x80004005): 42701: column name "cmax" conflicts
> with a system column name
>
> My first step was to google "Postgres Reserved Words"   will eventually get
> you to Appendix C in the doc but not mention about any system columns, url
> for help page:
>
> https://www.postgresql.org/docs/9.6/static/sql-keywords-appendix.html
>
The title is clear: *SQL* Key Words. It is not *PostgreSQL* Key Words.
BTW, system columns are documented [1].

> Got me thinking what the convention use to be in C starting out 30 years
> ago, should work good in Postgres as well?  Most all C compilers would
> prefix variables or items within system  struct that were considered
> system/reserved with an underscores or two.  This worked great before C++
> which brought classes and namespaces to assist in scoping.  This practice
> worked well to prevent name collisions between users code the compiler
> itself should work nicely in Postgres too.  Finally if this convention is
> adapted there could be one generic note in the docs saying "User columns
> should never being with an underscore."  This convention worked well for C
> before C++ exists, should work good for Postgres too?
>
This boat was already shipped a long time ago. You can't change that
without breaking tons of applications.


[1] https://www.postgresql.org/docs/9.6/static/ddl-system-columns.html


--   Euler Taveira                                   Timbira -
http://www.timbira.com.br/  PostgreSQL: Consultoria, Desenvolvimento, Suporte 24x7 e Treinamento


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

On Tue, Oct 3, 2017 at 6:13 AM, Euler Taveira <euler@timbira.com.br> wrote:
2017-10-02 22:30 GMT-03:00 dan d <dano2k3@hotmail.com>:
> Npgsql.PostgresException (0x80004005): 42701: column name "cmax" conflicts
> with a system column name
>
> My first step was to google "Postgres Reserved Words"   will eventually get
> you to Appendix C in the doc but not mention about any system columns, url
> for help page:
>
> https://www.postgresql.org/docs/9.6/static/sql-keywords-appendix.html
>
The title is clear: *SQL* Key Words. It is not *PostgreSQL* Key Words.
BTW, system columns are documented [1].

​That's a bit unconvincing since there are words on that table that are not marked as being reserved in any of the three​ SQL standard columns.

​Its user-error that an error message clearly indicating "system column name" prompted a search for reserved words instead of "system columns".  ​

The main distinction here is that "cmax" is a valid identifier expect when used as a table column name (you can, for instance, name columns of CTEs cmax) due to it already being in use.  Given the clarity of the error message and the fact that this is first I've seen the particular complaint in many years of monitoring these lists, I'd say trying to add system columns to the listing of reserved words in the interest of "comprehensiveness" doesn't outweigh the complexity that would be introduced explaining how it differs from other reserved words.

I wouldn't be opposed to adding a sentence and cross-reference link to the system columns page and pointing out that specific column names are also already reserved for system use.

​David J.

There is one other issue that I forgot to mention,  my program is taking existing databases and converting and loading them into a new Postgres DB, strange that one of my first test database actually had cmax as a column, unfortunately i'm not at liberty to go back and tell my customer you need to change your database columns...   I'd argue we can be assured SQL reserved words won't be used for the obvious reason, but not Postgres's system columns,  because of a short sighted design flaw.   Hard for me to go back and tell a MSSQL or Oracle DBA use you need to change your database and stop using Postgres's reserved column names...   That isn't gonna happen, and I would argue they have a valid point Postgres shouldn't impose any additional restrictions on reserved words besides what SQL imposes, especially when it would be easily for Postges just as compilers did to prefix system required column/variables with underscores.

The fix is really simply prefix system reserved column names with one or two underscores.  Simple clean prevents the issue between user code and database code...




From: David G. Johnston <david.g.johnston@gmail.com>
Sent: Tuesday, October 3, 2017 7:25 AM
To: Euler Taveira
Cc: dan d; pgsql-docs@postgresql.org
Subject: Re: [DOCS] Doc BUG and/or Change Request: Include Postgres's system columns in reserved words in docs, Chnage to previx sytem colums with 2 underscores?
 
On Tue, Oct 3, 2017 at 6:13 AM, Euler Taveira <euler@timbira.com.br> wrote:
2017-10-02 22:30 GMT-03:00 dan d <dano2k3@hotmail.com>:
> Npgsql.PostgresException (0x80004005): 42701: column name "cmax" conflicts
> with a system column name
>
> My first step was to google "Postgres Reserved Words"   will eventually get
> you to Appendix C in the doc but not mention about any system columns, url
> for help page:
>
> https://www.postgresql.org/docs/9.6/static/sql-keywords-appendix.html
>
The title is clear: *SQL* Key Words. It is not *PostgreSQL* Key Words.
BTW, system columns are documented [1].

That's a bit unconvincing since there are words on that table that are not marked as being reserved in any of the three SQL standard columns.

Its user-error that an error message clearly indicating "system column name" prompted a search for reserved words instead of "system columns". 

The main distinction here is that "cmax" is a valid identifier expect when used as a table column name (you can, for instance, name columns of CTEs cmax) due to it already being in use.  Given the clarity of the error message and the fact that this is first I've seen the particular complaint in many years of monitoring these lists, I'd say trying to add system columns to the listing of reserved words in the interest of "comprehensiveness" doesn't outweigh the complexity that would be introduced explaining how it differs from other reserved words.

I wouldn't be opposed to adding a sentence and cross-reference link to the system columns page and pointing out that specific column names are also already reserved for system use.

David J.

On Tue, Oct 3, 2017 at 10:34 AM, dan d <dano2k3@hotmail.com> wrote:

The fix is really simply prefix system reserved column names with one or two underscores.  Simple clean prevents the issue between user code and database code...

The challenge isn't in making the fix (though it would likely be extremely tedious).  The challenge is compatibility given the the current state has been in place for decades; system columns are user-visible.

This seems like a case where the cure is potentially worse than the disease.  It is difficult to please everyone and since this particular problem is very end-user specific whereas a change would impact many perfectly fine applications it seems like displeasing the unique end-user (and thus retain the status quo) is likely to prevail even were someone to invest the time to write a patch.  Though, unless someone is seriously offering to write such a patch wondering whether it would be accepted is not particularly productive.

David J.