dblink_error_message return value

=?utf-8?q?PG_Doc_comments_form?= <noreply@postgresql.org> writes:

The following documentation comment has been logged on the website:
Documentation says:

Return Value
Returns last error message, or an empty string if there has been no error in
this connection.
Which is invalid.
Actually it returns 'OK' string if no error was raised.

Good catch! The code's quite clear about it, but the SGML docs need
fixed.

Secondly
dblink_is_busy must be first called to make dblink_error_message returns an
error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

regards, tom lane

On Wed, Aug 8, 2018 at 8:57 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

Secondly
dblink_is_busy must be first called to make dblink_error_message returns

an

error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

I would suggest that while rewording might be useful a more beneficial
improvement would be separating the commands into "Sync/ASync" and link
back into the corresponding libpq documentation sections which dblink
provides direct SQL layer access to (and which the docs seem to have been
written under the assumption the reader was knowledgeable of that API). It
does abstract away some of the material in the libpq docs but I think it
would be a reasonable halfway measure in lieu of spending time enhancing
the mostly obsolete dblink extension documentation.

David J.

On 08/08/2018 11:57 AM, Tom Lane wrote:

=?utf-8?q?PG_Doc_comments_form?= <noreply@postgresql.org> writes:

The following documentation comment has been logged on the website:
Documentation says:

Return Value
Returns last error message, or an empty string if there has been no error in
this connection.
Which is invalid.
Actually it returns 'OK' string if no error was raised.

Good catch! The code's quite clear about it, but the SGML docs need
fixed.

As mentioned on the nearby thread, will fix. I suppose this ought to be
back-patched.

Secondly
dblink_is_busy must be first called to make dblink_error_message returns an
error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

Maybe a note, something like this?
------------
When asynchronous queries are initiated by dblink_send_query(), the
error message associated with the connection might not get updated until
the server's response message is consumed. This typically means that
dblink_is_busy() or dblink_get_result() should be called prior to
dblink_error_message(), so that any error generated by the asynchronous
query() will be visible.
------------

Joe

--
Crunchy Data - http://crunchydata.com
PostgreSQL Support for Secure Enterprises
Consulting, Training, & Open Source Development

On 09/25/2018 01:50 PM, Joe Conway wrote:

On 08/08/2018 11:57 AM, Tom Lane wrote:

=?utf-8?q?PG_Doc_comments_form?= <noreply@postgresql.org> writes:

The following documentation comment has been logged on the website:
Documentation says:

Return Value
Returns last error message, or an empty string if there has been no error in
this connection.
Which is invalid.
Actually it returns 'OK' string if no error was raised.

Good catch! The code's quite clear about it, but the SGML docs need
fixed.

As mentioned on the nearby thread, will fix. I suppose this ought to be
back-patched.

Secondly
dblink_is_busy must be first called to make dblink_error_message returns an
error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

Maybe a note, something like this?
------------
When asynchronous queries are initiated by dblink_send_query(), the
error message associated with the connection might not get updated until
the server's response message is consumed. This typically means that
dblink_is_busy() or dblink_get_result() should be called prior to
dblink_error_message(), so that any error generated by the asynchronous
query() will be visible.
------------

And now with the corresponding patch attached.

Thoughts/comments?

Joe
--
Crunchy Data - http://crunchydata.com
PostgreSQL Support for Secure Enterprises
Consulting, Training, & Open Source Development

On 09/25/2018 03:58 PM, Joe Conway wrote:

On 09/25/2018 01:50 PM, Joe Conway wrote:

On 08/08/2018 11:57 AM, Tom Lane wrote:

=?utf-8?q?PG_Doc_comments_form?= <noreply@postgresql.org> writes:

The following documentation comment has been logged on the website:
Documentation says:

Return Value
Returns last error message, or an empty string if there has been no error in
this connection.
Which is invalid.
Actually it returns 'OK' string if no error was raised.

Good catch! The code's quite clear about it, but the SGML docs need
fixed.

As mentioned on the nearby thread, will fix. I suppose this ought to be
back-patched.

Secondly
dblink_is_busy must be first called to make dblink_error_message returns an
error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

Maybe a note, something like this?
------------
When asynchronous queries are initiated by dblink_send_query(), the
error message associated with the connection might not get updated until
the server's response message is consumed. This typically means that
dblink_is_busy() or dblink_get_result() should be called prior to
dblink_error_message(), so that any error generated by the asynchronous
query() will be visible.
------------

And now with the corresponding patch attached.

Thoughts/comments?

Going once, going twice, ...
(if no complaints will commit soon)

Joe

--
Crunchy Data - http://crunchydata.com
PostgreSQL Support for Secure Enterprises
Consulting, Training, & Open Source Development

On 10/2/18 8:45 AM, Joe Conway wrote:

On 09/25/2018 03:58 PM, Joe Conway wrote:

On 09/25/2018 01:50 PM, Joe Conway wrote:

On 08/08/2018 11:57 AM, Tom Lane wrote:

=?utf-8?q?PG_Doc_comments_form?= <noreply@postgresql.org> writes:

The following documentation comment has been logged on the website:
Documentation says:

Return Value
Returns last error message, or an empty string if there has been no error in
this connection.
Which is invalid.
Actually it returns 'OK' string if no error was raised.

Good catch! The code's quite clear about it, but the SGML docs need
fixed.

As mentioned on the nearby thread, will fix. I suppose this ought to be
back-patched.

Secondly
dblink_is_busy must be first called to make dblink_error_message returns an
error message. (Tested on 9.6.9)

Meh. I see what you're getting at here, I think, but that seems like a
completely wrong/misleading statement of the issue. Joe, can you think of
better phraseology?

Maybe a note, something like this?
------------
When asynchronous queries are initiated by dblink_send_query(), the
error message associated with the connection might not get updated until
the server's response message is consumed. This typically means that
dblink_is_busy() or dblink_get_result() should be called prior to
dblink_error_message(), so that any error generated by the asynchronous
query() will be visible.
------------

And now with the corresponding patch attached.

Thoughts/comments?

Going once, going twice, ...
(if no complaints will commit soon)

Well, maybe not so soon, but now done.

Joe

--
Crunchy Data - http://crunchydata.com
PostgreSQL Support for Secure Enterprises
Consulting, Training, & Open Source Development